Compare commits

...
Author SHA1 Message Date
BaaRai 944c131e90 docs(context-telemetry): align Phase C expected-gain bullet with minimal split 2026-11-20 20:41:48 +01:00
BaaRai 0dbb6de8d5 feat(agent): opt-in worker system cache-prefix split (P3.5 phase C) 2026-11-20 20:29:17 +01:00
BaaRai 3559be6482 Merge pull request #426 from BaaRai/feat/context-telemetry
feat(llmrec): opt-in LLM context size telemetry and aggregation (P3.5 phase A)
2026-11-20 19:37:48 +01:00
BaaRai 3bbeb2c54e docs(context-telemetry): clarify aggregate vs turn-1 worker prefix share in fixture 2026-11-20 19:09:33 +01:00
BaaRai eecf8895aa docs(context-telemetry): correct built-in prompt volatility and Phase C worker/planner analysis 2026-11-20 19:03:18 +01:00
BaaRai dd5c371ead fix(llmrec): match norma joinSystem trim/drop-empty semantics and document retention 2026-11-20 18:43:03 +01:00
BaaRai e10c2b76aa Merge pull request #419 from ldevbr/contrib/next-telemetry-off
build(web): disable Next.js telemetry in the web build stage
2026-11-20 18:16:00 +01:00
BaaRai 4841016892 feat(llmrec): opt-in LLM context size telemetry and aggregation (P3.5 phase A) 2026-11-20 18:15:23 +01:00
BaaRai 8793f7b56c Merge pull request #408 from Lusitaniae/pr/tool-checksums
build: pin tool and node digests instead of fetching their checksums files
2026-11-20 18:15:21 +01:00
Luis ef25384063 build(web): disable Next.js telemetry in the web build stage
`next build` contacts telemetry.nextjs.org during the image build. Set
NEXT_TELEMETRY_DISABLED=1 in the web stage so the build makes no telemetry
request. No runtime behavior changes.
2026-11-20 10:26:01 -05:00
BaaRai 9465dfa67a Merge pull request #412 from BaaRai/feat/traffic-har-import-export
feat(traffic): HAR 1.2 import and export (P3.4)
2026-11-20 14:57:57 +01:00
BaaRai 61d567a771 fix(traffic): HAR 1.2 compliance, round-trip IDs, aligned caps, params/cookies, userinfo redaction 2026-11-20 14:33:57 +01:00
BaaRai 74b9843761 feat(traffic): HAR 1.2 import and export (P3.4) 2026-11-20 14:19:22 +01:00
Pedro Gomes b7bc9f9517 build: pin tool and node digests instead of fetching their checksums files
Each tool archive was verified against the *_checksums.txt of its own GitHub
release, fetched in the same build step as the archive. That catches a truncated
download, a stale CDN entry and a moved tag. It does not catch the publisher: the
digest and the artifact it describes ship from the same release, so a replaced
asset plus a replaced checksums file agree with each other and the build goes
green. The build was trusting the channel it downloads from to vouch for itself.

Pin the digests in the Dockerfile instead — one per asset per architecture, next
to the version ARG — and have fetch() take the expected digest as an argument.
The assets and their URLs are unchanged; a swapped asset now fails the build
unless the diff that swaps it also changes a digest a reviewer reads. Node's
tarball in the runtime stage gets the same treatment for the same reason; it was
the last fetched-SHASUMS256 check left.

The digests were taken from the artifacts themselves and cross-checked against
the published sums files, so this change is build-identical: same 14 bytes of
input, same output, different trust anchor. Verified by building the tools stage
for both arches with the pinned values.

Cost, and it is a real one: a version bump is three edits instead of one, and a
publisher re-uploading an asset under an existing tag breaks the build until the
digest is bumped deliberately. Two hashes per tool bump is the chore that buys
not trusting the release channel.

sqlmap already worked this way (pinned tag, asserted commit); this makes the
binary tools consistent with it.
2026-11-20 09:29:34 +00:00
BaaRai ab40834567 Merge pull request #399 from BaaRai/feat/archive-v2-traffic-rewrite-filter
feat(traffic): archive v2 rewrite provenance and rewritten list filter (P3.3b2b2)
2026-11-20 08:28:05 +01:00
BaaRai 79c5daf076 feat(traffic): archive v2 rewrite provenance and traffic list filter (P3.3b2b2) 2026-11-20 08:03:04 +01:00
BaaRai 64b90a746a Merge pull request #292 from henefesto/docs/local-systemd-deployment-32ebb38
docs: add local systemd deployment guide and verified reboot results
2026-11-20 07:46:02 +01:00
BaaRai dbbc287cd8 Merge pull request #395 from BaaRai/security/agent-exec-isolation
security: contain agent execution (exec-uid drop, run dirs, spills, Norma temp)
2026-11-20 07:42:38 +01:00
BaaRai 72fe3c3da1 Merge pull request #394 from BaaRai/feat/replay-rewrite-toggle
feat(traffic): per-replay rewrite-rule toggle and rewrite stats (P3.3b2b1)
2026-11-20 06:02:23 +01:00
BaaRai 15ed120758 test(traffic): adversarial coverage for the replay rewrite toggle
Duplicate and lower-case nonce headers are stripped without bypassing,
the registry is single-use and rejects multi-valued fields, a forged
caller header cannot steer either mode, concurrent bypass and apply
replays do not share state, SSE streams bypass by default, a cancelled
replay leaves no nonce or marker, and the route rejects a non-boolean
apply_rules.
2026-11-20 05:48:34 +01:00
BaaRai 221ad67549 feat(traffic): per-replay rewrite-rule toggle and RewriteStats API (P3.3b2b1) 2026-11-20 05:25:35 +01:00
BaaRai dbf5cc3186 docs: generalize systemd example unit and deployment guide (#292) 2026-11-20 03:26:50 +01:00
ParallelsandWarp 32ebb38e0f docs: document local systemd deployment and verified reboot persistence
Add the ARTEX systemd unit and document PostgreSQL setup, application deployment, operations, and successful automatic startup after reboot.

Co-Authored-By: Warp <agent@warp.dev>
2026-11-17 09:20:13 -04:00
41 changed files with 6504 additions and 133 deletions
+4 -1
View File
@@ -23,8 +23,11 @@ All notable changes to this project are documented in this file. The format is b
- **`ask_operator` target warning**: a request whose question names a host or URL outside the goal targets is tagged as containing target-derived text (#197).
- **Opt-in update banner**: `ARTEX_UPDATE_CHECK` makes the backend check the latest GitHub release at most once per 24 hours and the UI show a banner; off by default (#202).
- **Release provenance**: release archives carry GitHub build provenance attestations and a CycloneDX SBOM (`gh attestation verify <zip> --repo BaaRai/ARTEX`); release notes are the matching `CHANGELOG.md` section (#219, #201).
- **Opt-in worker cache-prefix split (P3.5 Phase C)**: setting `ARTEX_CACHE_PREFIX=1` (off by default) splits the `worker` system prompt at `DynamicBoundary = 1` into a sibling-shareable static prefix (`body + untrustedContentTail + workerTrafficBlock`) and a per-intent dynamic suffix (`workerArtifactSpec(runDir)` through `<available-deferred-tools>`) without reordering sections or altering the visible joined prompt text. Planner, mainagent, chat, and goals are unchanged; custom worker prompts referencing `{{.Now}}` remain volatile in the static segment. See `docs/context-telemetry.md`.
- **Opt-in LLM context size telemetry (P3.5 Phase A)**: setting `ARTEX_CONTEXT_TELEMETRY=1` (off by default) records per-call request section byte sizes (`system_static`/`system_dynamic` split at `DynamicBoundary`, serialized `tools`, `messages` by role, `tool_result` subset, and `last_message_bytes`), tool/message counts, and token/cache usage in the new `llm_context` table (migration `0009_llm_context.sql`), with zero prompt/behavior change and zero content stored, independent of `llm_record`. `GET /api/llm/context/summary` aggregates input percentiles (`p50`/`p90`/`p99`), pro-rata section shares over real `input_tokens`, `cache_read/input` ratios by `(worker, model)`, and top tasks. See `docs/context-telemetry.md`.
- **HAR 1.2 traffic import and export**: `POST /api/traffic/import/har` imports raw HAR 1.2 JSON captures (up to 32 MiB / 2,000 entries / 32 MiB decoded bodies) with deterministic `har-<sha256>` IDs (and `_artexExchangeId` preservation for ARTEX exports so distinct same-second exchanges round-trip without collapsing), base64 decoding, `postData.params` and `cookies` array fallbacks, URL userinfo stripping, blob spilling above 256 KiB, and full-text indexing, without applying capture rewrite rules (`exchange_rewrites` stays empty); per-entry errors are reported alongside `imported`/`skipped`/`failed` counts. `GET /api/traffic/export/har` exports filtered traffic (reusing the `GET /api/traffic` filters) as a compliant HAR 1.2 document (with explicit unknown placeholders for unrecorded timings and header sizes) in stable `(ts, id)` order, encoding UTF-8 bodies as text and binary bodies as base64, capped at 1,000 entries / 20 MiB raw headers+bodies (31 MiB JSON backstop, guaranteed under the 32 MiB import limit) with explicit truncation reporting via `X-ARTEX-HAR-*` headers and `_truncated`/`_truncatedReason`/`_totalMatched` fields. The Traffic page adds **Import HAR** and **Export HAR** controls. See `docs/traffic-har.md`.
- **Capture-proxy rewrite rules, storage and API**: `GET`, `POST`, `PUT` and `DELETE /api/traffic/rewrite-rules` and `POST /api/traffic/rewrite-rules/{id}/toggle` manage rules that match a request or a response (method, host, URL, headers, body, status) and describe a rewrite (method, path and query, headers, body, status). Rules are validated on save, limited to 100 of which 50 enabled, created disabled, and stored in the new table `traffic_rewrite_rules` (migration `0005_traffic_rewrite_rules.sql`) with `matchers` and `actions` encrypted at rest. The value of an action marked `sensitive` is never returned, and an update sent without it keeps the stored one. The server compiles the enabled rules at startup and after every change and hands the snapshot to the proxy. The proxy applies them from the next entry. See `docs/traffic-rewrite-rules.md`.
- **Capture-proxy rewrite rules applied**: the capture proxy applies enabled rules to each HTTP request before it is sent to the target and to each response before it reaches the client. Each action that ran is recorded in the new table `exchange_rewrites` in the same transaction as the exchange, `traffic_get` shows a `REWRITTEN` line, and `traffic_search` never finds a value a rule wrote. SSE status and headers are rewritten at the start of the stream. **WebSocket handshakes are not rewritten**: the capture library runs its hook after the upgrade, so the status and headers cannot be changed, and frames pass unchanged. **Responses of 5 MiB or more** are forwarded unchanged and not recorded in full, because the library does not call the response hook for them. Snapshots of original requests are capped at 1024 entries and 64 MiB. Replays go through the proxy, so the rules apply to them. Archives, a per-replay rule toggle, the web UI and the proof flag come later.
- **Capture-proxy rewrite rules applied**: the capture proxy applies enabled rules to each HTTP request before it is sent to the target and to each response before it reaches the client. Each action that ran is recorded in the new table `exchange_rewrites` in the same transaction as the exchange, `traffic_get` shows a `REWRITTEN` line, and `traffic_search` never finds a value a rule wrote. SSE status and headers are rewritten at the start of the stream. **WebSocket handshakes are not rewritten**: the capture library runs its hook after the upgrade, so the status and headers cannot be changed, and frames pass unchanged. **Responses of 5 MiB or more** are forwarded unchanged and not recorded in full, because the library does not call the response hook for them. Snapshots of original requests are capped at 1024 entries and 64 MiB. Replays (`POST /api/traffic/replay`) bypass capture rewrite rules by default via an internal `X-ARTEX-Replay-Nonce` header stripped unconditionally by the proxy, and apply the rules when opted in with `apply_rules: true` (the **Apply capture rewrite rules** checkbox in the Replay panel). `GET /api/traffic` exposes the proxy's `rewrite` counters (`large_responses`, `capacity_refusals`, `tombstone_drops`, `conversions`, `sweeps`), flags each exchange row with `rewritten`, and filters with `rewritten=all|rewritten|original` (also accepting `true`/`false`), surfaced with a filter control and row badge on the Traffic page. Traffic archives export version 2 with per-exchange `rewritten` and sorted distinct applied `rewrite_rule_ids` while remaining deterministic and backward-compatible with version 1 archives. The rule-management web UI comes later.
- **Bulk finding status** (#298): select findings in the findings table and apply one disposition to all of them, up to 500 per call, in one transaction; each row still goes through the single-decision notification semantics and the change is mirrored onto its exploration node (#289).
- **Stream message direction filter** (#379): `GET /api/traffic/stream` and the `traffic_messages` agent tool take an optional `direction` (`c2s` or `s2c`) that lists only one side of a WebSocket or SSE conversation. `total` and `offset` apply to the filtered list, each row keeps its `seq` in the whole stream, and the `meta` row (messages dropped past the recorder limits) stays in a filtered page. Any other value is a 400 (API) or a tool error.
- **Traffic replay** (#266): replay a recorded request from the traffic detail with its method, URL, headers and stored body; replay is disabled when only the body preview is loaded.
+57 -26
View File
@@ -14,6 +14,7 @@
# Pinned to the Node 22 line the CI/release workflows use. On the build platform
# only; the frontend static export is architecture-independent.
FROM --platform=$BUILDPLATFORM node:22-bookworm-slim AS web
ENV NEXT_TELEMETRY_DISABLED=1
WORKDIR /src/web
COPY web/package.json web/package-lock.json ./
RUN npm ci --ignore-scripts
@@ -47,19 +48,44 @@ RUN ARTEX_SKIP_FRONTEND=1 \
# Web-recon toolbox (drost-community take 1, #100): pinned release binaries, staged
# on the build platform and copied into /usr/local/bin of the runtime image so hunt
# workers can run them directly through their Bash tool. Versions pinned per tool;
# upgrade deliberately. Each archive is verified against the checksums file of its
# own GitHub release (exact asset name, sha256). That file comes from the same
# release channel as the archive, so it catches corrupted or swapped downloads and
# moved tags, not a compromised publisher. sqlmap has no release binaries: it is a
# shallow clone of a pinned tag, checked against the expected commit.
# upgrade deliberately.
#
# Every archive is verified against a SHA-256 written down here, per asset and per
# architecture, not against the *_checksums.txt of its own GitHub release. Checking
# against that file catches a corrupted or swapped download and a moved tag, but the
# digest and the artifact it describes come from the same publisher and the same
# release: a replacement asset and a replacement checksums file agree with each
# other and the build passes. Pinning the digest here moves that decision out of the
# build, so a swapped asset fails unless the diff that swaps it also changes the
# digest a reviewer reads.
#
# Cost: a version bump is now three edits (the version ARG and the two digests), and
# a publisher re-uploading an asset under the same tag breaks the build until the
# digest is bumped on purpose. That is the trade: the build stops trusting the
# channel it downloads from.
#
# sqlmap has no release binaries: it is a shallow clone of a pinned tag, checked
# against the expected commit, which is the same idea.
FROM --platform=$BUILDPLATFORM debian:bookworm-slim AS tools
ARG TARGETARCH
ARG NUCLEI_VERSION=v3.11.1
ARG NUCLEI_SHA256_AMD64=ea63d4ae232808cd7c6bc00d0142428e231fab59dae01042246097d195835ab6
ARG NUCLEI_SHA256_ARM64=8044e3d9768ba0a744b2872c1a87e813006f013da97ca9f50f7661a4203bec07
ARG HTTPX_VERSION=v1.12.0
ARG HTTPX_SHA256_AMD64=9d8439e8b6c9aa7d1e2314817a392e00d5178da3af5652f7475f88868f418f76
ARG HTTPX_SHA256_ARM64=fd7b123c1dfbc3d69f19f524e4eebcd6ec06b9a6cbd56813c76f11645197331e
ARG SUBFINDER_VERSION=v2.17.0
ARG SUBFINDER_SHA256_AMD64=a39c3e5c7a26ef3c430915c4d6cbf7716a00f26769a75e9405aba796142eba37
ARG SUBFINDER_SHA256_ARM64=24913d49dddc1d3a5e99a4a7580da5be980959b9415b96f52c4f7f4124392190
ARG KATANA_VERSION=v1.8.0
ARG KATANA_SHA256_AMD64=cb027ae99d0757061b70c07a15c7a80f62c5a535ab52c4a329f80c4b0a7309d2
ARG KATANA_SHA256_ARM64=80e269210727a855511ffac8dfb01297b2344742acec71a35e8ba4a67a90f0f2
ARG FFUF_VERSION=v2.3.0
ARG FFUF_SHA256_AMD64=b2a3c725fcb9da175159682f54d6e9149f2905b00d84d155e7efc5d599975ceb
ARG FFUF_SHA256_ARM64=e2d712615d53da797ee929006e41f2df5a7960a5353133db9c150ddaa22ae4e8
ARG DALFOX_VERSION=v3.2.4
ARG DALFOX_SHA256_AMD64=c11cfd2f8261973f0a2a2a1eba5a147ce77712a29acfa6343f9847fb8bb7eca2
ARG DALFOX_SHA256_ARM64=fd56b0b7da2d06868cd73b504f41d21314629feffcd5bdc97718db1c9dd2fa70
ARG SQLMAP_VERSION=1.9.9
ARG SQLMAP_COMMIT=cceda39208b5aad478de2c3f731df4239d9881f6
RUN apt-get update && apt-get install -y --no-install-recommends \
@@ -67,17 +93,21 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
&& rm -rf /var/lib/apt/lists/*
RUN set -eux; \
case "${TARGETARCH:-amd64}" in \
amd64) PD_ARCH=amd64; FFUF_ARCH=amd64; DALFOX_ARCH=x86_64 ;; \
arm64) PD_ARCH=arm64; FFUF_ARCH=arm64; DALFOX_ARCH=aarch64 ;; \
amd64) PD_ARCH=amd64; FFUF_ARCH=amd64; DALFOX_ARCH=x86_64; \
NUCLEI_SUM="$NUCLEI_SHA256_AMD64"; HTTPX_SUM="$HTTPX_SHA256_AMD64"; \
SUBFINDER_SUM="$SUBFINDER_SHA256_AMD64"; KATANA_SUM="$KATANA_SHA256_AMD64"; \
FFUF_SUM="$FFUF_SHA256_AMD64"; DALFOX_SUM="$DALFOX_SHA256_AMD64" ;; \
arm64) PD_ARCH=arm64; FFUF_ARCH=arm64; DALFOX_ARCH=aarch64; \
NUCLEI_SUM="$NUCLEI_SHA256_ARM64"; HTTPX_SUM="$HTTPX_SHA256_ARM64"; \
SUBFINDER_SUM="$SUBFINDER_SHA256_ARM64"; KATANA_SUM="$KATANA_SHA256_ARM64"; \
FFUF_SUM="$FFUF_SHA256_ARM64"; DALFOX_SUM="$DALFOX_SHA256_ARM64" ;; \
*) echo "unsupported architecture: ${TARGETARCH}" >&2; exit 1 ;; \
esac; \
mkdir -p /out; \
fetch() { base="$1"; asset="$2"; sums="$3"; name="$4"; \
fetch() { base="$1"; asset="$2"; want="$3"; name="$4"; \
rm -rf /tmp/t; mkdir -p /tmp/t/x; \
curl -fsSL "$base/$asset" -o "/tmp/t/$asset"; \
curl -fsSL "$base/$sums" -o /tmp/t/sums.txt; \
want="$(awk -v a="$asset" '$2 == a || $2 == "*" a { print $1 }' /tmp/t/sums.txt)"; \
if [ -z "$want" ]; then echo "no checksum entry for $asset" >&2; exit 1; fi; \
if [ -z "$want" ]; then echo "no pinned digest for $asset" >&2; exit 1; fi; \
got="$(sha256sum "/tmp/t/$asset" | cut -d' ' -f1)"; \
if [ "$want" != "$got" ]; then echo "checksum mismatch for $asset: want $want got $got" >&2; exit 1; fi; \
case "$asset" in *.zip) unzip -q "/tmp/t/$asset" -d /tmp/t/x ;; *) tar -xzf "/tmp/t/$asset" -C /tmp/t/x ;; esac; \
@@ -86,17 +116,17 @@ RUN set -eux; \
rm -rf /tmp/t; }; \
gh="https://github.com"; \
fetch "$gh/projectdiscovery/nuclei/releases/download/$NUCLEI_VERSION" \
"nuclei_${NUCLEI_VERSION#v}_linux_${PD_ARCH}.zip" "nuclei_${NUCLEI_VERSION#v}_checksums.txt" nuclei; \
"nuclei_${NUCLEI_VERSION#v}_linux_${PD_ARCH}.zip" "$NUCLEI_SUM" nuclei; \
fetch "$gh/projectdiscovery/httpx/releases/download/$HTTPX_VERSION" \
"httpx_${HTTPX_VERSION#v}_linux_${PD_ARCH}.zip" "httpx_${HTTPX_VERSION#v}_checksums.txt" httpx; \
"httpx_${HTTPX_VERSION#v}_linux_${PD_ARCH}.zip" "$HTTPX_SUM" httpx; \
fetch "$gh/projectdiscovery/subfinder/releases/download/$SUBFINDER_VERSION" \
"subfinder_${SUBFINDER_VERSION#v}_linux_${PD_ARCH}.zip" "subfinder_${SUBFINDER_VERSION#v}_checksums.txt" subfinder; \
"subfinder_${SUBFINDER_VERSION#v}_linux_${PD_ARCH}.zip" "$SUBFINDER_SUM" subfinder; \
fetch "$gh/projectdiscovery/katana/releases/download/$KATANA_VERSION" \
"katana_${KATANA_VERSION#v}_linux_${PD_ARCH}.zip" "katana-${KATANA_VERSION#v}-checksums.txt" katana; \
"katana_${KATANA_VERSION#v}_linux_${PD_ARCH}.zip" "$KATANA_SUM" katana; \
fetch "$gh/ffuf/ffuf/releases/download/$FFUF_VERSION" \
"ffuf_${FFUF_VERSION#v}_linux_${FFUF_ARCH}.tar.gz" "ffuf_${FFUF_VERSION#v}_checksums.txt" ffuf; \
"ffuf_${FFUF_VERSION#v}_linux_${FFUF_ARCH}.tar.gz" "$FFUF_SUM" ffuf; \
fetch "$gh/hahwul/dalfox/releases/download/$DALFOX_VERSION" \
"dalfox-${DALFOX_VERSION}-linux-${DALFOX_ARCH}-musl.tar.gz" "dalfox-${DALFOX_VERSION}-linux-${DALFOX_ARCH}-musl.tar.gz.sha256" dalfox; \
"dalfox-${DALFOX_VERSION}-linux-${DALFOX_ARCH}-musl.tar.gz" "$DALFOX_SUM" dalfox; \
git clone --quiet --depth 1 --branch "$SQLMAP_VERSION" https://github.com/sqlmapproject/sqlmap /opt/sqlmap; \
commit="$(git -C /opt/sqlmap rev-parse HEAD)"; \
if [ "$commit" != "$SQLMAP_COMMIT" ]; then echo "sqlmap commit mismatch: want $SQLMAP_COMMIT got $commit" >&2; exit 1; fi; \
@@ -110,26 +140,27 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
ca-certificates ripgrep curl wget vim git jq unzip xz-utils \
dnsutils iputils-ping netcat-openbsd inetutils-telnet whois nmap \
&& rm -rf /var/lib/apt/lists/*
# Node 22 from the official nodejs.org tarball, verified against its SHASUMS256.txt:
# bookworm's apt nodejs is 18, Playwright requires >=20, and 20 is end-of-life.
# 22 matches the CI and release workflows.
# Node 22 from the official nodejs.org tarball: bookworm's apt nodejs is 18,
# Playwright requires >=20, and 20 is end-of-life. 22 matches the CI and release
# workflows. Verified against a digest pinned here rather than against the fetched
# SHASUMS256.txt, for the reason given at the tools stage.
ARG NODE_VERSION=22.23.4
ARG NODE_SHA256_AMD64=1805bcd35e6815559bd2eb97db438d3fdb0eb3a5e6a977b7c1badca76337e9ce
ARG NODE_SHA256_ARM64=6287aa46b6c375d816e63ae66d024fbbfd70404825aabdfdf898a970871d2f4b
RUN set -eux; \
case "${TARGETARCH:-amd64}" in \
amd64) NODE_ARCH=x64 ;; \
arm64) NODE_ARCH=arm64 ;; \
amd64) NODE_ARCH=x64; want="$NODE_SHA256_AMD64" ;; \
arm64) NODE_ARCH=arm64; want="$NODE_SHA256_ARM64" ;; \
*) echo "unsupported architecture: ${TARGETARCH}" >&2; exit 1 ;; \
esac; \
base="https://nodejs.org/dist/v${NODE_VERSION}"; \
asset="node-v${NODE_VERSION}-linux-${NODE_ARCH}.tar.xz"; \
curl -fsSL "$base/$asset" -o "/tmp/$asset"; \
curl -fsSL "$base/SHASUMS256.txt" -o /tmp/SHASUMS256.txt; \
want="$(awk -v a="$asset" '$2 == a { print $1 }' /tmp/SHASUMS256.txt)"; \
if [ -z "$want" ]; then echo "no checksum entry for $asset" >&2; exit 1; fi; \
[ -n "$want" ] || { echo "no pinned digest for $asset" >&2; exit 1; }; \
got="$(sha256sum "/tmp/$asset" | cut -d' ' -f1)"; \
if [ "$want" != "$got" ]; then echo "checksum mismatch for $asset: want $want got $got" >&2; exit 1; fi; \
tar -xJf "/tmp/$asset" -C /usr/local --strip-components=1 --no-same-owner; \
rm -f "/tmp/$asset" /tmp/SHASUMS256.txt; \
rm -f "/tmp/$asset"; \
node --version; npm --version
# Pinned, checksum-verified recon binaries and the sqlmap checkout from the tools stage.
COPY --from=tools /out/ /usr/local/bin/
+54 -8
View File
@@ -387,12 +387,57 @@ func IsRunDirName(name string) bool {
return true
}
func workerSystem(proxyAddr, caCert, dataDir, runDir string) string {
// CachePrefixEnv is the opt-in environment flag enabling the worker system
// prompt cache-prefix split (P3.5 phase C). Off by default: when unset/false,
// worker system segments and DynamicBoundary are byte-identical to the legacy
// layout. Planner, mainagent, chat, and goals are unaffected.
const CachePrefixEnv = "ARTEX_CACHE_PREFIX"
func cachePrefixEnabled() bool {
switch strings.ToLower(strings.TrimSpace(os.Getenv(CachePrefixEnv))) {
case "1", "true", "on", "yes":
return true
}
return false
}
// workerSystemParts splits workerSystem at the first per-intent volatile
// segment (workerArtifactSpec(runDir)) without reordering any text:
// - staticPart = body + untrustedContentTail + workerTrafficBlock
// - dynamicHead = workerArtifactSpec(runDir) + noiseProfileBlock
//
// Custom DB prompts that reference {{.Now}} remain volatile inside staticPart.
func workerSystemParts(proxyAddr, caCert, dataDir, runDir string) (staticPart, dynamicHead string) {
body := renderSystem("worker", workerDefaultTmpl, WorkerVars{ProxyAddr: proxyAddr, DataDir: dataDir, Now: nowStr()})
// caCert is present only when the recording MITM is on, which is exactly when
// the traffic_* tools are registered — so it gates the traffic-tool note.
// Optional finding guidance is added for every role after tool resolution.
return body + untrustedContentTail(body) + workerTrafficBlock(caCert != "") + workerArtifactSpec(runDir) + noiseProfileBlock()
staticPart = body + untrustedContentTail(body) + workerTrafficBlock(caCert != "")
dynamicHead = workerArtifactSpec(runDir) + noiseProfileBlock()
return staticPart, dynamicHead
}
func workerSystem(proxyAddr, caCert, dataDir, runDir string) string {
staticPart, dynamicHead := workerSystemParts(proxyAddr, caCert, dataDir, runDir)
return staticPart + dynamicHead
}
// assembleWorkerSystem builds the worker's SystemPrompt segments and
// DynamicBoundary. When split is false (default), it delegates directly to
// deferredSystem(staticPart+dynamicTail, def). When split is true
// (ARTEX_CACHE_PREFIX=1), it emits two segments [staticPart, dynamic] with
// DynamicBoundary=1 so sibling workers in the same task share the cached static
// prefix, absorbing the leading "\n\n" separator from workerArtifactSpec so
// strings.Join(system, "\n\n") is byte-identical to the unsplit prompt.
func assembleWorkerSystem(staticPart, dynamicTail string, def DeferredInfo, split bool) (system []string, boundary int) {
if !split {
return deferredSystem(staticPart+dynamicTail, def)
}
dyn := strings.TrimPrefix(dynamicTail+def.FindingGuidance, "\n\n")
if block := actool.RenderDeferredToolsBlock(def.GlobalNames); block != "" {
dyn += "\n\n" + block
}
return []string{staticPart, dyn}, 1
}
// renderIntentTask formats the claimed intent for the worker's launch USER message:
@@ -524,18 +569,19 @@ func (w *Worker) execute(ctx context.Context, name string, taskID int64, as *db.
// inherit a parent run's background into the action reviewer.
ctx = intercept.WithReviewContext(ctx, runDir, intercept.ReviewBackground{})
overview := renderWorkerGraphOverview(tsx.graphOverviewData(), newDataEnvelope())
sysBody := workerSystem(w.proxyAddr, w.proxyCACert, w.workDir, runDir) + frameLessonsBlock(w.lessonsBlock)
staticPart, dynamicTail := workerSystemParts(w.proxyAddr, w.proxyCACert, w.workDir, runDir)
dynamicTail += frameLessonsBlock(w.lessonsBlock)
if w.wantConstraints() {
sysBody += constraintBlock(ts) // inject the operation constraints (if any) into the system prompt; the worker strictly obeys them during execution
dynamicTail += constraintBlock(ts) // inject the operation constraints (if any) into the system prompt; the worker strictly obeys them during execution
}
// intent block → intent-anchored assets block → launch instruction, appended to
// the system tail in that order (same append pattern as constraintBlock).
sysBody += renderIntentTask(intent)
dynamicTail += renderIntentTask(intent)
if as != nil {
if ids := intentAssetIDs(intent); len(ids) > 0 {
if assets, err := as.GetByIDs(ids); err == nil && len(assets) > 0 {
if b, err := json.Marshal(assets); err == nil {
sysBody += "\n\nThe target assets corresponding to this intent's asset_ids:\n" + string(b)
dynamicTail += "\n\nThe target assets corresponding to this intent's asset_ids:\n" + string(b)
}
// The assets this intent explicitly targets → automatically join the task's
// test scope (same conservative granularity as insertAssets). upsertTaskScope's
@@ -551,8 +597,8 @@ func (w *Worker) execute(ctx context.Context, name string, taskID int64, as *db.
}
}
}
sysBody += "\n\nStart executing the intent above: do only it, produce only facts, assets, findings, stop when done."
system, boundary := deferredSystem(sysBody, def)
dynamicTail += "\n\nStart executing the intent above: do only it, produce only facts, assets, findings, stop when done."
system, boundary := assembleWorkerSystem(staticPart, dynamicTail, def, cachePrefixEnabled())
// The task-level deadline (injected via ctx) clamps this run's wall-clock budget +
// decides the wrap-up prompt (see taskclock.go).
tc := taskClockFrom(ctx)
+326
View File
@@ -0,0 +1,326 @@
package agent
import (
"context"
"encoding/json"
"fmt"
"iter"
"path/filepath"
"slices"
"strings"
"sync"
"testing"
"time"
"github.com/Autumn-27/artex/db"
"github.com/Autumn-27/artex/llmrec"
"github.com/Autumn-27/norma/llm"
actool "github.com/Autumn-27/norma/tool"
)
type captureReqProvider struct {
mu sync.Mutex
reqs []llm.CompletionRequest
}
func (p *captureReqProvider) Stream(_ context.Context, req llm.CompletionRequest) iter.Seq2[llm.StreamEvent, error] {
p.mu.Lock()
p.reqs = append(p.reqs, req)
p.mu.Unlock()
return func(yield func(llm.StreamEvent, error) bool) {
if !yield(llm.StreamEvent{Type: llm.SEMessageStart, Usage: llm.Usage{InputTokens: 100}}, nil) {
return
}
if !yield(llm.StreamEvent{Type: llm.SETextDelta, Text: "done"}, nil) {
return
}
yield(llm.StreamEvent{Type: llm.SEMessageDelta, StopReason: "end_turn", Usage: llm.Usage{OutputTokens: 4}}, nil)
}
}
func (p *captureReqProvider) Complete(_ context.Context, req llm.CompletionRequest) (llm.Message, string, llm.Usage, error) {
p.mu.Lock()
p.reqs = append(p.reqs, req)
p.mu.Unlock()
return llm.Message{Role: llm.RoleAssistant, Content: []llm.ContentBlock{llm.TextBlock("done")}}, "end_turn", llm.Usage{InputTokens: 100, OutputTokens: 4}, nil
}
func (p *captureReqProvider) last() llm.CompletionRequest {
p.mu.Lock()
defer p.mu.Unlock()
if len(p.reqs) == 0 {
return llm.CompletionRequest{}
}
return p.reqs[len(p.reqs)-1]
}
func TestAssembleWorkerSystemFlagOffAndOnByteExact(t *testing.T) {
for _, tc := range []struct {
name string
caCert string
globalNames []string
guidance string
}{
{name: "no_deferred_no_traffic", caCert: "", globalNames: nil, guidance: ""},
{name: "with_deferred_and_traffic", caCert: "/ca.pem", globalNames: []string{"mcp__zap__scan", "mcp__burp__send"}, guidance: "\n\n[Finding guidance]"},
} {
t.Run(tc.name, func(t *testing.T) {
def := DeferredInfo{
GlobalNames: tc.globalNames,
FindingGuidance: tc.guidance,
}
buildTail := func(runDir, intentSummary, assetDomain string) (string, string) {
staticPart, dynamicTail := workerSystemParts("127.0.0.1:8788", tc.caCert, "/data", runDir)
dynamicTail += "\n\n## Prior evidence\n- lesson A"
dynamicTail += "\n\n[Constraints]\n- do not touch prod"
dynamicTail += renderIntentTask(&db.Node{ID: 7, Payload: json.RawMessage(`{"summary":"` + intentSummary + `"}`)})
dynamicTail += "\n\nThe target assets corresponding to this intent's asset_ids:\n[{\"domain\":\"" + assetDomain + "\"}]"
dynamicTail += "\n\nStart executing the intent above: do only it, produce only facts, assets, findings, stop when done."
return staticPart, dynamicTail
}
s1, d1 := buildTail("/data/tasks/1/i10", "probe auth", "alpha.example")
s2, d2 := buildTail("/data/tasks/1/i20", "probe billing", "beta.example")
// Legacy baseline (equivalent to flag off).
legacySys1, legacyBound1 := deferredSystem(workerSystem("127.0.0.1:8788", tc.caCert, "/data", "/data/tasks/1/i10")+strings.TrimPrefix(d1, workerArtifactSpec("/data/tasks/1/i10")+noiseProfileBlock()), def)
offSys1, offBound1 := assembleWorkerSystem(s1, d1, def, false)
if offBound1 != legacyBound1 || !slices.Equal(offSys1, legacySys1) {
t.Fatalf("flag off diverged from legacy: got (%d, len=%d) want (%d, len=%d)", offBound1, len(offSys1), legacyBound1, len(legacySys1))
}
// Flag ON: split into [static, dynamic] with DynamicBoundary=1.
onSys1, onBound1 := assembleWorkerSystem(s1, d1, def, true)
onSys2, onBound2 := assembleWorkerSystem(s2, d2, def, true)
if onBound1 != 1 || onBound2 != 1 || len(onSys1) != 2 || len(onSys2) != 2 {
t.Fatalf("flag on shape = (%d, %d) / (%d, %d), want boundary=1 len=2", onBound1, len(onSys1), onBound2, len(onSys2))
}
// Static segment identical across sibling workers.
if onSys1[0] != onSys2[0] {
t.Fatalf("static prefix differs between sibling workers")
}
// Dynamic segment differs and carries per-intent data + deferred block.
if onSys1[1] == onSys2[1] {
t.Fatalf("dynamic segment unexpectedly identical")
}
for _, want := range []string{"/data/tasks/1/i10", "probe auth", "alpha.example", "Start executing the intent above:"} {
if !strings.Contains(onSys1[1], want) {
t.Errorf("dynamic[1] missing %q", want)
}
if strings.Contains(onSys1[0], want) {
t.Errorf("static[0] unexpectedly contains volatile %q", want)
}
}
if len(tc.globalNames) > 0 {
if !strings.Contains(onSys1[1], "<available-deferred-tools>") || !strings.Contains(onSys1[1], "mcp__zap__scan") {
t.Errorf("deferred tools block missing from dynamic segment")
}
if strings.Contains(onSys1[0], "<available-deferred-tools>") {
t.Errorf("deferred tools block unexpectedly in static segment")
}
}
// Byte-exact visible text: Join(onSys, "\n\n") == Join(offSys, "\n\n").
wantVisible1 := strings.Join(offSys1, "\n\n")
gotVisible1 := strings.Join(onSys1, "\n\n")
if gotVisible1 != wantVisible1 {
t.Fatalf("visible text mismatch (len got=%d want=%d)", len(gotVisible1), len(wantVisible1))
}
offSys2, _ := assembleWorkerSystem(s2, d2, def, false)
if strings.Join(onSys2, "\n\n") != strings.Join(offSys2, "\n\n") {
t.Fatalf("worker 2 visible text mismatch")
}
})
}
}
func TestWorkerExecuteCachePrefixEndToEnd(t *testing.T) {
d := testDB(t)
t.Cleanup(func() { _ = d.Close() })
stamp := time.Now().UnixNano()
task, err := d.CreateTask(fmt.Sprintf("cache-prefix-%d", stamp), "verify worker cache prefix", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
t.Cleanup(func() { _ = d.DeleteTask(task.ID) })
ts := d.Exploration(task.ExplorationID)
as := d.Assets()
var asset1ID, asset2ID int64
dom1 := fmt.Sprintf("alpha-%d.cache.test", stamp)
dom2 := fmt.Sprintf("beta-%d.cache.test", stamp)
if err := d.QueryRow(`INSERT INTO assets(type,domain,root_domain,task_ids) VALUES('subdomain',$1,$1,ARRAY[$2]::bigint[]) RETURNING id`, dom1, task.ID).Scan(&asset1ID); err != nil {
t.Fatal(err)
}
if err := d.QueryRow(`INSERT INTO assets(type,domain,root_domain,task_ids) VALUES('subdomain',$1,$1,ARRAY[$2]::bigint[]) RETURNING id`, dom2, task.ID).Scan(&asset2ID); err != nil {
t.Fatal(err)
}
t.Cleanup(func() {
_, _ = d.Exec(`DELETE FROM assets WHERE id IN ($1,$2)`, asset1ID, asset2ID)
})
if _, err := ts.AddConstraint("deny", "stay within test scope", "human"); err != nil {
t.Fatal(err)
}
p1Map := map[string]any{"summary": "inspect alpha auth", "asset_ids": []int64{asset1ID}}
p2Map := map[string]any{"summary": "inspect beta checkout", "asset_ids": []int64{asset2ID}}
i1ID, err := ts.AddNode("intent", p1Map, 0, "running", "planner", nil)
if err != nil {
t.Fatal(err)
}
i2ID, err := ts.AddNode("intent", p2Map, 0, "running", "planner", nil)
if err != nil {
t.Fatal(err)
}
raw1, _ := json.Marshal(p1Map)
raw2, _ := json.Marshal(p2Map)
intent1 := &db.Node{ID: i1ID, Payload: raw1}
intent2 := &db.Node{ID: i2ID, Payload: raw2}
// Inject a global deferred MCP name so the deferred block is exercised end-to-end.
prevAugment := ToolAugment()
SetToolAugment(func(_ context.Context, _ string) ([]actool.CoreTool, DeferredInfo, func()) {
return nil, DeferredInfo{
GlobalNames: []string{"mcp__demo__scanner"},
}, func() {}
})
t.Cleanup(func() { SetToolAugment(prevAugment) })
workDir := t.TempDir()
runWorker := func(intent *db.Node, lessons string) llm.CompletionRequest {
t.Helper()
prov := &captureReqProvider{}
w := NewWorker(prov, "test-model", workDir, nil, 0, 2)
w.SetProxy("127.0.0.1:8788", "/ca.pem")
w.SetLessonsBlock(lessons)
if _, _, err := w.Execute(t.Context(), "work#1", task.ID, as, ts, intent, passHooks{}, nil, nil, nil); err != nil {
t.Fatalf("Worker.Execute(intent=%d): %v", intent.ID, err)
}
return prov.last()
}
// 1. Flag OFF (default): legacy segments and boundary. With a global deferred
// tool block, deferredSystem emits [sysText, deferredBlock] with boundary=2,
// and norma's UntrustedOutputBoundary (on by default) appends its fixed note as
// System[2], giving len=3 and boundary=2.
t.Setenv(CachePrefixEnv, "")
req1Off := runWorker(intent1, "")
req2Off := runWorker(intent2, "")
if req1Off.DynamicBoundary != 2 || len(req1Off.System) != 3 {
t.Fatalf("flag off worker 1: boundary=%d len=%d, want 2/3", req1Off.DynamicBoundary, len(req1Off.System))
}
if req1Off.System[0] == req2Off.System[0] {
t.Fatalf("expected flag-off worker system[0] to differ because runDir/intent are inside it")
}
// 2. Flag ON: static prefix identical across sibling workers, dynamic differs,
// visible text byte-identical, tools unchanged, deferred block inside dynamic,
// norma's appended untrusted-output note preserved as System[2].
t.Setenv(CachePrefixEnv, "1")
req1On := runWorker(intent1, "")
req2On := runWorker(intent2, "")
if req1On.DynamicBoundary != 1 || req2On.DynamicBoundary != 1 || len(req1On.System) != 3 || len(req2On.System) != 3 {
t.Fatalf("flag on shape: w1=(%d,%d) w2=(%d,%d), want (1,3)",
req1On.DynamicBoundary, len(req1On.System), req2On.DynamicBoundary, len(req2On.System))
}
if req1On.System[2] != req1Off.System[2] || req2On.System[2] != req2Off.System[2] {
t.Fatalf("norma untrusted-output note segment changed under ARTEX_CACHE_PREFIX=1")
}
if req1On.System[0] != req2On.System[0] {
t.Fatalf("flag on static prefix differs between sibling workers")
}
if req1On.System[1] == req2On.System[1] {
t.Fatalf("flag on dynamic suffix should differ between sibling workers")
}
runDir1 := filepath.Join(workDir, "tasks", fmt.Sprint(task.ID), fmt.Sprintf("i%d", intent1.ID))
runDir2 := filepath.Join(workDir, "tasks", fmt.Sprint(task.ID), fmt.Sprintf("i%d", intent2.ID))
for _, pair := range []struct {
name string
req llm.CompletionRequest
runDir string
intent string
domain string
}{
{"worker1", req1On, runDir1, "inspect alpha auth", dom1},
{"worker2", req2On, runDir2, "inspect beta checkout", dom2},
} {
for _, needle := range []string{pair.runDir, pair.intent, pair.domain, "stay within test scope", "Start executing the intent above:", "<available-deferred-tools>", "mcp__demo__scanner"} {
if !strings.Contains(pair.req.System[1], needle) {
t.Errorf("%s dynamic segment missing %q", pair.name, needle)
}
if strings.Contains(pair.req.System[0], needle) {
t.Errorf("%s static segment unexpectedly contains %q", pair.name, needle)
}
}
}
if strings.Join(req1On.System, "\n\n") != strings.Join(req1Off.System, "\n\n") {
t.Fatalf("worker 1 visible prompt changed under ARTEX_CACHE_PREFIX=1")
}
if strings.Join(req2On.System, "\n\n") != strings.Join(req2Off.System, "\n\n") {
t.Fatalf("worker 2 visible prompt changed under ARTEX_CACHE_PREFIX=1")
}
// With RSI lessons enabled (framed with a per-run random nonce), lessons stay in
// System[1] so System[0] remains byte-identical across sibling workers.
const sampleLessons = "- [positive, 1 evidence] sqli: verified"
req1Lessons := runWorker(intent1, sampleLessons)
req2Lessons := runWorker(intent2, sampleLessons)
if req1Lessons.System[0] != req1On.System[0] || req2Lessons.System[0] != req1On.System[0] {
t.Fatalf("lessons block leaked into static prefix")
}
if !strings.Contains(req1Lessons.System[1], sampleLessons) || !strings.Contains(req2Lessons.System[1], sampleLessons) {
t.Fatalf("lessons block missing from dynamic segment")
}
toolsOffJSON, _ := json.Marshal(req1Off.Tools)
toolsOnJSON, _ := json.Marshal(req1On.Tools)
if string(toolsOnJSON) != string(toolsOffJSON) {
t.Fatalf("tools changed under ARTEX_CACHE_PREFIX=1")
}
// Telemetry observes non-zero static & dynamic split with identical static bytes.
m1 := llmrec.MeasureContext(req1On)
m2 := llmrec.MeasureContext(req2On)
if m1.SystemStaticBytes <= 0 || m1.SystemDynamicBytes <= 0 || m1.SystemStaticBytes != m2.SystemStaticBytes {
t.Fatalf("MeasureContext mismatch: m1=(%d,%d) m2=(%d,%d)",
m1.SystemStaticBytes, m1.SystemDynamicBytes, m2.SystemStaticBytes, m2.SystemDynamicBytes)
}
// 3. Planner and mainagent are completely unchanged by ARTEX_CACHE_PREFIX=1.
runPlanner := func() llm.CompletionRequest {
prov := &captureReqProvider{}
if _, _, err := NewPlanner(prov, "test-model", workDir, nil, 0, 2).Plan(t.Context(), task.ID, as, ts, "goal", nil, passHooks{}, nil); err != nil {
t.Fatal(err)
}
return prov.last()
}
runMain := func() llm.CompletionRequest {
prov := &captureReqProvider{}
if _, err := NewMainAgent(prov, "test-model", workDir, nil, 0, 2).Chat(t.Context(), task.ID, 0, as, ts, "goal", "status?", passHooks{}, nil, nil, nil, nil, nil); err != nil {
t.Fatal(err)
}
return prov.last()
}
t.Setenv(CachePrefixEnv, "")
planOff := runPlanner()
mainOff := runMain()
t.Setenv(CachePrefixEnv, "1")
planOn := runPlanner()
mainOn := runMain()
if planOn.DynamicBoundary != planOff.DynamicBoundary || !slices.Equal(planOn.System, planOff.System) {
t.Fatalf("planner changed under ARTEX_CACHE_PREFIX=1: off=(%d,len=%d) on=(%d,len=%d)",
planOff.DynamicBoundary, len(planOff.System), planOn.DynamicBoundary, len(planOn.System))
}
if mainOn.DynamicBoundary != mainOff.DynamicBoundary || !slices.Equal(mainOn.System, mainOff.System) {
t.Fatalf("mainagent changed under ARTEX_CACHE_PREFIX=1: off=(%d,len=%d) on=(%d,len=%d)",
mainOff.DynamicBoundary, len(mainOff.System), mainOn.DynamicBoundary, len(mainOn.System))
}
}
+1 -1
View File
@@ -100,7 +100,7 @@ func TestEvidenceRewriteFlagMigrationAndRoundTrip(t *testing.T) {
t.Fatal(err)
}
for _, e := range entries {
if e.IsDir() || strings.HasPrefix(e.Name(), "0008_") {
if e.IsDir() || e.Name() >= "0008_" {
continue
}
data, err := fs.ReadFile(migrationsFS(), e.Name())
+316
View File
@@ -0,0 +1,316 @@
package db
import (
"fmt"
"strings"
)
// LLMContext is one opt-in context-size telemetry row written when
// ARTEX_CONTEXT_TELEMETRY=1. It records only byte sizes, counts, token usage,
// and call metadata — never any prompt, tool, or message content.
type LLMContext struct {
TaskID string `json:"task_id"`
Worker string `json:"worker"`
Model string `json:"model"`
Profile string `json:"profile"`
ToolCount int `json:"tool_count"`
MessageCount int `json:"message_count"`
SystemStaticBytes int64 `json:"system_static_bytes"`
SystemDynamicBytes int64 `json:"system_dynamic_bytes"`
ToolsBytes int64 `json:"tools_bytes"`
MessagesUserBytes int64 `json:"messages_user_bytes"`
MessagesAssistantBytes int64 `json:"messages_assistant_bytes"`
MessagesOtherBytes int64 `json:"messages_other_bytes"`
ToolResultBytes int64 `json:"tool_result_bytes"`
LastMessageBytes int64 `json:"last_message_bytes"`
InputTokens int `json:"input_tokens"`
OutputTokens int `json:"output_tokens"`
CacheRead int `json:"cache_read"`
CacheWrite int `json:"cache_write"`
Status string `json:"status"`
}
// TotalMeasuredBytes returns the partition sum of request section bytes
// (excluding ToolResultBytes, which is a subset of MessagesUserBytes).
func (c LLMContext) TotalMeasuredBytes() int64 {
return c.SystemStaticBytes + c.SystemDynamicBytes + c.ToolsBytes +
c.MessagesUserBytes + c.MessagesAssistantBytes + c.MessagesOtherBytes
}
// LLMContextSectionShares holds the pro-rata share (0..1) of real input_tokens
// attributed to each request section by byte proportion. The six partition
// fields (SystemStatic through MessagesOther) sum to 1.0 whenever at least one
// call has both measured bytes > 0 and input_tokens > 0. ToolResultSubset is an
// informational sub-share of MessagesUser (not added to the sum).
type LLMContextSectionShares struct {
SystemStatic float64 `json:"system_static"`
SystemDynamic float64 `json:"system_dynamic"`
Tools float64 `json:"tools"`
MessagesUser float64 `json:"messages_user"`
MessagesAsst float64 `json:"messages_assistant"`
MessagesOther float64 `json:"messages_other"`
ToolResultSubset float64 `json:"tool_result_subset"`
}
// ComputeLLMContextShares deterministically allocates each call's real
// input_tokens across sections proportionally to measured bytes, then divides
// each section's attributed tokens by the total attributed input tokens.
// Rows with zero measured bytes or zero input tokens are ignored so they never
// produce NaN or distort the proportions.
func ComputeLLMContextShares(rows []LLMContext) LLMContextSectionShares {
var (
totInput float64
sysStat float64
sysDyn float64
tools float64
msgUser float64
msgAsst float64
msgOther float64
toolRes float64
)
for _, r := range rows {
totalBytes := r.TotalMeasuredBytes()
if totalBytes <= 0 || r.InputTokens <= 0 {
continue
}
in := float64(r.InputTokens)
tb := float64(totalBytes)
totInput += in
sysStat += in * (float64(r.SystemStaticBytes) / tb)
sysDyn += in * (float64(r.SystemDynamicBytes) / tb)
tools += in * (float64(r.ToolsBytes) / tb)
msgUser += in * (float64(r.MessagesUserBytes) / tb)
msgAsst += in * (float64(r.MessagesAssistantBytes) / tb)
msgOther += in * (float64(r.MessagesOtherBytes) / tb)
toolRes += in * (float64(r.ToolResultBytes) / tb)
}
if totInput <= 0 {
return LLMContextSectionShares{}
}
return LLMContextSectionShares{
SystemStatic: sysStat / totInput,
SystemDynamic: sysDyn / totInput,
Tools: tools / totInput,
MessagesUser: msgUser / totInput,
MessagesAsst: msgAsst / totInput,
MessagesOther: msgOther / totInput,
ToolResultSubset: toolRes / totInput,
}
}
// LLMContextGroupStat aggregates llm_context rows for one (worker, model) pair.
type LLMContextGroupStat struct {
Worker string `json:"worker"`
Model string `json:"model"`
Calls int `json:"calls"`
InputTokensTotal int64 `json:"input_tokens_total"`
OutputTokensTotal int64 `json:"output_tokens_total"`
CacheReadTotal int64 `json:"cache_read_total"`
CacheWriteTotal int64 `json:"cache_write_total"`
CacheReadRatio float64 `json:"cache_read_ratio"`
InputP50 float64 `json:"input_p50"`
InputP90 float64 `json:"input_p90"`
InputP99 float64 `json:"input_p99"`
InputMax int `json:"input_max"`
AvgToolCount float64 `json:"avg_tool_count"`
AvgMessageCount float64 `json:"avg_message_count"`
AvgLastMessageBytes float64 `json:"avg_last_message_bytes"`
Shares LLMContextSectionShares `json:"shares"`
}
// LLMContextTaskStat summarizes context consumption for one task_id.
type LLMContextTaskStat struct {
TaskID string `json:"task_id"`
Calls int `json:"calls"`
InputTokensTotal int64 `json:"input_tokens_total"`
CacheReadTotal int64 `json:"cache_read_total"`
CacheReadRatio float64 `json:"cache_read_ratio"`
InputP90 float64 `json:"input_p90"`
InputMax int `json:"input_max"`
Shares LLMContextSectionShares `json:"shares"`
}
// LLMContextSummary is the aggregation output for Phase A context measurement.
type LLMContextSummary struct {
ByWorkerModel []LLMContextGroupStat `json:"by_worker_model"`
TopTasks []LLMContextTaskStat `json:"top_tasks"`
}
// LLMContextQuery scopes LLMContextSummary.
type LLMContextQuery struct {
Days int // > 0 restricts to ts >= now() - Days; <= 0 = all time
TaskID string // optional exact task_id filter
TopTasks int // max rows in TopTasks (default 20, clamped 1..100)
}
// InsertLLMContext appends one context-telemetry row. Best-effort: callers log
// and continue on error so telemetry never breaks an LLM call.
func (d *DB) InsertLLMContext(c *LLMContext) error {
status := strings.TrimSpace(c.Status)
if status == "" {
status = "ok"
}
_, err := d.Exec(`
INSERT INTO llm_context(
task_id, worker, model, profile,
tool_count, message_count,
system_static_bytes, system_dynamic_bytes, tools_bytes,
messages_user_bytes, messages_assistant_bytes, messages_other_bytes,
tool_result_bytes, last_message_bytes,
input_tokens, output_tokens, cache_read, cache_write, status
) VALUES (
NULLIF($1,''), NULLIF($2,''), NULLIF($3,''), NULLIF($4,''),
$5, $6,
$7, $8, $9,
$10, $11, $12,
$13, $14,
$15, $16, $17, $18, $19
)`,
c.TaskID, c.Worker, c.Model, c.Profile,
c.ToolCount, c.MessageCount,
c.SystemStaticBytes, c.SystemDynamicBytes, c.ToolsBytes,
c.MessagesUserBytes, c.MessagesAssistantBytes, c.MessagesOtherBytes,
c.ToolResultBytes, c.LastMessageBytes,
c.InputTokens, c.OutputTokens, c.CacheRead, c.CacheWrite, status)
return err
}
const llmContextTotalBytesExpr = `(system_static_bytes + system_dynamic_bytes + tools_bytes + messages_user_bytes + messages_assistant_bytes + messages_other_bytes)`
// llmContextSharesSelect produces the 7 pro-rata section-share SQL expressions
// matching ComputeLLMContextShares.
const llmContextSharesSelect = `
COALESCE(SUM(CASE WHEN ` + llmContextTotalBytesExpr + ` > 0 AND input_tokens > 0 THEN input_tokens * (system_static_bytes::double precision / ` + llmContextTotalBytesExpr + `::double precision) ELSE 0 END)
/ NULLIF(SUM(CASE WHEN ` + llmContextTotalBytesExpr + ` > 0 AND input_tokens > 0 THEN input_tokens::double precision ELSE 0 END), 0), 0),
COALESCE(SUM(CASE WHEN ` + llmContextTotalBytesExpr + ` > 0 AND input_tokens > 0 THEN input_tokens * (system_dynamic_bytes::double precision / ` + llmContextTotalBytesExpr + `::double precision) ELSE 0 END)
/ NULLIF(SUM(CASE WHEN ` + llmContextTotalBytesExpr + ` > 0 AND input_tokens > 0 THEN input_tokens::double precision ELSE 0 END), 0), 0),
COALESCE(SUM(CASE WHEN ` + llmContextTotalBytesExpr + ` > 0 AND input_tokens > 0 THEN input_tokens * (tools_bytes::double precision / ` + llmContextTotalBytesExpr + `::double precision) ELSE 0 END)
/ NULLIF(SUM(CASE WHEN ` + llmContextTotalBytesExpr + ` > 0 AND input_tokens > 0 THEN input_tokens::double precision ELSE 0 END), 0), 0),
COALESCE(SUM(CASE WHEN ` + llmContextTotalBytesExpr + ` > 0 AND input_tokens > 0 THEN input_tokens * (messages_user_bytes::double precision / ` + llmContextTotalBytesExpr + `::double precision) ELSE 0 END)
/ NULLIF(SUM(CASE WHEN ` + llmContextTotalBytesExpr + ` > 0 AND input_tokens > 0 THEN input_tokens::double precision ELSE 0 END), 0), 0),
COALESCE(SUM(CASE WHEN ` + llmContextTotalBytesExpr + ` > 0 AND input_tokens > 0 THEN input_tokens * (messages_assistant_bytes::double precision / ` + llmContextTotalBytesExpr + `::double precision) ELSE 0 END)
/ NULLIF(SUM(CASE WHEN ` + llmContextTotalBytesExpr + ` > 0 AND input_tokens > 0 THEN input_tokens::double precision ELSE 0 END), 0), 0),
COALESCE(SUM(CASE WHEN ` + llmContextTotalBytesExpr + ` > 0 AND input_tokens > 0 THEN input_tokens * (messages_other_bytes::double precision / ` + llmContextTotalBytesExpr + `::double precision) ELSE 0 END)
/ NULLIF(SUM(CASE WHEN ` + llmContextTotalBytesExpr + ` > 0 AND input_tokens > 0 THEN input_tokens::double precision ELSE 0 END), 0), 0),
COALESCE(SUM(CASE WHEN ` + llmContextTotalBytesExpr + ` > 0 AND input_tokens > 0 THEN input_tokens * (tool_result_bytes::double precision / ` + llmContextTotalBytesExpr + `::double precision) ELSE 0 END)
/ NULLIF(SUM(CASE WHEN ` + llmContextTotalBytesExpr + ` > 0 AND input_tokens > 0 THEN input_tokens::double precision ELSE 0 END), 0), 0)`
// LLMContextSummary aggregates llm_context by (worker, model) and returns the
// top tasks by total input_tokens, with input percentiles (p50/p90/p99),
// pro-rata section shares over real input_tokens, and cache_read/input ratio.
func (d *DB) LLMContextSummary(q LLMContextQuery) (*LLMContextSummary, error) {
topLimit := q.TopTasks
if topLimit <= 0 {
topLimit = 20
}
if topLimit > 100 {
topLimit = 100
}
var conds []string
var args []any
addCond := func(expr string, val any) {
args = append(args, val)
conds = append(conds, fmt.Sprintf(expr, len(args)))
}
if q.Days > 0 {
addCond("ts >= now() - make_interval(days => $%d)", q.Days)
}
if tid := strings.TrimSpace(q.TaskID); tid != "" {
addCond("COALESCE(task_id,'') = $%d", tid)
}
where := ""
if len(conds) > 0 {
where = " WHERE " + strings.Join(conds, " AND ")
}
out := &LLMContextSummary{
ByWorkerModel: []LLMContextGroupStat{},
TopTasks: []LLMContextTaskStat{},
}
groupSQL := `
SELECT COALESCE(NULLIF(worker,''),'(unknown)') AS w,
COALESCE(NULLIF(model,''),'(unknown)') AS m,
COUNT(*),
COALESCE(SUM(input_tokens), 0),
COALESCE(SUM(output_tokens), 0),
COALESCE(SUM(cache_read), 0),
COALESCE(SUM(cache_write), 0),
COALESCE(SUM(cache_read)::double precision / NULLIF(SUM(input_tokens)::double precision, 0), 0),
COALESCE(PERCENTILE_CONT(0.50) WITHIN GROUP (ORDER BY input_tokens), 0),
COALESCE(PERCENTILE_CONT(0.90) WITHIN GROUP (ORDER BY input_tokens), 0),
COALESCE(PERCENTILE_CONT(0.99) WITHIN GROUP (ORDER BY input_tokens), 0),
COALESCE(MAX(input_tokens), 0),
COALESCE(AVG(tool_count)::double precision, 0),
COALESCE(AVG(message_count)::double precision, 0),
COALESCE(AVG(last_message_bytes)::double precision, 0),` + llmContextSharesSelect + `
FROM llm_context` + where + `
GROUP BY w, m
ORDER BY SUM(input_tokens) DESC, w ASC, m ASC`
rows, err := d.Query(groupSQL, args...)
if err != nil {
return nil, err
}
defer rows.Close()
for rows.Next() {
var g LLMContextGroupStat
if err := rows.Scan(
&g.Worker, &g.Model, &g.Calls,
&g.InputTokensTotal, &g.OutputTokensTotal, &g.CacheReadTotal, &g.CacheWriteTotal,
&g.CacheReadRatio,
&g.InputP50, &g.InputP90, &g.InputP99, &g.InputMax,
&g.AvgToolCount, &g.AvgMessageCount, &g.AvgLastMessageBytes,
&g.Shares.SystemStatic, &g.Shares.SystemDynamic, &g.Shares.Tools,
&g.Shares.MessagesUser, &g.Shares.MessagesAsst, &g.Shares.MessagesOther,
&g.Shares.ToolResultSubset,
); err != nil {
return nil, err
}
out.ByWorkerModel = append(out.ByWorkerModel, g)
}
if err := rows.Err(); err != nil {
return nil, err
}
taskConds := append([]string{}, conds...)
taskConds = append(taskConds, "COALESCE(task_id,'') <> ''")
taskArgs := append([]any{}, args...)
taskArgs = append(taskArgs, topLimit)
limitPlaceholder := fmt.Sprintf("$%d", len(taskArgs))
taskSQL := `
SELECT task_id,
COUNT(*),
COALESCE(SUM(input_tokens), 0),
COALESCE(SUM(cache_read), 0),
COALESCE(SUM(cache_read)::double precision / NULLIF(SUM(input_tokens)::double precision, 0), 0),
COALESCE(PERCENTILE_CONT(0.90) WITHIN GROUP (ORDER BY input_tokens), 0),
COALESCE(MAX(input_tokens), 0),` + llmContextSharesSelect + `
FROM llm_context
WHERE ` + strings.Join(taskConds, " AND ") + `
GROUP BY task_id
ORDER BY SUM(input_tokens) DESC, task_id ASC
LIMIT ` + limitPlaceholder
trows, err := d.Query(taskSQL, taskArgs...)
if err != nil {
return nil, err
}
defer trows.Close()
for trows.Next() {
var ts LLMContextTaskStat
if err := trows.Scan(
&ts.TaskID, &ts.Calls, &ts.InputTokensTotal, &ts.CacheReadTotal,
&ts.CacheReadRatio, &ts.InputP90, &ts.InputMax,
&ts.Shares.SystemStatic, &ts.Shares.SystemDynamic, &ts.Shares.Tools,
&ts.Shares.MessagesUser, &ts.Shares.MessagesAsst, &ts.Shares.MessagesOther,
&ts.Shares.ToolResultSubset,
); err != nil {
return nil, err
}
out.TopTasks = append(out.TopTasks, ts)
}
return out, trows.Err()
}
+287
View File
@@ -0,0 +1,287 @@
package db
import (
"fmt"
"io/fs"
"math"
"testing"
"testing/fstest"
"time"
)
func TestComputeLLMContextSharesDeterministic(t *testing.T) {
// Call 1: 1000 bytes (400 sys_static, 100 sys_dyn, 200 tools, 200 user [150 tool_result], 100 asst), 1000 input tokens.
// Call 2: 2000 bytes (200 sys_static, 0 sys_dyn, 200 tools, 1400 user [1200 tool_result], 200 asst), 3000 input tokens.
// Call 3: zero tokens / zero bytes -> ignored safely (no NaN).
rows := []LLMContext{
{
SystemStaticBytes: 400, SystemDynamicBytes: 100, ToolsBytes: 200,
MessagesUserBytes: 200, MessagesAssistantBytes: 100, ToolResultBytes: 150,
InputTokens: 1000,
},
{
SystemStaticBytes: 200, SystemDynamicBytes: 0, ToolsBytes: 200,
MessagesUserBytes: 1400, MessagesAssistantBytes: 200, ToolResultBytes: 1200,
InputTokens: 3000,
},
{InputTokens: 0, SystemStaticBytes: 0},
}
got := ComputeLLMContextShares(rows)
// Exact hand-computed attributed tokens (total = 4000):
// sys_static: 400 + 300 = 700 -> 0.175
// sys_dynamic: 100 + 0 = 100 -> 0.025
// tools: 200 + 300 = 500 -> 0.125
// msg_user: 200 + 2100 = 2300 -> 0.575
// msg_asst: 100 + 300 = 400 -> 0.100
// msg_other: 0 = 0 -> 0.000
// tool_result: 150 + 1800 = 1950 -> 0.4875
assertNear := func(name string, v, want float64) {
t.Helper()
if math.IsNaN(v) || math.IsInf(v, 0) || math.Abs(v-want) > 1e-9 {
t.Errorf("%s = %.12f, want %.12f", name, v, want)
}
}
assertNear("SystemStatic", got.SystemStatic, 0.175)
assertNear("SystemDynamic", got.SystemDynamic, 0.025)
assertNear("Tools", got.Tools, 0.125)
assertNear("MessagesUser", got.MessagesUser, 0.575)
assertNear("MessagesAsst", got.MessagesAsst, 0.100)
assertNear("MessagesOther", got.MessagesOther, 0.0)
assertNear("ToolResultSubset", got.ToolResultSubset, 0.4875)
sum := got.SystemStatic + got.SystemDynamic + got.Tools + got.MessagesUser + got.MessagesAsst + got.MessagesOther
assertNear("partition sum", sum, 1.0)
// Empty input produces zero struct without NaN.
zero := ComputeLLMContextShares(nil)
if zero != (LLMContextSectionShares{}) {
t.Fatalf("empty shares = %+v, want zero", zero)
}
}
func TestLLMContextMigration0009AndSummaryAggregation(t *testing.T) {
dsn := scratchDSN(t, "ctx_mig_0009")
// 1. Open database with migrations 0001..0008 only (pre-0009 database).
pre0009 := fstest.MapFS{}
entries, err := fs.ReadDir(migrationsFS(), ".")
if err != nil {
t.Fatal(err)
}
for _, e := range entries {
if e.IsDir() || e.Name() >= "0009_" {
continue
}
data, err := fs.ReadFile(migrationsFS(), e.Name())
if err != nil {
t.Fatal(err)
}
pre0009[e.Name()] = &fstest.MapFile{Data: data}
}
d0, err := openWithMigrations(dsn, pre0009)
if err != nil {
t.Fatalf("open pre-0009: %v", err)
}
var existsBefore bool
if err := d0.QueryRow(`SELECT to_regclass('llm_context') IS NOT NULL`).Scan(&existsBefore); err != nil || existsBefore {
d0.Close()
t.Fatalf("llm_context should not exist before 0009: exists=%v err=%v", existsBefore, err)
}
d0.Close()
// 2. Re-open with full embedded migrations: 0009 must apply cleanly.
d, err := Open(dsn)
if err != nil {
t.Fatalf("Open with 0009: %v", err)
}
defer d.Close()
// 3. Seed deterministic rows across two workers and two tasks.
seed := []LLMContext{
{
TaskID: "t-heavy", Worker: "worker", Model: "claude-sonnet", Profile: "main",
ToolCount: 15, MessageCount: 4,
SystemStaticBytes: 400, SystemDynamicBytes: 100, ToolsBytes: 200,
MessagesUserBytes: 200, MessagesAssistantBytes: 100, ToolResultBytes: 150,
LastMessageBytes: 120,
InputTokens: 1000, OutputTokens: 200, CacheRead: 600, CacheWrite: 100, Status: "ok",
},
{
TaskID: "t-heavy", Worker: "worker", Model: "claude-sonnet", Profile: "main",
ToolCount: 15, MessageCount: 12,
SystemStaticBytes: 200, SystemDynamicBytes: 0, ToolsBytes: 200,
MessagesUserBytes: 1400, MessagesAssistantBytes: 200, ToolResultBytes: 1200,
LastMessageBytes: 800,
InputTokens: 3000, OutputTokens: 300, CacheRead: 1800, CacheWrite: 200, Status: "ok",
},
{
TaskID: "t-light", Worker: "planner", Model: "claude-sonnet", Profile: "main",
ToolCount: 10, MessageCount: 2,
SystemStaticBytes: 600, SystemDynamicBytes: 0, ToolsBytes: 200,
MessagesUserBytes: 200, MessagesAssistantBytes: 0, ToolResultBytes: 0,
LastMessageBytes: 200,
InputTokens: 500, OutputTokens: 100, CacheRead: 250, CacheWrite: 50, Status: "ok",
},
}
for i := range seed {
if err := d.InsertLLMContext(&seed[i]); err != nil {
t.Fatalf("InsertLLMContext[%d]: %v", i, err)
}
}
summary, err := d.LLMContextSummary(LLMContextQuery{Days: 7, TopTasks: 10})
if err != nil {
t.Fatalf("LLMContextSummary: %v", err)
}
if len(summary.ByWorkerModel) != 2 {
t.Fatalf("ByWorkerModel = %d, want 2: %+v", len(summary.ByWorkerModel), summary.ByWorkerModel)
}
// First group is worker/claude-sonnet (total input = 4000).
w := summary.ByWorkerModel[0]
wantShares := ComputeLLMContextShares(seed[:2])
near := func(name string, got, want float64) {
t.Helper()
if math.Abs(got-want) > 1e-9 {
t.Errorf("%s = %.12f, want %.12f", name, got, want)
}
}
if w.Worker != "worker" || w.Model != "claude-sonnet" || w.Calls != 2 ||
w.InputTokensTotal != 4000 || w.OutputTokensTotal != 500 ||
w.CacheReadTotal != 2400 || w.CacheWriteTotal != 300 || w.InputMax != 3000 {
t.Fatalf("worker group = %+v", w)
}
near("CacheReadRatio", w.CacheReadRatio, 2400.0/4000.0)
near("InputP50", w.InputP50, 2000.0)
near("InputP90", w.InputP90, 2800.0)
near("InputP99", w.InputP99, 2980.0)
near("Shares.SystemStatic", w.Shares.SystemStatic, wantShares.SystemStatic)
near("Shares.SystemDynamic", w.Shares.SystemDynamic, wantShares.SystemDynamic)
near("Shares.Tools", w.Shares.Tools, wantShares.Tools)
near("Shares.MessagesUser", w.Shares.MessagesUser, wantShares.MessagesUser)
near("Shares.MessagesAsst", w.Shares.MessagesAsst, wantShares.MessagesAsst)
near("Shares.ToolResultSubset", w.Shares.ToolResultSubset, wantShares.ToolResultSubset)
// TopTasks ordered by InputTokensTotal DESC: t-heavy (4000) then t-light (500).
if len(summary.TopTasks) != 2 || summary.TopTasks[0].TaskID != "t-heavy" || summary.TopTasks[1].TaskID != "t-light" {
t.Fatalf("TopTasks = %+v", summary.TopTasks)
}
// Task filter narrows to t-light only.
filtered, err := d.LLMContextSummary(LLMContextQuery{Days: 7, TaskID: "t-light", TopTasks: 5})
if err != nil {
t.Fatal(err)
}
if len(filtered.ByWorkerModel) != 1 || filtered.ByWorkerModel[0].Worker != "planner" || len(filtered.TopTasks) != 1 {
t.Fatalf("filtered summary = %+v", filtered)
}
}
// TestLLMContextRepresentativeFixtureReport runs a realistic multi-turn ARTEX
// task fixture (planner rounds + deep worker intent execution + main-agent
// follow-up) through InsertLLMContext and LLMContextSummary, verifying the
// aggregation pipeline end-to-end and logging reproducible Phase A numbers.
func TestLLMContextRepresentativeFixtureReport(t *testing.T) {
d, err := Open(testDSN(t))
if err != nil {
SkipPGUnavailable(t, err)
}
defer d.Close()
taskID := fmt.Sprintf("fixture-p35a-%d", time.Now().UnixNano())
t.Cleanup(func() { _, _ = d.Exec(`DELETE FROM llm_context WHERE task_id=$1`, taskID) })
// Representative byte sizes modeled from ARTEX's actual plannerSystem (~12 KiB),
// workerSystem (~11 KiB), tool schemas (~8-10 KiB for 12-16 tools), and real
// multi-turn exploration trajectories (token estimate ~3.8 bytes/token).
var fixture []LLMContext
// 5 planner rounds: ~12.5 KiB static system (includes {{.Now}} so cross-round cache misses
// on first turn of each round), ~9.2 KiB tools (14 tools), situational graph overview growing
// from 4 KiB to 14 KiB across rounds, 2 turns per round.
for round := range 5 {
overviewBytes := int64(4000 + round*2500)
// Turn 1 of round: fresh session, cache miss (cache_write ~system+tools, cache_read 0)
sysBytes := int64(12500)
toolsBytes := int64(9200)
in1 := int((sysBytes + toolsBytes + overviewBytes) / 4)
fixture = append(fixture, LLMContext{
TaskID: taskID, Worker: "planner", Model: "claude-sonnet-4", Profile: "fixture",
ToolCount: 14, MessageCount: 1,
SystemStaticBytes: sysBytes, ToolsBytes: toolsBytes,
MessagesUserBytes: overviewBytes, LastMessageBytes: overviewBytes,
InputTokens: in1, OutputTokens: 450, CacheRead: 0, CacheWrite: int((sysBytes + toolsBytes) / 4), Status: "ok",
})
// Turn 2 of round: after add_intent tool_result, prefix hits cache
asstBytes := int64(1400)
toolResBytes := int64(450)
in2 := int((sysBytes + toolsBytes + overviewBytes + asstBytes + toolResBytes) / 4)
fixture = append(fixture, LLMContext{
TaskID: taskID, Worker: "planner", Model: "claude-sonnet-4", Profile: "fixture",
ToolCount: 14, MessageCount: 3,
SystemStaticBytes: sysBytes, ToolsBytes: toolsBytes,
MessagesUserBytes: overviewBytes + toolResBytes, MessagesAssistantBytes: asstBytes,
ToolResultBytes: toolResBytes, LastMessageBytes: toolResBytes,
InputTokens: in2, OutputTokens: 320, CacheRead: in1, CacheWrite: 0, Status: "ok",
})
}
// 2 worker intents x 10 turns each: ~11 KiB system (embeds per-intent runDir + {{.Now}}),
// ~10.5 KiB tools (16 tools), growing Bash/WebFetch tool_result outputs (avg ~5 KiB/turn).
for range 2 {
sysBytes := int64(11000)
toolsBytes := int64(10500)
userBytes := int64(6500) // launch overview + claimed intent
asstBytes := int64(0)
toolResTotal := int64(0)
prevInput := 0
for turn := range 10 {
lastBytes := int64(6500)
if turn > 0 {
trChunk := int64(3500 + (turn%3)*2500) // realistic nmap/curl/ffuf tool outputs
asstChunk := int64(900)
userBytes += trChunk
toolResTotal += trChunk
asstBytes += asstChunk
lastBytes = trChunk
}
totalB := sysBytes + toolsBytes + userBytes + asstBytes
inTok := int(totalB / 4)
cacheRead := prevInput
cacheWrite := 0
if turn == 0 {
cacheRead = 0
cacheWrite = int((sysBytes + toolsBytes) / 4)
}
fixture = append(fixture, LLMContext{
TaskID: taskID, Worker: "worker", Model: "claude-sonnet-4", Profile: "fixture",
ToolCount: 16, MessageCount: 1 + turn*2,
SystemStaticBytes: sysBytes, ToolsBytes: toolsBytes,
MessagesUserBytes: userBytes, MessagesAssistantBytes: asstBytes,
ToolResultBytes: toolResTotal, LastMessageBytes: lastBytes,
InputTokens: inTok, OutputTokens: 380, CacheRead: cacheRead, CacheWrite: cacheWrite, Status: "ok",
})
prevInput = inTok
}
}
for i := range fixture {
if err := d.InsertLLMContext(&fixture[i]); err != nil {
t.Fatal(err)
}
}
summary, err := d.LLMContextSummary(LLMContextQuery{Days: 1, TaskID: taskID, TopTasks: 5})
if err != nil {
t.Fatal(err)
}
if len(summary.ByWorkerModel) != 2 || len(summary.TopTasks) != 1 {
t.Fatalf("unexpected fixture summary shape: %+v", summary)
}
for _, g := range summary.ByWorkerModel {
t.Logf("FIXTURE [%s/%s] calls=%d in_total=%d p50=%.0f p90=%.0f max=%d cache_read_ratio=%.3f | sys_static=%.1f%% sys_dyn=%.1f%% tools=%.1f%% user=%.1f%% (tool_result=%.1f%%) asst=%.1f%%",
g.Worker, g.Model, g.Calls, g.InputTokensTotal, g.InputP50, g.InputP90, g.InputMax, g.CacheReadRatio,
g.Shares.SystemStatic*100, g.Shares.SystemDynamic*100, g.Shares.Tools*100,
g.Shares.MessagesUser*100, g.Shares.ToolResultSubset*100, g.Shares.MessagesAsst*100)
}
}
+3 -3
View File
@@ -123,7 +123,7 @@ func readApplied(t *testing.T, d *DB) []appliedMigration {
// embedded migration is recorded with the SHA-256 of its file.
func assertEmbeddedApplied(t *testing.T, d *DB) []appliedMigration {
t.Helper()
for _, table := range []string{"llm_records", "llm_usage"} {
for _, table := range []string{"llm_records", "llm_usage", "llm_context"} {
var exists bool
if err := d.QueryRow(`SELECT to_regclass($1) IS NOT NULL`, table).Scan(&exists); err != nil {
t.Fatal(err)
@@ -138,9 +138,9 @@ func assertEmbeddedApplied(t *testing.T, d *DB) []appliedMigration {
name string
}{{1, llmRecordsMigration}, {2, llmUsageMigration}, {3, "0003_task_scope_icp_index.sql"}, {4, "0004_llm_usage_ts_index.sql"},
{5, "0005_traffic_rewrite_rules.sql"}, {6, "0006_activity_outbox.sql"}, {7, "0007_finding_report_state.sql"},
{8, "0008_traffic_evidence_rewrite_flag.sql"}}
{8, "0008_traffic_evidence_rewrite_flag.sql"}, {9, "0009_llm_context.sql"}}
if len(applied) != len(want) {
t.Fatalf("schema_migrations = %+v, want versions 1 to 8", applied)
t.Fatalf("schema_migrations = %+v, want versions 1 to 9", applied)
}
for i, w := range want {
content, err := fs.ReadFile(migrationsFS(), w.name)
+29
View File
@@ -0,0 +1,29 @@
-- Opt-in per-call LLM context size telemetry (ARTEX_CONTEXT_TELEMETRY=1).
-- Records only byte sizes, counts, token usage, and metadata — never any
-- prompt, tool, or message content.
CREATE TABLE IF NOT EXISTS llm_context (
id BIGSERIAL PRIMARY KEY,
ts TIMESTAMPTZ NOT NULL DEFAULT now(),
task_id TEXT,
worker TEXT,
model TEXT,
profile TEXT,
tool_count INTEGER NOT NULL DEFAULT 0,
message_count INTEGER NOT NULL DEFAULT 0,
system_static_bytes BIGINT NOT NULL DEFAULT 0,
system_dynamic_bytes BIGINT NOT NULL DEFAULT 0,
tools_bytes BIGINT NOT NULL DEFAULT 0,
messages_user_bytes BIGINT NOT NULL DEFAULT 0,
messages_assistant_bytes BIGINT NOT NULL DEFAULT 0,
messages_other_bytes BIGINT NOT NULL DEFAULT 0,
tool_result_bytes BIGINT NOT NULL DEFAULT 0,
last_message_bytes BIGINT NOT NULL DEFAULT 0,
input_tokens INTEGER NOT NULL DEFAULT 0,
output_tokens INTEGER NOT NULL DEFAULT 0,
cache_read INTEGER NOT NULL DEFAULT 0,
cache_write INTEGER NOT NULL DEFAULT 0,
status TEXT NOT NULL DEFAULT 'ok'
);
CREATE INDEX IF NOT EXISTS idx_llm_context_ts ON llm_context(ts);
CREATE INDEX IF NOT EXISTS idx_llm_context_task ON llm_context(task_id);
CREATE INDEX IF NOT EXISTS idx_llm_context_worker_model ON llm_context(worker, model);
+148
View File
@@ -0,0 +1,148 @@
# LLM Context Size Telemetry (P3.5 Phase A)
`ARTEX_CONTEXT_TELEMETRY=1` enables opt-in, size-only per-call telemetry for every `llmrec` `Stream` and `Complete` call. It is **disabled by default**: when off, no request sizing runs and no rows are written to PostgreSQL.
This phase measures context composition without changing any prompt, compaction policy, or model behavior, so Phases B (`ARTEX_BOUNDED_CONTEXT`) and C (`ARTEX_CACHE_PREFIX` / `ARTEX_COMPACTION_MEMO`) can be evaluated against real numbers first.
---
## Configuration
| Environment variable | Values | Default |
|---|---|---|
| `ARTEX_CONTEXT_TELEMETRY` | `1`, `true`, `on`, `yes` (case-insensitive) | off |
Telemetry runs independently of `llm_record` (`recordBodies`): enabling `ARTEX_CONTEXT_TELEMETRY=1` works even when full request/response body recording is disabled, and stores **no prompt, tool, or message content** — only byte lengths, counts, token usage, and metadata.
### Stopping, retention, and write cost
When `ARTEX_CONTEXT_TELEMETRY=1` is active, each LLM call incurs one lightweight `INSERT` into `llm_context` (~200 bytes/row plus index entries, no bodies), so the table grows by one row per completion call. Unset `ARTEX_CONTEXT_TELEMETRY` (and restart) once the measurement campaign is finished to stop all writes and sizing work, or purge old rows periodically:
```sql
DELETE FROM llm_context WHERE ts < now() - interval '30 days';
```
---
## Measurement Semantics (`llmrec.MeasureContext`)
Every enabled `Stream` and `Complete` call measures the assembled `llm.CompletionRequest` right before it is handed to the inner provider, and persists one row to `llm_context` (migration `0009_llm_context.sql`) on completion or error:
- **`system_static_bytes` & `system_dynamic_bytes`**:
- `norma`'s `CompletionRequest.System` is a slice of segments that `llm.joinSystem` trims (`strings.TrimSpace`), drops when empty, and joins with blank lines (`"\n\n"`). `MeasureContext` applies the exact same trim, drop-empty, and `"\n\n"` join rules per slice.
- When `DynamicBoundary > 0`, `System` is split at `b = min(DynamicBoundary, len(System))`: `System[:b]` is measured into `system_static_bytes` and `System[b:]` into `system_dynamic_bytes`.
- When `DynamicBoundary <= 0`, there is no provider-visible cache split, so the entire `System` slice is measured as one block in `system_static_bytes` and `system_dynamic_bytes` is `0`.
- **Why this choice**: `agent/deferred.go` returns `boundary = 0` when no global deferred-MCP block exists and `boundary = len(system)` when one does. Placing the single unsplit block in `system_static_bytes` ensures `system_static_bytes + system_dynamic_bytes` always equals total system bytes and `system_dynamic_bytes > 0` strictly signals a dynamic suffix placed after a cache breakpoint.
- **`tool_count` & `tools_bytes`**: `len(req.Tools)` and `len(json.Marshal(req.Tools))` (`0` when no tools). Go's `encoding/json` sorts map keys deterministically.
- **`message_count`, `messages_user_bytes`, `messages_assistant_bytes`, `messages_other_bytes`, `last_message_bytes`**: each `llm.Message` is measured by `len(json.Marshal(msg))` and accumulated by role (`user`, `assistant`, or other); `last_message_bytes` is the serialized byte size of the final message.
- **`tool_result_bytes`**: serialized byte length of `tool_result` content blocks inside messages. Reported as an **informational subset** of `messages_user_bytes` (not added to the partition denominator, avoiding double counting).
### Pro-rata token attribution
Rather than shipping per-provider tokenizers, each section's share of real `input_tokens` is computed per call by byte proportion:
```text
total_bytes(call) = system_static_bytes + system_dynamic_bytes + tools_bytes
+ messages_user_bytes + messages_assistant_bytes + messages_other_bytes
share(section) = SUM(input_tokens * (section_bytes / total_bytes))
/ SUM(input_tokens) -- over calls with total_bytes > 0 and input_tokens > 0
```
The six partition shares sum to `1.0`. `cache_read_ratio = SUM(cache_read) / NULLIF(SUM(input_tokens), 0)` (in ARTEX/Norma, `input_tokens` already includes cached tokens).
---
## Querying & Aggregation
### HTTP endpoint (admin JWT required, no UI in Phase A)
```text
GET /api/llm/context/summary?days=30&top=20&task=<optional-task-id>
```
Returns `enabled`, `by_worker_model` (calls, totals, `cache_read_ratio`, `input_p50`/`p90`/`p99`/`max`, avg tool/message/last-message size, and pro-rata `shares`), and `top_tasks` ordered by total `input_tokens`.
### Standalone SQL query (`psql`)
```sql
WITH base AS (
SELECT *,
(system_static_bytes + system_dynamic_bytes + tools_bytes
+ messages_user_bytes + messages_assistant_bytes + messages_other_bytes) AS total_bytes
FROM llm_context
WHERE ts >= now() - interval '30 days'
)
SELECT COALESCE(NULLIF(worker,''), '(unknown)') AS worker,
COALESCE(NULLIF(model,''), '(unknown)') AS model,
COUNT(*) AS calls,
SUM(input_tokens) AS input_tokens,
ROUND((SUM(cache_read)::numeric / NULLIF(SUM(input_tokens), 0))::numeric, 3) AS cache_read_ratio,
PERCENTILE_CONT(0.50) WITHIN GROUP (ORDER BY input_tokens) AS p50_input,
PERCENTILE_CONT(0.90) WITHIN GROUP (ORDER BY input_tokens) AS p90_input,
PERCENTILE_CONT(0.99) WITHIN GROUP (ORDER BY input_tokens) AS p99_input,
ROUND((SUM(CASE WHEN total_bytes > 0 AND input_tokens > 0 THEN input_tokens * (system_static_bytes::double precision / total_bytes) ELSE 0 END)
/ NULLIF(SUM(CASE WHEN total_bytes > 0 AND input_tokens > 0 THEN input_tokens::double precision ELSE 0 END), 0))::numeric, 3) AS share_sys_static,
ROUND((SUM(CASE WHEN total_bytes > 0 AND input_tokens > 0 THEN input_tokens * (system_dynamic_bytes::double precision / total_bytes) ELSE 0 END)
/ NULLIF(SUM(CASE WHEN total_bytes > 0 AND input_tokens > 0 THEN input_tokens::double precision ELSE 0 END), 0))::numeric, 3) AS share_sys_dynamic,
ROUND((SUM(CASE WHEN total_bytes > 0 AND input_tokens > 0 THEN input_tokens * (tools_bytes::double precision / total_bytes) ELSE 0 END)
/ NULLIF(SUM(CASE WHEN total_bytes > 0 AND input_tokens > 0 THEN input_tokens::double precision ELSE 0 END), 0))::numeric, 3) AS share_tools,
ROUND((SUM(CASE WHEN total_bytes > 0 AND input_tokens > 0 THEN input_tokens * (messages_user_bytes::double precision / total_bytes) ELSE 0 END)
/ NULLIF(SUM(CASE WHEN total_bytes > 0 AND input_tokens > 0 THEN input_tokens::double precision ELSE 0 END), 0))::numeric, 3) AS share_user,
ROUND((SUM(CASE WHEN total_bytes > 0 AND input_tokens > 0 THEN input_tokens * (tool_result_bytes::double precision / total_bytes) ELSE 0 END)
/ NULLIF(SUM(CASE WHEN total_bytes > 0 AND input_tokens > 0 THEN input_tokens::double precision ELSE 0 END), 0))::numeric, 3) AS share_tool_result_subset,
ROUND((SUM(CASE WHEN total_bytes > 0 AND input_tokens > 0 THEN input_tokens * (messages_assistant_bytes::double precision / total_bytes) ELSE 0 END)
/ NULLIF(SUM(CASE WHEN total_bytes > 0 AND input_tokens > 0 THEN input_tokens::double precision ELSE 0 END), 0))::numeric, 3) AS share_assistant
FROM base
GROUP BY 1, 2
ORDER BY input_tokens DESC;
```
---
## Phase A Baseline Report & Recommendation on Phases B and C
> **Measurement scope note**: No live production LLM run was executed in this local verification environment. The figures below come from the reproducible representative fixture in `db/llm_context_test.go` (`TestLLMContextRepresentativeFixtureReport`), parameterized from ARTEX's actual `plannerSystem` (~12.5 KiB), `workerSystem` (~11 KiB), tool schema sets (14–16 tools, ~9–10.5 KiB), 5 planner rounds (2 turns each), and 2 worker intents (10 turns each with 3.5–8.5 KiB Bash/tool outputs), combined with a static audit of `agent/planner.go`, `agent/worker.go`, `agent/prompt.go`, and `agent/deferred.go`. Operators should capture 3–5 real tasks with `ARTEX_CONTEXT_TELEMETRY=1` before flipping Phases B or C on by default.
### Representative Fixture Breakdown
| Lane / Model | Calls | Total Input | p50 / p90 / Max | Cache Read Ratio | `sys_static` | `sys_dynamic` | `tools` | `user` (`tool_result` subset) | `assistant` |
|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|
| `worker` / `claude-sonnet-4` | 20 | 299,000 | 15,075 / 21,535 / 22,525 | 84.9% | 18.4% | 0.0% | 17.6% | 57.3% (`46.4%`) | 6.8% |
| `planner` / `claude-sonnet-4` | 10 | 79,060 | 7,906 / 8,971 / 9,387 | 48.5% | 39.5% | 0.0% | 29.1% | 29.2% (`0.7%`) | 2.2% |
### Explicit Recommendations
#### Phase B — Bounded Context (`ARTEX_BOUNDED_CONTEXT`)
- **Recommendation: Proceed (opt-in first), highest ROI on deep `worker` runs.**
- **Expected gain**: In the representative worker trajectory, `messages_user` accounts for **57.3%** of total input tokens, with **`tool_result` outputs alone representing 46.4%** (and growing quadratically across turns on long Bash/scanner sessions). Bounding/compacting old tool outputs on workers (keeping recent turns verbatim and summarizing/truncating stale `tool_result` payloads) targets the largest single token sink (realistic 25–45% input-token reduction on 10–30 turn workers, and lower p90/p99 latency).
- **Risk**: Aggressive trimming of earlier tool outputs can drop credentials, endpoints, or exact error lines needed for exploit chaining or evidence binding, and mutating earlier messages inside a live multi-turn session **invalidates the prompt cache prefix** (turning cheap `cache_read` turns into uncached re-prefills if triggered too frequently). Bound only at explicit compaction boundaries or spill large `tool_result` bodies at insertion time before they enter the cached prefix.
#### Phase C — Cache Prefix Stabilization (`ARTEX_CACHE_PREFIX` / `ARTEX_COMPACTION_MEMO`)
- **Recommendation: Proceed with a surgical `worker` prefix split (`ARTEX_CACHE_PREFIX`); verify `planner` cross-round cache behavior and `ARTEX_COMPACTION_MEMO` with live `ARTEX_CONTEXT_TELEMETRY=1` data before changing them.**
- **Actual system-prompt volatility by lane (built-in prompts)**:
`system_dynamic` is **0.0%** across all lanes today because `agent/deferred.go` sets `DynamicBoundary = 0` (no global deferred MCP) or `len(system)` (everything before the boundary). Auditing the built-in templates and assembly in `agent/worker.go` and `agent/planner.go` shows what actually varies:
| Lane | Stable per task | Volatile per intent / round |
|---|---|---|
| `worker` | Base `workerDefaultTmpl` (`ProxyAddr`, `DataDir`), `untrustedContentTail`, `workerTrafficBlock`, `noiseProfileBlock`, `frameLessonsBlock` (task-level), `constraintBlock` (task-level, stable unless operator edits constraints), `def.FindingGuidance`, `<available-deferred-tools>` | `workerArtifactSpec(runDir)` (`tasks/<id>/i<intentID>`), `renderIntentTask(intent)`, intent target-assets JSON block, closing launch instruction |
| `planner` | `plannerSystem` (`goal`, `dataDir`, `workDir`/`taskDir`, `untrustedDataRules`, `artifactSpec(taskDir)`, `noiseProfileBlock`), `frameLessonsBlock`, `def.FindingGuidance`, `<available-deferred-tools>` | `constraintBlock(ts)` is recomputed each round (stable unless operator changes constraints; situational graph overview already lives in the user turn) |
| Custom prompts only | — | `{{.Now}}` (`agent/prompt.go:46`, 1-second resolution). Built-in templates do **not** reference `{{.Now}}` (Go's `text/template` ignores the unused `Now` field); it only affects operator-customized prompts that explicitly use `{{.Now}}`. |
- **Expected gain**:
- **`worker` turn-1 cache reuse (real, code-backed)**: Today `workerArtifactSpec(runDir)` is concatenated into the single `sysBody` before `noiseProfileBlock`, lessons, constraints, `renderIntentTask(intent)`, and the target-assets block, so every sibling worker spawn busts the prompt cache from `runDir` onward. The minimal `ARTEX_CACHE_PREFIX=1` split shipped in Phase C places the prefix before `workerArtifactSpec(runDir)` (`body` + `untrustedContentTail` + `workerTrafficBlock`) **before** `DynamicBoundary = 1` (`system_static`) and leaves everything from `workerArtifactSpec(runDir)` onward (`workerArtifactSpec(runDir)` + `noiseProfileBlock` + lessons + constraints + `renderIntentTask(intent)` + target assets JSON + closing launch line + `def.FindingGuidance` + `<available-deferred-tools>`) **after** `DynamicBoundary = 1` (`system_dynamic`) in its exact original order. This keeps the visible joined prompt byte-identical and keeps the intent protected from compaction inside `System` while allowing sibling workers in the same task to share the cached `system_static + tools` prefix on turn 1 (in the fixture, `system + tools` is 36.0% of aggregate worker input and ~76.8% of turn-1 input, 21.5 KiB / 28.0 KiB, though the fixture does not isolate the per-intent suffix; measure the exact shareable prefix live). Moving the remaining task-stable blocks (`noiseProfileBlock`, lessons, constraints, `def.FindingGuidance`, `<available-deferred-tools>`) before the boundary would require reordering sections relative to `workerArtifactSpec(runDir)` and is left as a potential future optimization outside this PR.
- **`planner` cross-round cache ratio (unverified hypothesis)**: The fixture above assumed a turn-1 cache miss on every planner round (yielding 48.5%), which is **not** demonstrated by the built-in code because `plannerSystem` is stable per task. Real cross-round misses could still happen from provider cache TTL expiry between rounds, constraint edits, or tool-RAG tool-list churn — treat any planner cache gain as a hypothesis to measure with `ARTEX_CONTEXT_TELEMETRY=1` on real tasks rather than a proven savings.
- **Risk**: Low for the `worker` `ARTEX_CACHE_PREFIX` split (pure boundary split using `norma`'s existing `DynamicBoundary` contract, with zero text reordering; custom prompts using `{{.Now}}` stay an explicit edge case). Medium for `ARTEX_COMPACTION_MEMO` (injecting compaction memos alters model reasoning continuity and only pays off on tasks that repeatedly trigger compaction — verify with live `ARTEX_CONTEXT_TELEMETRY=1` data first).
---
## Phase C: Worker Cache Prefix Split (`ARTEX_CACHE_PREFIX=1`)
`ARTEX_CACHE_PREFIX=1` (accepts `1`, `true`, `on`, `yes`, case-insensitive; **off by default**) enables a minimal, zero-reordering system-prompt split for the **`worker`** lane only so sibling workers in the same task share a cacheable static prefix without changing the text visible to the model:
- **Static segment (`System[0]`, before `DynamicBoundary = 1`)**: `body + untrustedContentTail + workerTrafficBlock`.
- **Dynamic segment (`System[1]`, after `DynamicBoundary = 1`)**: `workerArtifactSpec(runDir) + noiseProfileBlock + lessons + constraints + renderIntentTask(intent) + target assets + launch instruction + FindingGuidance + <available-deferred-tools>` (in exact original order). The leading `"\n\n"` from `workerArtifactSpec` is absorbed because providers join segments with `"\n\n"`, keeping `strings.Join(System, "\n\n")` byte-identical to the legacy unsplit prompt.
- **Unchanged**: When `ARTEX_CACHE_PREFIX` is unset/false, worker segments and `DynamicBoundary` are byte-identical to the legacy layout. `planner`, `mainagent`, `chat`, and `goals` are completely unchanged regardless of the flag.
- **Custom prompts caveat**: The split covers only the built-in suffix starting at `workerArtifactSpec(runDir)`. If an operator customizes the `worker` prompt in the database to include `{{.Now}}` (1-second resolution), that timestamp remains inside `System[0]` and will keep the static prefix volatile across spawns.
- **Measuring live impact**: Enable both `ARTEX_CONTEXT_TELEMETRY=1` and `ARTEX_CACHE_PREFIX=1` on real tasks and compare `system_static` / `system_dynamic` shares and `cache_read_ratio` via `GET /api/llm/context/summary`.
+25
View File
@@ -0,0 +1,25 @@
[Unit]
Description=ARTEX application
After=network.target
StartLimitIntervalSec=60
StartLimitBurst=5
[Service]
Type=simple
User=artex
Group=artex
WorkingDirectory=/opt/artex
Environment=ARTEX_JWT_KEY_DIR=/opt/artex/keys
ExecStart=/opt/artex/artex -addr 127.0.0.1:8787 -proxy 127.0.0.1:8788
Restart=on-failure
RestartSec=5s
TimeoutStopSec=30s
KillSignal=SIGTERM
KillMode=control-group
UMask=0002
StandardOutput=journal
StandardError=journal
SyslogIdentifier=artex
[Install]
WantedBy=multi-user.target
+178
View File
@@ -0,0 +1,178 @@
# Local ARTEX systemd deployment
This guide describes how to deploy ARTEX locally under `systemd` using the example unit at `docs/examples/artex.service`.
## Placeholders and defaults
The example unit uses these defaults; replace them if your host uses different paths or names:
- Install directory (`<install-dir>`): `/opt/artex` (executable at `/opt/artex/artex`).
- System user and group (`<artex-user>`): `artex` (non-root).
- PostgreSQL service unit (`<pg-unit>`): distribution-specific — for example `postgresql.service` on RHEL/Fedora/Arch or `postgresql@<pg-version>-<pg-cluster>.service` on Debian/Ubuntu (where `<pg-version>/<pg-cluster>` comes from `pg_lsclusters`, such as `18/main`).
- Configuration: `<install-dir>/config.json`, ignored by Git, owner-only mode `0600`.
- Web UI: `http://127.0.0.1:8787`; recording proxy: `127.0.0.1:8788`.
- Example unit file: `docs/examples/artex.service`; installed at `/etc/systemd/system/artex.service`.
## System user and directory permissions
Create a dedicated non-login system user and group before installing the binary:
```bash
sudo useradd --system --user-group --home-dir /opt/artex --shell /usr/sbin/nologin artex
```
Create the install, runtime data, and key directories with explicit ownership and modes:
```bash
sudo install -d -o root -g artex -m 0755 /opt/artex
sudo install -d -o artex -g artex -m 0770 /opt/artex/data
sudo install -d -o artex -g artex -m 0700 /opt/artex/keys
```
Directory ownership and modes:
- `<install-dir>` (`/opt/artex`): `root:<artex-user>`, mode `0755` (executable `/opt/artex/artex` mode `0755`).
- `<install-dir>/data` (`/opt/artex/data`): `<artex-user>:<artex-user>`, mode `0770` so the service user (and, in a root-run deployment using an executor group, that group) can write task run directories while blocking world access.
- `<install-dir>/keys` (`/opt/artex/keys`): `<artex-user>:<artex-user>`, mode `0700`. Holds `jwt.key` and `secret.key` (mode `0600`) outside the browsable `data/` directory. `docs/examples/artex.service` sets `Environment=ARTEX_JWT_KEY_DIR=/opt/artex/keys` so the non-root service writes key files here rather than into `root`-owned `/opt/artex` (if `ARTEX_JWT_KEY_DIR` were left unset, ARTEX would fall back to `<install-dir>` and fail to create `jwt.key` and `secret.key` when `<install-dir>` is `root:<artex-user>` `0755`).
- `<install-dir>/config.json` (`/opt/artex/config.json`): `<artex-user>:<artex-user>`, mode `0600`.
## Database and credentials
Install PostgreSQL on the host (or use an existing reachable PostgreSQL instance) and create a dedicated non-superuser login role and database named `artex` with no role-creation or database-creation privileges.
On Debian/Ubuntu, confirm the cluster name (`<pg-version>/<pg-cluster>`) with `pg_lsclusters`, start `<pg-unit>`, and create the role and database if absent:
```bash
sudo systemctl start <pg-unit>
sudo -u postgres createuser --pwprompt artex
sudo -u postgres createdb --owner=artex artex
```
Enter a unique password at the prompt. In `<install-dir>/config.json`, set `database.host` to `127.0.0.1`, `port` to `5432`, `user` and `dbname` to `artex`, `password` to that secret, and `sslmode` to `disable` (disabling TLS is appropriate only for loopback traffic). Preserve other settings and restrict access:
```bash
sudo chown artex:artex /opt/artex/config.json
sudo chmod 0600 /opt/artex/config.json
```
`ARTEX_PG_DSN`, if set in a systemd drop-in override, takes precedence over `config.json`. Never commit configuration, credentials, encryption/signing keys, or startup logs.
## Build and install
The binary embeds the UI when built with `-tags embedui`. To rebuild the UI from a source checkout, run `npm ci` and `npm run build:static` in `web/`, create the parent directory with `mkdir -p server/webui`, and replace `server/webui/dist` with `web/out`. Build the binary with:
```bash
CGO_ENABLED=0 go build -tags embedui -trimpath -o artex ./cmd/artex
```
Install the binary into `<install-dir>`:
```bash
sudo install -o root -g artex -m 0755 artex /opt/artex/artex
```
For a running deployment, build into a separate staging path and stop the service before replacing `/opt/artex/artex` or generated assets. Avoid rerunning `install.sh` without reviewing its prompts: it regenerates `config.json` and starts a separate process.
Review `User`, `Group`, `WorkingDirectory`, `Environment=ARTEX_JWT_KEY_DIR=...`, and `ExecStart` in `docs/examples/artex.service` before installing it on your host:
```bash
systemd-analyze verify docs/examples/artex.service
sudo install -o root -g root -m 0644 docs/examples/artex.service /etc/systemd/system/artex.service
sudo systemctl daemon-reload
```
### Database dependency and environment overrides
`docs/examples/artex.service` intentionally declares only `After=network.target` and does not hardcode a distribution-specific PostgreSQL unit (such as Debian's `postgresql@<pg-version>-<pg-cluster>.service`), so it works across Linux distributions and with remote PostgreSQL servers.
When PostgreSQL runs locally on the same host, add a systemd drop-in (`sudo systemctl edit artex.service`) so ARTEX starts after and depends on your local PostgreSQL unit:
```ini
[Unit]
Requires=<pg-unit>
After=<pg-unit>
```
For example, set `<pg-unit>` to `postgresql.service` on RHEL/Fedora/Arch, or to `postgresql@<pg-version>-<pg-cluster>.service` on Debian/Ubuntu (and verify `/etc/postgresql/<pg-version>/<pg-cluster>/start.conf` is `auto`, not `disabled`). `ARTEX_JWT_KEY_DIR=/opt/artex/keys` is already set in `docs/examples/artex.service`; override it in the drop-in only if you install ARTEX under a different `<install-dir>`. Run `sudo systemctl daemon-reload` after editing the drop-in.
### Enable and start
If ARTEX is running outside `systemd`, identify its PID/executable and send that verified process `SIGTERM`. Wait for shutdown and release of ports `8787` and `8788` before enabling the service:
```bash
ss -ltnp '( sport = :8787 or sport = :8788 )'
sudo systemctl enable --now artex.service
systemctl status artex.service --no-pager
curl --fail --silent --show-error http://127.0.0.1:8787/api/health
```
The unit runs `/opt/artex/artex` directly without an extra supervisor script. It starts after `network.target`, restarts on failure after five seconds, allows 30 seconds for graceful shutdown, logs to the journal, and uses `UMask=0002` (rather than `0077`, which would block group writes in a root-run deployment where executor privilege drop sets umask `002` for the executor group; sensitive files such as `config.json`, `keys/jwt.key`, and `keys/secret.key` are protected explicitly with `0600`/`0700` permissions). Five starts within 60 seconds trigger the start limit.
### Agent isolation
Because `docs/examples/artex.service` runs the server as the non-root `<artex-user>` (`User=artex`), the `-exec-uid`/`-exec-gid` privilege drop cannot be applied: `ResolveExecPrivilege` disables the drop when the server is not root (and errors if those flags are passed explicitly). The example unit deliberately omits `-exec-uid`/`-exec-gid`, so agents and spawned tools run under the `<artex-user>` service identity rather than a separate executor account. The root-owned run-directory layout and executor-group containment exist only when the server runs as root with explicit `-exec-uid`/`-exec-gid`; if you require unprivileged-executor containment, use the Docker deployment, which is the supported path for that isolation model.
## Web setup
On a new database, visit `http://127.0.0.1:8787/setup`, provide the one-time setup token from the private startup log or journal (`journalctl -u artex.service`), and choose a unique password of at least 12 characters. The username is `ARTEX`. Store the password in your password manager rather than in a plaintext file on disk.
## Reboot persistence verification (example)
Pre-reboot checks confirm persistent ARTEX enablement, active PostgreSQL and ARTEX services, HTTP 200 health, and initialized authentication, but do not by themselves prove survival across a reboot.
To verify reboot persistence on a host, record the current boot ID and timestamp in an ignored runtime file such as `<install-dir>/data/reboot-verification-baseline.json` before rebooting, and do not overwrite that baseline until the test is complete:
```bash
sudo -u artex python3 -c 'import json, time; from pathlib import Path; Path("/opt/artex/data/reboot-verification-baseline.json").write_text(json.dumps({"boot_id": Path("/proc/sys/kernel/random/boot_id").read_text().strip(), "ts": time.time()}))'
```
Save other work before rebooting; this disconnects local and remote sessions:
```bash
sudo systemctl reboot
```
After reconnecting, do not manually start ARTEX before checking automatic startup. Compare the current boot ID with the saved baseline and check the services:
```bash
cd /opt/artex
python3 -c 'import json; from pathlib import Path; old=json.loads(Path("data/reboot-verification-baseline.json").read_text())["boot_id"]; new=Path("/proc/sys/kernel/random/boot_id").read_text().strip(); print("PASS: a reboot occurred" if old != new else "FAIL: no reboot occurred"); raise SystemExit(0 if old != new else 1)'
systemctl is-enabled artex.service
systemctl is-active <pg-unit> artex.service
curl --fail --silent --show-error http://127.0.0.1:8787/api/health
python3 -c 'import json, urllib.request; r=json.load(urllib.request.urlopen("http://127.0.0.1:8787/api/auth/status", timeout=5)); print("Authentication initialized:", r.get("initialized")); raise SystemExit(0 if r.get("initialized") else 1)'
systemctl show artex.service --no-pager -p ActiveState -p SubState -p MainPID -p ActiveEnterTimestamp
```
Success requires a changed boot ID, enabled ARTEX, both services active without manual startup, HTTP 200, and authentication still initialized. A new PID alone is not proof of reboot. If the system has just booted, allow startup time and repeat health checks. Refresh the baseline before running a future independent reboot test.
### Example reboot verification record
Example of a completed post-reboot verification (recorded from the original Debian/PostgreSQL 18 reference deployment on 2026-11-17, without manually starting or restarting either service):
- Current boot ID differed from the saved pre-reboot baseline, confirming an actual reboot.
- `artex.service` remained enabled and was `active (running)` under its non-root service user with `NRestarts=0`.
- The local PostgreSQL cluster service was `active`.
- `/api/health` returned HTTP 200.
- `/api/auth/status` confirmed initialized authentication was preserved.
Observed timestamps, PIDs, and restart counts are host-specific examples, not fixed configuration values.
## Operations and troubleshooting
```bash
sudo systemctl restart artex.service
sudo systemctl stop artex.service
sudo systemctl start artex.service
journalctl -b -u artex.service --no-pager
journalctl -u artex.service -f
```
Treat journal output as potentially sensitive; never share unredacted setup tokens or credentials. If startup fails, check PostgreSQL status, database settings and permissions on `<install-dir>/config.json`, `<install-dir>/data`, and `<install-dir>/keys`, occupied ports (`8787`/`8788`), and the journal. After a PostgreSQL dependency failure, restart ARTEX explicitly once the database is healthy. To recover from the start limit:
```bash
sudo systemctl reset-failed artex.service
sudo systemctl start artex.service
```
Back up PostgreSQL, `<install-dir>/config.json`, `<install-dir>/data`, and `<install-dir>/keys` (`jwt.key` and `secret.key`) securely before upgrades. The application stays loopback-only; remote access requires a separately reviewed tunnel or reverse proxy.
+130
View File
@@ -0,0 +1,130 @@
# Traffic HAR 1.2 Import and Export
ARTEX supports importing and exporting captured HTTP traffic using the **HAR 1.2** JSON format so operators can bring captures from browser DevTools or external proxies into the Traffic store and export filtered subsets out of ARTEX.
Both routes require the standard administrator JWT (`Authorization: Bearer <token>`) like the rest of `/api/traffic`, and return `503` when the traffic store is unavailable.
---
## Import: `POST /api/traffic/import/har`
### Request format
Send a raw HAR 1.2 JSON document directly as the HTTP request body (`Content-Type: application/json`), capped at the standard API JSON body limit (**32 MiB**; oversized bodies return `413 Request Entity Too Large`).
```bash
curl -X POST \
-H "Authorization: Bearer $TOKEN" \
-H "Content-Type: application/json" \
--data-binary @capture.har \
http://127.0.0.1:8787/api/traffic/import/har
```
The Traffic page also provides an **Import HAR** file picker (with the same 32 MiB limit) that reads a `.har` / `.json` file, posts it to this endpoint, displays loading/error state and a count summary, and refreshes the table.
### Validation & error handling
The request is rejected with `400 Bad Request` naming the reason when:
- the body is empty or malformed JSON (including trailing content after the root object),
- the `log` object is missing,
- `log.version` is not `"1.2"`,
- `log.entries` is missing or `null`,
- `log.entries` exceeds the per-import entry cap (`2000` entries).
Individual malformed entries (invalid `startedDateTime`, non-HTTP(S) URL, invalid HTTP method, header/cookie names or values containing CR/LF/NUL, out-of-range status, invalid base64 or unsupported body encoding) do **not** fail the whole import: they are counted in `failed` and reported in `errors` (up to 50 error details) while valid entries continue importing. HTTP/2 pseudo-headers starting with `:` (such as `:authority` or `:method` from browser DevTools) are skipped cleanly.
### Response
```json
{
"imported": 12,
"skipped": 3,
"failed": 1,
"truncated": false,
"errors": [
{ "index": 4, "url": "notaurl", "reason": "invalid request url \"notaurl\"" }
]
}
```
### Deterministic IDs, round-trip fidelity, & idempotence
- **ARTEX exports (`_artexExchangeId`)**: Every entry exported by ARTEX carries the HAR 1.2 custom field `_artexExchangeId` set to the exchange's stored ID. When a valid `_artexExchangeId` (`[A-Za-z0-9._:-]`, 1–128 chars) is present on import, ARTEX preserves it. This guarantees that two distinct ARTEX exchanges recorded within the same second with identical method, URL, status, and bodies do **not** collapse into one row when imported into a fresh store, and that re-importing an export into the source store stays idempotent (`skipped`).
- **External HARs (no `_artexExchangeId`)**: Each entry receives a deterministic content-based ID:
```text
har-<hex(sha256(ts_unix_seconds + NUL + METHOD + NUL + canonical_url + NUL + status + NUL + sha256(req_body) + NUL + sha256(resp_body)))>
```
`ts_unix_seconds` is `startedDateTime` parsed as RFC 3339 and truncated to UTC Unix seconds (matching `exchanges.ts`). Re-importing the same external HAR skips existing IDs (`INSERT OR IGNORE`, hot rows always win). Identical entries in the same external HAR sharing `(ts_unix_seconds, method, canonical_url, status, req_body, resp_body)` are deduplicated (the first is imported, duplicates are counted in `skipped`).
### Storage, bodies, cookies, `postData.params`, and userinfo
- **URL userinfo**: Any `user:pass@` userinfo in `request.url` is stripped before storing, indexing, listing, exporting, or reporting errors (matching ARTEX's asset store).
- `host` is derived from `URL.Hostname()` and `url_template` from `db.TemplatePath(URL.EscapedPath())`; `path` stays empty.
- **Bodies & base64**: `response.content.text` and `request.postData.text` are stored as raw UTF-8 bytes when `encoding` is empty, and decoded via standard base64 when `encoding` is `"base64"`. When `request.postData.text` is empty and `request.postData.params` is populated, ARTEX reconstructs the form body as `application/x-www-form-urlencoded` (preserving parameter order).
- **Cookies arrays**: When `request.headers` does not already include a `Cookie` header and `request.cookies` is non-empty, ARTEX synthesizes a `Cookie` header from `request.cookies`. Likewise, when `response.headers` does not already include a `Set-Cookie` header and `response.cookies` is non-empty, ARTEX synthesizes `Set-Cookie` headers from `response.cookies`. When the headers are already present, they are kept as-is without duplication.
- `req_len` and `resp_len` record the actual decoded byte lengths.
- `content_type` comes from `response.content.mimeType` (falling back to the response `Content-Type` header).
- Bodies larger than the 256 KiB inline threshold spill to the content-addressed blob store (`_blobs/sha256/...`) with `blob_refs` rows, and searchable text bodies are added to the `ex_fts` full-text index — identical to proxy capture and `ImportArchive`.
### Capture rewrite rules are NOT applied
HAR import writes directly to the traffic store without passing through the MITM capture proxy. **Capture rewrite rules are never applied to imported exchanges**, no `exchange_rewrites` rows are created, and imported rows always have `rewritten: false`.
### Import limits
| Limit | Default | Behavior when exceeded |
|---|---|---|
| Raw HTTP request body | 32 MiB | `413 Request Entity Too Large` |
| Entries per HAR | 2,000 | `400 Bad Request` |
| Total decoded request + response body bytes | 32 MiB | Stops importing remaining entries, sets `"truncated": true`, counts remaining entries in `skipped`, and appends an explanatory item in `errors`. |
---
## Export: `GET /api/traffic/export/har`
### Request & filters
Returns an `application/json` HAR 1.2 document (`Content-Disposition: attachment; filename="artex-traffic.har"`) ordered deterministically by `(ts ASC, id ASC)`.
It reuses the same filter parser and SQL predicate builder as `GET /api/traffic`, supporting:
- `host`, `method`, `path`, `status`, `q`, `body`, `rewritten` (`all` | `rewritten` | `original` | `true` | `false`), `resp_min`, `resp_max`.
### HAR 1.2 compliance & unknown placeholders
- `log.version` is `"1.2"` and `log.creator` is `{"name": "ARTEX", "version": "<build-version>"}`.
- `startedDateTime` is formatted from `exchanges.ts` as UTC RFC 3339, and `_artexExchangeId` carries the exchange's stored ID.
- Request and response headers are parsed from the stored raw heads (`req_head` / `resp_head`). ARTEX does not invent `serverIPAddress` or headers that are not present in the store.
- Required HAR 1.2 fields whose values are not recorded by ARTEX are emitted with explicit **unknown placeholders** rather than fabricated measurements:
- `entry.time`: `0`
- `entry.cache`: `{}`
- `entry.timings`: `{"send": -1, "wait": -1, "receive": -1}` (unknown/unavailable per HAR 1.2)
- `request.cookies`, `request.queryString`, `response.cookies`: `[]`
- `request.headersSize`, `response.headersSize`: `-1` (unknown per HAR 1.2)
- `response.redirectURL`: `""`
- Bodies are read in full from inline columns or spilled blobs: valid NUL-free UTF-8 bodies are emitted as plain `text`, and binary / non-UTF-8 bodies are emitted with `encoding: "base64"`.
### Export limits, import-compatibility invariant, & truncation reporting
| Limit | Default | `truncatedReason` |
|---|---|---|
| Exported entries | 1,000 | `max_entries` |
| Total raw request + response headers and bodies | 20 MiB | `max_bytes` |
| Marshaled JSON size backstop | 31 MiB | `max_bytes` |
**Cap invariant**: Every HAR produced by `GET /api/traffic/export/har` fits inside `POST /api/traffic/import/har`'s **32 MiB** HTTP body limit (and the Traffic page's 32 MiB file limit). Raw headers and bodies are capped at **20 MiB** so worst-case base64 expansion (`4/3x ≈ 26.7 MiB`) plus JSON structure stays comfortably under 32 MiB, backed by an explicit **31 MiB** marshaled-JSON cap so even heavily JSON-escaped payloads cannot trigger a `413` on re-import.
Truncation is never silent. Every export reports it in both HTTP response headers and HAR 1.2 custom (`_`-prefixed) `log` fields:
- Headers:
- `X-ARTEX-HAR-Truncated: true|false`
- `X-ARTEX-HAR-Truncated-Reason: max_entries|max_bytes` (when truncated)
- `X-ARTEX-HAR-Exported: <count>`
- `X-ARTEX-HAR-Matched: <total matching rows>`
- JSON (`log` object, omitted when not truncated):
- `_truncated: true`
- `_truncatedReason: "max_entries" | "max_bytes"`
- `_totalMatched: <total matching rows>`
+54 -5
View File
@@ -4,9 +4,11 @@ A rewrite rule matches a request or a response that goes through the capture pro
how to change it: the method, the path and query, headers, the body, the status code.
**Status: the capture proxy applies enabled rules to HTTP traffic (request and response phases),
and records what it did.** Not yet in this change: archive format, a per-replay rule toggle, the web UI,
and the proof flag. See [Application in the capture proxy](#application-in-the-capture-proxy) for
what is covered and what is not.
records what it did, tags bound evidence snapshots, supports per-replay rule opt-in, carries
rewrite provenance in traffic archive v2, exposes rewrite counters and a rewritten filter/flag on
`GET /api/traffic`, and shows the filter and badge on the Traffic page.** Not yet in this change:
the rule-management web UI. See [Application in the capture proxy](#application-in-the-capture-proxy)
for what is covered and what is not.
Rules are a convenience for the operator, not a security control: a rule that fails leaves the
message unchanged (fail open).
@@ -207,5 +209,52 @@ the slot may be converted into a tombstone if a new reservation needs the room.
### Replay
Replay goes through the recording proxy like any other traffic, so the enabled rules apply to a
replay and the resulting exchange is traced like a captured one. There is no per-replay toggle yet.
`POST /api/traffic/replay` sends the edited request through the recording proxy so the exchange is
captured in Traffic, but **bypasses capture rewrite rules by default** (`apply_rules: false`). To
run enabled capture rules against a replay and trace them like captured traffic, opt in with
`"apply_rules": true` (or check **Apply capture rewrite rules** in the Replay panel).
When `apply_rules` is false, `traffic.Replay` registers a single-use random 32-byte hex nonce in an
in-memory registry, attaches the internal header `X-ARTEX-Replay-Nonce: <nonce>`, and removes the
nonce when the call returns. The capture proxy strips `X-ARTEX-Replay-Nonce` unconditionally in the
request hook so the target never sees it; only a value matching a registered nonce bypasses both
request and response rule phases, while any forged header is stripped and ignored.
### Stats
`GET /api/traffic` includes a `rewrite` object exposing the proxy's `RewriteStats` counters since
start: `large_responses`, `capacity_refusals`, `tombstone_drops`, `conversions`, and `sweeps`.
### Traffic list
Each exchange returned by `GET /api/traffic` carries `"rewritten": true|false`, computed with an
`EXISTS` check against `exchange_rewrites` for rows with `outcome = 'applied'` (skipped or failed
rules do not mark an exchange as rewritten).
`GET /api/traffic` accepts an optional `rewritten` query parameter:
- `all` (default when omitted or unrecognized): returns all exchanges.
- `rewritten` (alias `true`): returns only exchanges with at least one applied rewrite row.
- `original` (alias `false`): returns only exchanges with no applied rewrite row.
The filter combines with `host`, `method`, `path`, `status`, `q`, `body`, `resp_min`, and
`resp_max` using `AND`, and `total` and pagination reflect the filtered set. The Traffic page shows
the filter next to the existing filters and renders a `Rewritten` badge on rewritten rows.
### Task archives (v2)
Task archives export captured traffic into `traffic/traffic.json` using **format version 2**:
- Each `ArchiveExchange` carries optional `rewritten` (`bool`) and `rewrite_rule_ids` (`[]int64`),
matching `RewriteInfo` (applied rows only, sorted ascending, distinct). Exchanges with no applied
rewrite omit both fields.
- Output stays deterministic: exchanges are ordered by `(ts, id)`, `rewrite_rule_ids` are sorted
and distinct, and `blobs` are sorted so re-exporting the same data produces a byte-identical
`traffic.json`.
- `ImportArchive` accepts versions 1 and 2 and refuses any other version. Version 1 entries import
as unrewritten with no rule IDs. Version 2 entries marked `rewritten: true` reconstruct minimal
applied rows in `exchange_rewrites` (`phase = 'archived'`, `outcome = 'applied'`,
`reason = 'archived'`, `rule_snapshot = NULL`) for each distinct rule ID (or a single `NULL`
`rule_id` row when `rewrite_rule_ids` is empty), preserving `RewriteInfo`, `traffic_get`'s
`=== REWRITTEN ===` header, evidence-binding provenance, and the traffic list filter without
fabricating a rule snapshot or request/response phase. Existing hot rows always win on import.
+169
View File
@@ -0,0 +1,169 @@
package llmrec
import (
"encoding/json"
"log"
"os"
"strings"
"github.com/Autumn-27/artex/db"
"github.com/Autumn-27/norma/llm"
)
// ContextTelemetryEnv is the opt-in environment flag enabling per-call context
// size telemetry into llm_context. Off by default: when unset/false, no sizing
// work runs and no rows are written.
const ContextTelemetryEnv = "ARTEX_CONTEXT_TELEMETRY"
// ContextTelemetryEnabled reports whether ARTEX_CONTEXT_TELEMETRY is enabled
// (1, true, on, yes; case-insensitive). Default is false.
func ContextTelemetryEnabled() bool {
switch strings.ToLower(strings.TrimSpace(os.Getenv(ContextTelemetryEnv))) {
case "1", "true", "on", "yes":
return true
}
return false
}
// ContextMetrics holds size-only measurements of an assembled CompletionRequest
// as it leaves for the provider. Never retains any prompt, tool, or message
// content.
type ContextMetrics struct {
ToolCount int
MessageCount int
SystemStaticBytes int64
SystemDynamicBytes int64
ToolsBytes int64
MessagesUserBytes int64
MessagesAssistantBytes int64
MessagesOtherBytes int64
ToolResultBytes int64
LastMessageBytes int64
}
// MeasureContext computes deterministic byte sizes and counts for req without
// retaining any content:
//
// - System: when DynamicBoundary > 0, req.System is split at
// min(DynamicBoundary, len(req.System)) into SystemStaticBytes ([:b]) and
// SystemDynamicBytes ([b:]). Within each slice, each segment is trimmed with
// strings.TrimSpace, empty segments are dropped, and remaining segments are
// joined with "\n\n", matching norma's llm.joinSystem. When
// DynamicBoundary <= 0, the whole req.System is measured as a single static
// block (SystemStaticBytes) and SystemDynamicBytes is 0.
// - Tools: ToolCount = len(req.Tools); ToolsBytes = len(json.Marshal(req.Tools))
// (0 when no tools). Go's encoding/json sorts map keys deterministically.
// - Messages: MessageCount = len(req.Messages). Each message is measured by
// len(json.Marshal(msg)) and summed into MessagesUserBytes (role "user"),
// MessagesAssistantBytes (role "assistant"), or MessagesOtherBytes (any
// other role). ToolResultBytes records the serialized size of tool_result
// content blocks inside messages as an informational subset (not added to
// the partition denominator). LastMessageBytes is the serialized byte size
// of the final message (0 when no messages).
func MeasureContext(req llm.CompletionRequest) ContextMetrics {
var m ContextMetrics
// 1. System prompt bytes (split at DynamicBoundary when > 0, else single static block).
if len(req.System) > 0 {
if req.DynamicBoundary > 0 {
b := req.DynamicBoundary
if b > len(req.System) {
b = len(req.System)
}
m.SystemStaticBytes = joinedSegmentBytes(req.System[:b])
m.SystemDynamicBytes = joinedSegmentBytes(req.System[b:])
} else {
m.SystemStaticBytes = joinedSegmentBytes(req.System)
}
}
// 2. Serialized tool schemas and count.
m.ToolCount = len(req.Tools)
if len(req.Tools) > 0 {
if b, err := json.Marshal(req.Tools); err == nil {
m.ToolsBytes = int64(len(b))
}
}
// 3. Messages by role, tool_result subset, and last message size.
m.MessageCount = len(req.Messages)
for i, msg := range req.Messages {
var msgBytes int64
if b, err := json.Marshal(msg); err == nil {
msgBytes = int64(len(b))
}
switch msg.Role {
case llm.RoleUser:
m.MessagesUserBytes += msgBytes
case llm.RoleAssistant:
m.MessagesAssistantBytes += msgBytes
default:
m.MessagesOtherBytes += msgBytes
}
if i == len(req.Messages)-1 {
m.LastMessageBytes = msgBytes
}
for _, blk := range msg.Content {
if blk.Type == llm.BlockToolResult {
if b, err := json.Marshal(blk); err == nil {
m.ToolResultBytes += int64(len(b))
}
}
}
}
return m
}
// joinedSegmentBytes returns the byte length of segs after trimming whitespace
// from each segment, dropping empty segments, and joining the remaining ones
// with "\n\n" — matching norma's llm.joinSystem without allocating a joined
// string.
func joinedSegmentBytes(segs []string) int64 {
var total int64
nonEmpty := 0
for _, s := range segs {
s = strings.TrimSpace(s)
if s == "" {
continue
}
if nonEmpty > 0 {
total += 2 // "\n\n"
}
total += int64(len(s))
nonEmpty++
}
return total
}
// recordContext appends one row to llm_context when r.pg is configured.
// Best-effort: logs and continues on error so telemetry never disrupts calls.
func (r *Recorder) recordContext(taskID, worker string, m ContextMetrics, usage llm.Usage, status string) {
if r.pg == nil {
return
}
err := r.pg.InsertLLMContext(&db.LLMContext{
TaskID: taskID,
Worker: worker,
Model: r.model,
Profile: r.prof,
ToolCount: m.ToolCount,
MessageCount: m.MessageCount,
SystemStaticBytes: m.SystemStaticBytes,
SystemDynamicBytes: m.SystemDynamicBytes,
ToolsBytes: m.ToolsBytes,
MessagesUserBytes: m.MessagesUserBytes,
MessagesAssistantBytes: m.MessagesAssistantBytes,
MessagesOtherBytes: m.MessagesOtherBytes,
ToolResultBytes: m.ToolResultBytes,
LastMessageBytes: m.LastMessageBytes,
InputTokens: usage.InputTokens,
OutputTokens: usage.OutputTokens,
CacheRead: usage.CacheReadTokens,
CacheWrite: usage.CacheWriteTokens,
Status: status,
})
if err != nil {
log.Printf("[llmcontext] insert: %v", err)
}
}
+371
View File
@@ -0,0 +1,371 @@
package llmrec
import (
"context"
"encoding/json"
"errors"
"fmt"
"iter"
"os"
"testing"
"time"
"github.com/Autumn-27/artex/db"
"github.com/Autumn-27/norma/llm"
"github.com/Autumn-27/norma/transcript"
)
type syntheticTelemetryProvider struct {
usage llm.Usage
streamErr error
completeEr error
}
func (p syntheticTelemetryProvider) Stream(context.Context, llm.CompletionRequest) iter.Seq2[llm.StreamEvent, error] {
return func(yield func(llm.StreamEvent, error) bool) {
if !yield(llm.StreamEvent{Type: llm.SEMessageStart, Usage: llm.Usage{
InputTokens: p.usage.InputTokens,
CacheReadTokens: p.usage.CacheReadTokens,
CacheWriteTokens: p.usage.CacheWriteTokens,
}}, nil) {
return
}
if p.streamErr != nil {
yield(llm.StreamEvent{}, p.streamErr)
return
}
if !yield(llm.StreamEvent{Type: llm.SETextDelta, Text: "ok"}, nil) {
return
}
yield(llm.StreamEvent{
Type: llm.SEMessageDelta,
StopReason: "end_turn",
Usage: llm.Usage{OutputTokens: p.usage.OutputTokens},
}, nil)
}
}
func (p syntheticTelemetryProvider) Complete(context.Context, llm.CompletionRequest) (llm.Message, string, llm.Usage, error) {
if p.completeEr != nil {
return llm.Message{}, "", p.usage, p.completeEr
}
return llm.Message{Role: llm.RoleAssistant, Content: []llm.ContentBlock{llm.TextBlock("done")}}, "end_turn", p.usage, nil
}
func TestMeasureContextExactSizesAndBoundaryModes(t *testing.T) {
tools := []llm.ToolSchema{
{
Name: "Bash",
Description: "Run shell command",
InputSchema: map[string]any{
"type": "object",
"properties": map[string]any{
"command": map[string]any{"type": "string"},
},
},
},
{
Name: "record_fact",
Description: "Store a verified fact",
InputSchema: map[string]any{"type": "object"},
},
}
wantToolsBytes, err := json.Marshal(tools)
if err != nil {
t.Fatal(err)
}
msgUser1 := llm.UserText("first user prompt")
msgAsst := llm.Message{
Role: llm.RoleAssistant,
Content: []llm.ContentBlock{
{Type: llm.BlockToolUse, ID: "call_1", Name: "Bash", Input: json.RawMessage(`{"command":"id"}`)},
},
}
toolResBlock := llm.ToolResultText("call_1", "uid=0(root) gid=0(root)", false)
msgUser2 := llm.Message{
Role: llm.RoleUser,
Content: []llm.ContentBlock{toolResBlock},
}
msgOther := llm.Message{
Role: llm.Role("tool"),
Content: []llm.ContentBlock{llm.TextBlock("other role text")},
}
mustLen := func(v any) int64 {
b, err := json.Marshal(v)
if err != nil {
t.Fatal(err)
}
return int64(len(b))
}
// Case 1: DynamicBoundary == 0 -> all system segments as single static block (joined with "\n\n").
reqNoSplit := llm.CompletionRequest{
System: []string{"STATIC_ONE", "STATIC_TWO"},
DynamicBoundary: 0,
Tools: tools,
Messages: []llm.Message{msgUser1, msgAsst, msgUser2, msgOther},
}
got1 := MeasureContext(reqNoSplit)
if got1.SystemStaticBytes != int64(len("STATIC_ONE\n\nSTATIC_TWO")) || got1.SystemDynamicBytes != 0 {
t.Fatalf("DynamicBoundary=0 system = (%d, %d), want (%d, 0)", got1.SystemStaticBytes, got1.SystemDynamicBytes, len("STATIC_ONE\n\nSTATIC_TWO"))
}
if got1.ToolCount != 2 || got1.ToolsBytes != int64(len(wantToolsBytes)) {
t.Fatalf("tools = (count=%d, bytes=%d), want (2, %d)", got1.ToolCount, got1.ToolsBytes, len(wantToolsBytes))
}
wantUser := mustLen(msgUser1) + mustLen(msgUser2)
wantAsst := mustLen(msgAsst)
wantOther := mustLen(msgOther)
wantToolRes := mustLen(toolResBlock)
if got1.MessageCount != 4 ||
got1.MessagesUserBytes != wantUser ||
got1.MessagesAssistantBytes != wantAsst ||
got1.MessagesOtherBytes != wantOther ||
got1.ToolResultBytes != wantToolRes ||
got1.LastMessageBytes != wantOther {
t.Fatalf("messages = %+v, want count=4 user=%d asst=%d other=%d toolRes=%d last=%d",
got1, wantUser, wantAsst, wantOther, wantToolRes, wantOther)
}
// Case 2: 0 < DynamicBoundary < len(System) -> split at boundary.
reqSplit := llm.CompletionRequest{
System: []string{"AAA", "BBBB", "CCCCC", "DD"},
DynamicBoundary: 2,
}
got2 := MeasureContext(reqSplit)
if got2.SystemStaticBytes != int64(len("AAA\n\nBBBB")) || got2.SystemDynamicBytes != int64(len("CCCCC\n\nDD")) {
t.Fatalf("DynamicBoundary=2 system = (%d, %d), want (%d, %d)",
got2.SystemStaticBytes, got2.SystemDynamicBytes, len("AAA\n\nBBBB"), len("CCCCC\n\nDD"))
}
// Case 3: DynamicBoundary >= len(System) -> whole prompt static, dynamic == 0.
reqAllStatic := llm.CompletionRequest{
System: []string{"AAA", "BBBB"},
DynamicBoundary: 5,
}
got3 := MeasureContext(reqAllStatic)
if got3.SystemStaticBytes != int64(len("AAA\n\nBBBB")) || got3.SystemDynamicBytes != 0 {
t.Fatalf("DynamicBoundary=5 system = (%d, %d), want (%d, 0)",
got3.SystemStaticBytes, got3.SystemDynamicBytes, len("AAA\n\nBBBB"))
}
// Case 4: whitespace-padded and empty segments match norma's llm.joinSystem
// (strings.TrimSpace per segment, drop empty, join remaining with "\n\n").
for _, tc := range []struct {
name string
segs []string
boundary int
wantStat int64
wantDyn int64
}{
{
name: "padded and empty unsplit",
segs: []string{" STATIC ", "", "\n", "TAIL\n\n"},
boundary: 0,
wantStat: int64(len("STATIC\n\nTAIL")),
wantDyn: 0,
},
{
name: "padded split at boundary 2",
segs: []string{" A ", " B ", " C "},
boundary: 2,
wantStat: int64(len("A\n\nB")),
wantDyn: int64(len("C")),
},
{
name: "only empty and whitespace",
segs: []string{"", " "},
boundary: 1,
wantStat: 0,
wantDyn: 0,
},
{
name: "empty segments around boundary",
segs: []string{" ", "HEAD ", "", "\t\n", " TAIL"},
boundary: 2,
wantStat: int64(len("HEAD")),
wantDyn: int64(len("TAIL")),
},
} {
t.Run(tc.name, func(t *testing.T) {
got := MeasureContext(llm.CompletionRequest{System: tc.segs, DynamicBoundary: tc.boundary})
if got.SystemStaticBytes != tc.wantStat || got.SystemDynamicBytes != tc.wantDyn {
t.Fatalf("MeasureContext(%q, b=%d) = (%d, %d), want (%d, %d)",
tc.segs, tc.boundary, got.SystemStaticBytes, got.SystemDynamicBytes, tc.wantStat, tc.wantDyn)
}
})
}
}
func TestContextTelemetryFlagOffAndOnPostgres(t *testing.T) {
dsn := os.Getenv("ARTEX_PG_DSN")
if dsn == "" {
t.Skip("ARTEX_PG_DSN not set")
}
pg, err := db.Open(dsn)
if err != nil {
db.SkipPGUnavailable(t, err)
}
defer pg.Close()
stamp := time.Now().UnixNano()
taskOff := fmt.Sprintf("ctx-off-%d", stamp)
taskOn := fmt.Sprintf("ctx-on-%d", stamp)
t.Cleanup(func() {
_, _ = pg.Exec(`DELETE FROM llm_context WHERE task_id IN ($1,$2)`, taskOff, taskOn)
_, _ = pg.Exec(`DELETE FROM llm_usage WHERE task_id IN ($1,$2)`, taskOff, taskOn)
_, _ = pg.Exec(`DELETE FROM llm_records WHERE task_id IN ($1,$2)`, taskOff, taskOn)
})
sentinelPrompt := "SECRET_PROMPT_SENTINEL_NEVER_STORE"
sentinelMsg := "SECRET_USER_MSG_SENTINEL_NEVER_STORE"
req := llm.CompletionRequest{
System: []string{sentinelPrompt, "DYNAMIC_TAIL"},
DynamicBoundary: 1,
Tools: []llm.ToolSchema{
{Name: "Bash", Description: "SECRET_TOOL_DESC", InputSchema: map[string]any{"type": "object"}},
},
Messages: []llm.Message{
llm.UserText(sentinelMsg),
},
}
wantMetrics := MeasureContext(req)
wantUsage := llm.Usage{InputTokens: 1200, OutputTokens: 150, CacheReadTokens: 800, CacheWriteTokens: 100}
// 1. Flag OFF (default): zero rows in llm_context on both Stream and Complete,
// identical behavior regardless of recordBodies.
t.Setenv(ContextTelemetryEnv, "")
if ContextTelemetryEnabled() {
t.Fatal("expected telemetry disabled when env unset")
}
recOff := Wrap(syntheticTelemetryProvider{usage: wantUsage}, pg, "claude-test", "prof-a", "", "", func() bool { return false })
ctxOff := transcript.WithSessionID(WithTaskID(context.Background(), taskOff), "exp1-worker-i1")
var streamText string
for ev, err := range recOff.Stream(ctxOff, req) {
if err != nil {
t.Fatal(err)
}
if ev.Type == llm.SETextDelta {
streamText += ev.Text
}
}
if streamText != "ok" {
t.Fatalf("flag-off stream text = %q, want ok", streamText)
}
msgOff, stopOff, usageOff, err := recOff.Complete(ctxOff, req)
if err != nil || msgOff.Text() != "done" || stopOff != "end_turn" || usageOff != wantUsage {
t.Fatalf("flag-off Complete = (%q, %q, %+v, %v)", msgOff.Text(), stopOff, usageOff, err)
}
var offCount int
if err := pg.QueryRow(`SELECT COUNT(*) FROM llm_context WHERE task_id=$1`, taskOff).Scan(&offCount); err != nil {
t.Fatal(err)
}
if offCount != 0 {
t.Fatalf("flag-off wrote %d rows to llm_context, want 0", offCount)
}
// 2. Flag ON with recordBodies=FALSE: rows ARE written on Stream (ok + error)
// and Complete, with exact byte sizes and zero llm_records rows.
t.Setenv(ContextTelemetryEnv, "1")
if !ContextTelemetryEnabled() {
t.Fatal("expected telemetry enabled when ARTEX_CONTEXT_TELEMETRY=1")
}
recOn := Wrap(syntheticTelemetryProvider{usage: wantUsage}, pg, "claude-test", "prof-a", "", "", func() bool { return false })
ctxOn := transcript.WithSessionID(WithTaskID(context.Background(), taskOn), "exp2-planner")
for _, err := range recOn.Stream(ctxOn, req) {
if err != nil {
t.Fatal(err)
}
}
if _, _, _, err := recOn.Complete(ctxOn, req); err != nil {
t.Fatal(err)
}
// Error path also records status='error'.
recErr := Wrap(syntheticTelemetryProvider{usage: llm.Usage{InputTokens: 400}, streamErr: errors.New("upstream 500")}, pg, "claude-test", "prof-a", "", "", func() bool { return false })
for _, err := range recErr.Stream(ctxOn, req) {
if err != nil {
break
}
}
var recBodiesCount int
if err := pg.QueryRow(`SELECT COUNT(*) FROM llm_records WHERE task_id=$1`, taskOn).Scan(&recBodiesCount); err != nil {
t.Fatal(err)
}
if recBodiesCount != 0 {
t.Fatalf("llm_records count = %d with recordBodies=false, want 0", recBodiesCount)
}
rows, err := pg.Query(`
SELECT worker, model, profile, tool_count, message_count,
system_static_bytes, system_dynamic_bytes, tools_bytes,
messages_user_bytes, messages_assistant_bytes, messages_other_bytes,
tool_result_bytes, last_message_bytes,
input_tokens, output_tokens, cache_read, cache_write, status
FROM llm_context WHERE task_id=$1 ORDER BY id`, taskOn)
if err != nil {
t.Fatal(err)
}
defer rows.Close()
var got []db.LLMContext
for rows.Next() {
var c db.LLMContext
if err := rows.Scan(
&c.Worker, &c.Model, &c.Profile, &c.ToolCount, &c.MessageCount,
&c.SystemStaticBytes, &c.SystemDynamicBytes, &c.ToolsBytes,
&c.MessagesUserBytes, &c.MessagesAssistantBytes, &c.MessagesOtherBytes,
&c.ToolResultBytes, &c.LastMessageBytes,
&c.InputTokens, &c.OutputTokens, &c.CacheRead, &c.CacheWrite, &c.Status,
); err != nil {
t.Fatal(err)
}
got = append(got, c)
}
if len(got) != 3 {
t.Fatalf("llm_context rows = %d, want 3 (stream ok, complete ok, stream error)", len(got))
}
for i, c := range got {
if c.Worker != "planner" || c.Model != "claude-test" || c.Profile != "prof-a" ||
c.ToolCount != wantMetrics.ToolCount || c.MessageCount != wantMetrics.MessageCount ||
c.SystemStaticBytes != wantMetrics.SystemStaticBytes ||
c.SystemDynamicBytes != wantMetrics.SystemDynamicBytes ||
c.ToolsBytes != wantMetrics.ToolsBytes ||
c.MessagesUserBytes != wantMetrics.MessagesUserBytes ||
c.MessagesAssistantBytes != wantMetrics.MessagesAssistantBytes ||
c.MessagesOtherBytes != wantMetrics.MessagesOtherBytes ||
c.ToolResultBytes != wantMetrics.ToolResultBytes ||
c.LastMessageBytes != wantMetrics.LastMessageBytes {
t.Fatalf("row %d metrics = %+v, want %+v", i, c, wantMetrics)
}
}
if got[0].Status != "ok" || got[0].InputTokens != 1200 || got[0].CacheRead != 800 || got[0].CacheWrite != 100 ||
got[1].Status != "ok" || got[1].OutputTokens != 150 ||
got[2].Status != "error" || got[2].InputTokens != 400 {
t.Fatalf("unexpected usage/status rows: %+v", got)
}
// Verify schema contains zero content columns and stored rows hold none of the sentinels.
var textCols int
if err := pg.QueryRow(`SELECT COUNT(*) FROM information_schema.columns WHERE table_name='llm_context' AND column_name ~ '(body|content|prompt|text|raw|message$)'`).Scan(&textCols); err != nil || textCols != 0 {
t.Fatalf("unexpected content-like columns in llm_context: count=%d err=%v", textCols, err)
}
var serializedRows string
if err := pg.QueryRow(`SELECT COALESCE(string_agg(row_to_json(t)::text, ''), '') FROM llm_context t WHERE task_id=$1`, taskOn).Scan(&serializedRows); err != nil {
t.Fatal(err)
}
for _, secret := range []string{sentinelPrompt, sentinelMsg, "SECRET_TOOL_DESC", "DYNAMIC_TAIL"} {
if containsSubstr(serializedRows, secret) {
t.Fatalf("llm_context leaked request content %q", secret)
}
}
}
func containsSubstr(s, sub string) bool {
for i := 0; i+len(sub) <= len(s); i++ {
if s[i:i+len(sub)] == sub {
return true
}
}
return false
}
+19
View File
@@ -151,6 +151,7 @@ func (r *Recorder) Stream(ctx context.Context, req llm.CompletionRequest) iter.S
// stats and must be complete even for interrupted/failed runs, so it is NOT
// gated. Only body serialization + response accumulation are skipped when off.
recordBodies := r.enabled == nil || r.enabled()
collectContext := ContextTelemetryEnabled()
start := time.Now()
session := transcript.SessionIDFrom(ctx)
parsedID, worker := parseSession(session)
@@ -165,6 +166,11 @@ func (r *Recorder) Stream(ctx context.Context, req llm.CompletionRequest) iter.S
taskID = parsedID
}
var ctxMetrics ContextMetrics
if collectContext {
ctxMetrics = MeasureContext(req)
}
// Serialize the request only when storing bodies (this is the expensive part).
// Alongside it, attach a Capture so the HTTP transport can hand back the
// untouched wire bodies — the normalized view below cannot reconstruct them
@@ -198,6 +204,10 @@ func (r *Recorder) Stream(ctx context.Context, req llm.CompletionRequest) iter.S
}
// Lightweight metering row — always written.
r.recordUsage(ctx, taskID, expID, worker, usage, int(time.Since(start).Milliseconds()), status)
// Opt-in context size telemetry — independent of recordBodies.
if collectContext {
r.recordContext(taskID, worker, ctxMetrics, usage, status)
}
// Heavy trace row — only when body recording is on.
if recordBodies {
r.record(req, session, taskID, worker, reqBody, capt, start, textBuf.String(), thinkingBuf.String(), usage, stopReason, err)
@@ -252,6 +262,7 @@ func (r *Recorder) Stream(ctx context.Context, req llm.CompletionRequest) iter.S
// usage metering, normalized body recording and raw wire capture as Stream.
func (r *Recorder) Complete(ctx context.Context, req llm.CompletionRequest) (llm.Message, string, llm.Usage, error) {
recordBodies := r.enabled == nil || r.enabled()
collectContext := ContextTelemetryEnabled()
start := time.Now()
session := transcript.SessionIDFrom(ctx)
parsedID, worker := parseSession(session)
@@ -264,6 +275,11 @@ func (r *Recorder) Complete(ctx context.Context, req llm.CompletionRequest) (llm
taskID = parsedID
}
var ctxMetrics ContextMetrics
if collectContext {
ctxMetrics = MeasureContext(req)
}
reqBody := ""
var capt *Capture
if recordBodies {
@@ -279,6 +295,9 @@ func (r *Recorder) Complete(ctx context.Context, req llm.CompletionRequest) (llm
reportStop(ctx, stopReason)
}
r.recordUsage(ctx, taskID, expID, worker, usage, int(time.Since(start).Milliseconds()), status)
if collectContext {
r.recordContext(taskID, worker, ctxMetrics, usage, status)
}
if recordBodies {
r.record(req, session, taskID, worker, reqBody, capt, start, msg.Text(), thinkingText(msg), usage, stopReason, err)
}
+2
View File
@@ -168,6 +168,7 @@ type TrafficExchangeDTO struct {
Status int `json:"status"`
ContentType string `json:"content_type"`
RespLen int `json:"resp_len"`
Rewritten bool `json:"rewritten"`
}
func trafficDTOs(ex []traffic.ExchangeMeta) []TrafficExchangeDTO {
@@ -182,6 +183,7 @@ func trafficDTOs(ex []traffic.ExchangeMeta) []TrafficExchangeDTO {
Status: e.Status,
ContentType: e.ContentType,
RespLen: e.RespLen,
Rewritten: e.Rewritten,
})
}
return out
+40
View File
@@ -0,0 +1,40 @@
package server
import (
"net/http"
"strings"
"github.com/Autumn-27/artex/db"
"github.com/Autumn-27/artex/llmrec"
)
// pgLLMContextSummary returns aggregated context-size telemetry from
// llm_context: per-(worker, model) input percentiles, pro-rata section shares
// over real input_tokens, cache_read/input ratio, and top tasks by input
// tokens. Read-only reporting endpoint (no UI in Phase A).
func (s *Server) pgLLMContextSummary(w http.ResponseWriter, r *http.Request) {
pg := s.m.PG()
if pg == nil {
writeJSON(w, http.StatusOK, map[string]any{
"enabled": llmrec.ContextTelemetryEnabled(),
"by_worker_model": []db.LLMContextGroupStat{},
"top_tasks": []db.LLMContextTaskStat{},
})
return
}
q := r.URL.Query()
summary, err := pg.LLMContextSummary(db.LLMContextQuery{
Days: atoiDefault(q.Get("days"), 30),
TaskID: strings.TrimSpace(q.Get("task")),
TopTasks: atoiDefault(q.Get("top"), 20),
})
if err != nil {
writeErr(w, http.StatusInternalServerError, err.Error())
return
}
writeJSON(w, http.StatusOK, map[string]any{
"enabled": llmrec.ContextTelemetryEnabled(),
"by_worker_model": summary.ByWorkerModel,
"top_tasks": summary.TopTasks,
})
}
+62
View File
@@ -0,0 +1,62 @@
package server
import (
"encoding/json"
"fmt"
"net/http"
"net/http/httptest"
"testing"
"time"
"github.com/Autumn-27/artex/db"
"github.com/Autumn-27/artex/llmrec"
)
func TestLLMContextSummaryEndpointAuthAndAggregation(t *testing.T) {
s, request := ingestServer(t)
// Unauthenticated request -> 401.
rNoAuth := httptest.NewRequest(http.MethodGet, "/api/llm/context/summary", nil)
wNoAuth := httptest.NewRecorder()
s.Handler().ServeHTTP(wNoAuth, rNoAuth)
if wNoAuth.Code != http.StatusUnauthorized {
t.Fatalf("unauthenticated /api/llm/context/summary status=%d, want 401", wNoAuth.Code)
}
taskID := fmt.Sprintf("srv-ctx-%d", time.Now().UnixNano())
t.Cleanup(func() { _, _ = s.m.PG().Exec(`DELETE FROM llm_context WHERE task_id=$1`, taskID) })
if err := s.m.PG().InsertLLMContext(&db.LLMContext{
TaskID: taskID, Worker: "worker", Model: "test-model", Profile: "p1",
ToolCount: 8, MessageCount: 5,
SystemStaticBytes: 250, ToolsBytes: 250,
MessagesUserBytes: 400, MessagesAssistantBytes: 100, ToolResultBytes: 300,
LastMessageBytes: 200,
InputTokens: 1000, OutputTokens: 120, CacheRead: 700, CacheWrite: 100, Status: "ok",
}); err != nil {
t.Fatal(err)
}
t.Setenv(llmrec.ContextTelemetryEnv, "1")
w := request(http.MethodGet, "/api/llm/context/summary?days=7&top=5&task="+taskID, "")
if w.Code != http.StatusOK {
t.Fatalf("GET /api/llm/context/summary status=%d body=%s", w.Code, w.Body.String())
}
var resp struct {
Enabled bool `json:"enabled"`
ByWorkerModel []db.LLMContextGroupStat `json:"by_worker_model"`
TopTasks []db.LLMContextTaskStat `json:"top_tasks"`
}
if err := json.Unmarshal(w.Body.Bytes(), &resp); err != nil {
t.Fatal(err)
}
if !resp.Enabled || len(resp.ByWorkerModel) != 1 || len(resp.TopTasks) != 1 {
t.Fatalf("unexpected summary response: %+v", resp)
}
g := resp.ByWorkerModel[0]
if g.Worker != "worker" || g.Model != "test-model" || g.Calls != 1 || g.InputTokensTotal != 1000 ||
g.CacheReadRatio < 0.699 || g.CacheReadRatio > 0.701 ||
g.Shares.SystemStatic < 0.249 || g.Shares.Tools < 0.249 || g.Shares.MessagesUser < 0.399 {
t.Fatalf("unexpected group stat: %+v", g)
}
}
+14 -14
View File
@@ -898,9 +898,12 @@ func (s *Server) Handler() http.Handler {
mux.HandleFunc("GET /api/traffic/blob", s.getTrafficBlob)
mux.HandleFunc("POST /api/traffic/ingest", s.ingestTraffic)
mux.HandleFunc("POST /api/traffic/replay", s.replayTraffic)
mux.HandleFunc("POST /api/traffic/import/har", s.importTrafficHAR)
mux.HandleFunc("GET /api/traffic/export/har", s.exportTrafficHAR)
mux.HandleFunc("GET /api/commands", s.pgListCommands)
mux.HandleFunc("GET /api/commands/stats", s.pgToolStats) // aggregate call counts per tool
mux.HandleFunc("GET /api/llm/records", s.pgListLLMRecords)
mux.HandleFunc("GET /api/llm/context/summary", s.pgLLMContextSummary)
mux.HandleFunc("DELETE /api/llm/records", s.pgDeleteLLMRecords)
mux.HandleFunc("GET /api/llm/records/tasks", s.pgLLMTasks)
mux.HandleFunc("GET /api/llm/records/by-model", s.pgTokenByModel) // aggregate this task's token usage by model
@@ -3535,24 +3538,13 @@ func taskActivityDetail(t *Task, seq int64) (string, error) {
func (s *Server) getTraffic(w http.ResponseWriter, r *http.Request) {
tr := s.m.Traffic()
if tr == nil {
writeJSON(w, 200, map[string]any{"enabled": false, "exchanges": []any{}})
writeJSON(w, 200, map[string]any{"enabled": false, "exchanges": []any{}, "rewrite": traffic.RewriteStats{}})
return
}
q := r.URL.Query()
page := atoiDefault(q.Get("page"), 0)
size := atoiDefault(q.Get("size"), 100)
ex, matched, _ := tr.Page(traffic.PageQuery{
Host: q.Get("host"),
Method: q.Get("method"),
Query: q.Get("q"),
Body: q.Get("body"),
Path: q.Get("path"),
Status: q.Get("status"),
RespMin: int64(atoiDefault(q.Get("resp_min"), -1)),
RespMax: int64(atoiDefault(q.Get("resp_max"), -1)),
Sort: q.Get("sort"),
Order: q.Get("order"),
}, page, size)
ex, matched, _ := tr.Page(parseTrafficPageQuery(q), page, size)
count, _ := tr.Count() // global total, for the stat card
writeJSON(w, 200, map[string]any{
"enabled": s.m.TrafficEnabled(), // reflect the capture toggle
@@ -3562,6 +3554,7 @@ func (s *Server) getTraffic(w http.ResponseWriter, r *http.Request) {
"page": page,
"size": size,
"exchanges": trafficDTOs(ex),
"rewrite": tr.RewriteStats(),
})
}
@@ -3740,6 +3733,7 @@ func (s *Server) replayTraffic(w http.ResponseWriter, r *http.Request) {
Headers []traffic.ReplayHeader `json:"headers"`
Body string `json:"body"`
BodyBase64 bool `json:"body_base64"`
ApplyRules bool `json:"apply_rules"`
}
if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
var tooLarge *http.MaxBytesError
@@ -3761,7 +3755,13 @@ func (s *Server) replayTraffic(w http.ResponseWriter, r *http.Request) {
}
ctx, cancel := context.WithTimeout(r.Context(), 60*time.Second)
defer cancel()
res, err := tr.Replay(ctx, traffic.ReplayRequest{Method: body.Method, URL: body.URL, Headers: body.Headers, Body: payload})
res, err := tr.Replay(ctx, traffic.ReplayRequest{
Method: body.Method,
URL: body.URL,
Headers: body.Headers,
Body: payload,
ApplyRules: body.ApplyRules,
})
switch {
case errors.Is(err, traffic.ErrReplayTooLarge):
writeErr(w, 413, err.Error())
+96
View File
@@ -0,0 +1,96 @@
package server
import (
"encoding/json"
"errors"
"io"
"net/http"
"net/url"
"strconv"
"github.com/Autumn-27/artex/traffic"
)
func parseTrafficPageQuery(q url.Values) traffic.PageQuery {
return traffic.PageQuery{
Host: q.Get("host"),
Method: q.Get("method"),
Query: q.Get("q"),
Body: q.Get("body"),
Path: q.Get("path"),
Status: q.Get("status"),
Rewritten: q.Get("rewritten"),
RespMin: int64(atoiDefault(q.Get("resp_min"), -1)),
RespMax: int64(atoiDefault(q.Get("resp_max"), -1)),
Sort: q.Get("sort"),
Order: q.Get("order"),
}
}
// importTrafficHAR accepts a raw HAR 1.2 JSON document (capped at decodeMaxBytes,
// 32 MiB) and imports valid entries into the traffic store with deterministic
// IDs. Imported traffic never passes through capture rewrite rules.
func (s *Server) importTrafficHAR(w http.ResponseWriter, r *http.Request) {
tr := s.m.Traffic()
if tr == nil {
writeErr(w, http.StatusServiceUnavailable, "traffic store is not available")
return
}
body := r.Body
if _, already := body.(*limitedBody); !already {
body = http.MaxBytesReader(w, body, decodeMaxBytes)
}
raw, err := io.ReadAll(body)
if err != nil {
var tooLarge *http.MaxBytesError
if errors.As(err, &tooLarge) {
writeErr(w, http.StatusRequestEntityTooLarge, "HAR payload exceeds 32 MiB limit")
return
}
writeErr(w, http.StatusBadRequest, "failed to read HAR body: "+err.Error())
return
}
res, err := tr.ImportHAR(raw, traffic.HARImportOptions{})
if err != nil {
if errors.Is(err, traffic.ErrInvalidHAR) {
writeErr(w, http.StatusBadRequest, err.Error())
return
}
writeErr(w, http.StatusInternalServerError, err.Error())
return
}
writeJSON(w, http.StatusOK, res)
}
// exportTrafficHAR exports recorded traffic matching the query filters as an
// application/json HAR 1.2 document in stable chronological order. Entry and
// total-byte caps are enforced and reported via both X-ARTEX-HAR-* headers and
// HAR 1.2 custom fields (_truncated, _truncatedReason, _totalMatched).
func (s *Server) exportTrafficHAR(w http.ResponseWriter, r *http.Request) {
tr := s.m.Traffic()
if tr == nil {
writeErr(w, http.StatusServiceUnavailable, "traffic store is not available")
return
}
pq := parseTrafficPageQuery(r.URL.Query())
doc, meta, err := tr.ExportHAR(pq, traffic.HARExportOptions{CreatorVersion: BuildVersion})
if err != nil {
writeErr(w, http.StatusInternalServerError, err.Error())
return
}
payload, err := json.Marshal(doc)
if err != nil {
writeErr(w, http.StatusInternalServerError, err.Error())
return
}
w.Header().Set("Content-Type", "application/json; charset=utf-8")
w.Header().Set("Content-Disposition", `attachment; filename="artex-traffic.har"`)
w.Header().Set("X-ARTEX-HAR-Truncated", strconv.FormatBool(meta.Truncated))
w.Header().Set("X-ARTEX-HAR-Exported", strconv.Itoa(meta.Exported))
w.Header().Set("X-ARTEX-HAR-Matched", strconv.Itoa(meta.TotalMatched))
if meta.Truncated {
w.Header().Set("X-ARTEX-HAR-Truncated-Reason", meta.TruncatedReason)
}
w.WriteHeader(http.StatusOK)
_, _ = w.Write(payload)
}
+256
View File
@@ -0,0 +1,256 @@
package server
import (
"bytes"
"encoding/base64"
"encoding/json"
"net/http"
"net/http/httptest"
"strings"
"testing"
"github.com/Autumn-27/artex/traffic"
)
func TestTrafficHAREndpointsRoundTripFilterAndErrors(t *testing.T) {
s, request := ingestServer(t)
binPayload := []byte{0x00, 0x01, 0x89, 0x50, 0xfe, 0xff}
harIn := traffic.HAR{
Log: &traffic.HARLog{
Version: "1.2",
Creator: traffic.HARCreator{Name: "TestSuite", Version: "1.0"},
Entries: []traffic.HAREntry{
{
StartedDateTime: "2026-06-01T08:00:00Z",
Request: traffic.HARRequest{
Method: "POST",
URL: "https://har-a.example/api/login",
Headers: []traffic.HARHeader{{Name: "Content-Type", Value: "application/json"}},
PostData: &traffic.HARPostData{
MimeType: "application/json",
Text: `{"user":"alice"}`,
},
},
Response: traffic.HARResponse{
Status: 200,
Headers: []traffic.HARHeader{{Name: "Content-Type", Value: "application/json"}},
Content: traffic.HARContent{
MimeType: "application/json",
Text: `{"ok":true}`,
},
},
},
{
StartedDateTime: "2026-06-01T08:00:10Z",
Request: traffic.HARRequest{
Method: "GET",
URL: "https://har-b.example/blob.bin",
Headers: []traffic.HARHeader{},
},
Response: traffic.HARResponse{
Status: 200,
Headers: []traffic.HARHeader{{Name: "Content-Type", Value: "application/octet-stream"}},
Content: traffic.HARContent{
MimeType: "application/octet-stream",
Text: base64.StdEncoding.EncodeToString(binPayload),
Encoding: "base64",
},
},
},
},
},
}
rawIn, err := json.Marshal(harIn)
if err != nil {
t.Fatal(err)
}
// Import via HTTP endpoint.
w := request(http.MethodPost, "/api/traffic/import/har", string(rawIn))
if w.Code != http.StatusOK {
t.Fatalf("POST /api/traffic/import/har status=%d body=%s", w.Code, w.Body.String())
}
var importRes traffic.HARImportResult
if err := json.Unmarshal(w.Body.Bytes(), &importRes); err != nil {
t.Fatal(err)
}
if importRes.Imported != 2 || importRes.Skipped != 0 || importRes.Failed != 0 {
t.Fatalf("importRes = %+v, want 2/0/0", importRes)
}
// Re-import is idempotent.
w2 := request(http.MethodPost, "/api/traffic/import/har", string(rawIn))
if w2.Code != http.StatusOK {
t.Fatalf("re-import status=%d body=%s", w2.Code, w2.Body.String())
}
var importRes2 traffic.HARImportResult
if err := json.Unmarshal(w2.Body.Bytes(), &importRes2); err != nil {
t.Fatal(err)
}
if importRes2.Imported != 0 || importRes2.Skipped != 2 || importRes2.Failed != 0 {
t.Fatalf("re-import res = %+v, want 0/2/0", importRes2)
}
// Imported rows are listed on GET /api/traffic with rewritten=false and zero exchange_rewrites rows.
wList := request(http.MethodGet, "/api/traffic?host=har-", "")
if wList.Code != http.StatusOK {
t.Fatalf("GET /api/traffic status=%d body=%s", wList.Code, wList.Body.String())
}
var listResp struct {
Total int `json:"total"`
Exchanges []TrafficExchangeDTO `json:"exchanges"`
}
if err := json.Unmarshal(wList.Body.Bytes(), &listResp); err != nil {
t.Fatal(err)
}
if listResp.Total != 2 || len(listResp.Exchanges) != 2 {
t.Fatalf("listResp = %+v, want 2 exchanges", listResp)
}
for _, row := range listResp.Exchanges {
if row.Rewritten {
t.Fatalf("imported row %s marked rewritten", row.ID)
}
}
var rwCount int
if err := s.m.Traffic().DB().QueryRow(`SELECT COUNT(*) FROM exchange_rewrites`).Scan(&rwCount); err != nil || rwCount != 0 {
t.Fatalf("exchange_rewrites count after HTTP HAR import = %d err=%v, want 0", rwCount, err)
}
// Filtered export by host returns only har-b.example with binary payload preserved as base64.
wExp := request(http.MethodGet, "/api/traffic/export/har?host=har-b.example", "")
if wExp.Code != http.StatusOK {
t.Fatalf("GET /api/traffic/export/har status=%d body=%s", wExp.Code, wExp.Body.String())
}
if ct := wExp.Header().Get("Content-Type"); !strings.HasPrefix(ct, "application/json") {
t.Fatalf("export Content-Type = %q, want application/json", ct)
}
if wExp.Header().Get("X-ARTEX-HAR-Truncated") != "false" || wExp.Header().Get("X-ARTEX-HAR-Exported") != "1" || wExp.Header().Get("X-ARTEX-HAR-Matched") != "1" {
t.Fatalf("unexpected export headers: %+v", wExp.Header())
}
var exported traffic.HAR
if err := json.Unmarshal(wExp.Body.Bytes(), &exported); err != nil {
t.Fatal(err)
}
if exported.Log == nil || exported.Log.Version != "1.2" || len(exported.Log.Entries) != 1 {
t.Fatalf("filtered export = %+v", exported.Log)
}
entry := exported.Log.Entries[0]
if entry.Request.URL != "https://har-b.example/blob.bin" || entry.Response.Content.Encoding != "base64" {
t.Fatalf("exported binary entry = %+v", entry)
}
decoded, err := base64.StdEncoding.DecodeString(entry.Response.Content.Text)
if err != nil || !bytes.Equal(decoded, binPayload) {
t.Fatalf("exported binary bytes mismatch: got=%x err=%v", decoded, err)
}
// Re-importing the exported HAR via HTTP is idempotent.
wReimp := request(http.MethodPost, "/api/traffic/import/har", wExp.Body.String())
if wReimp.Code != http.StatusOK || !strings.Contains(wReimp.Body.String(), `"imported":0`) || !strings.Contains(wReimp.Body.String(), `"skipped":1`) {
t.Fatalf("re-import exported HAR status=%d body=%s", wReimp.Code, wReimp.Body.String())
}
// Error cases -> 400 naming the reason.
for _, tc := range []struct {
name string
body string
want string
}{
{"malformed", `{"log":`, "malformed JSON"},
{"missing entries", `{"log":{"version":"1.2"}}`, "missing log.entries"},
{"unsupported version", `{"log":{"version":"2.0","entries":[]}}`, "unsupported HAR version"},
} {
wErr := request(http.MethodPost, "/api/traffic/import/har", tc.body)
if wErr.Code != http.StatusBadRequest || !strings.Contains(wErr.Body.String(), tc.want) {
t.Fatalf("%s: status=%d body=%s, want 400 containing %q", tc.name, wErr.Code, wErr.Body.String(), tc.want)
}
}
// Traffic store unavailable -> 503 on both import and export.
sOff := &Server{m: &Manager{}}
wOffImp := httptest.NewRecorder()
sOff.importTrafficHAR(wOffImp, httptest.NewRequest(http.MethodPost, "/api/traffic/import/har", strings.NewReader(`{}`)))
if wOffImp.Code != http.StatusServiceUnavailable {
t.Fatalf("import with traffic off status=%d, want 503", wOffImp.Code)
}
wOffExp := httptest.NewRecorder()
sOff.exportTrafficHAR(wOffExp, httptest.NewRequest(http.MethodGet, "/api/traffic/export/har", nil))
if wOffExp.Code != http.StatusServiceUnavailable {
t.Fatalf("export with traffic off status=%d, want 503", wOffExp.Code)
}
}
func TestTrafficHAREndpointsRequireAuth(t *testing.T) {
s, _ := ingestServer(t)
rImp := httptest.NewRequest(http.MethodPost, "/api/traffic/import/har", strings.NewReader(`{"log":{"version":"1.2","entries":[]}}`))
wImp := httptest.NewRecorder()
s.Handler().ServeHTTP(wImp, rImp)
if wImp.Code != http.StatusUnauthorized {
t.Fatalf("unauthenticated HAR import status=%d, want 401", wImp.Code)
}
rExp := httptest.NewRequest(http.MethodGet, "/api/traffic/export/har", nil)
wExp := httptest.NewRecorder()
s.Handler().ServeHTTP(wExp, rExp)
if wExp.Code != http.StatusUnauthorized {
t.Fatalf("unauthenticated HAR export status=%d, want 401", wExp.Code)
}
}
func TestTrafficHARExportNearCapReimportsThroughHTTPHandler(t *testing.T) {
s, request := ingestServer(t)
tr := s.m.Traffic()
// Seed a binary exchange near the 20 MiB export raw cap (~19.5 MiB).
const bodySize = 19*1024*1024 + 512*1024
binBody := make([]byte, bodySize)
for i := range binBody {
binBody[i] = byte((i*17 + 0x81) & 0xff)
}
binBody[0] = 0x00
rawSeed, err := json.Marshal(traffic.HAR{
Log: &traffic.HARLog{
Version: "1.2",
Entries: []traffic.HAREntry{{
StartedDateTime: "2026-06-02T00:00:00Z",
Request: traffic.HARRequest{Method: "GET", URL: "https://nearcap.example/bin", Headers: []traffic.HARHeader{}},
Response: traffic.HARResponse{
Status: 200,
Headers: []traffic.HARHeader{{Name: "Content-Type", Value: "application/octet-stream"}},
Content: traffic.HARContent{
Size: len(binBody),
MimeType: "application/octet-stream",
Text: base64.StdEncoding.EncodeToString(binBody),
Encoding: "base64",
},
},
}},
},
})
if err != nil {
t.Fatal(err)
}
if _, err := tr.ImportHAR(rawSeed, traffic.HARImportOptions{}); err != nil {
t.Fatal(err)
}
wExp := request(http.MethodGet, "/api/traffic/export/har?host=nearcap.example", "")
if wExp.Code != http.StatusOK {
t.Fatalf("export status=%d", wExp.Code)
}
if wExp.Header().Get("X-ARTEX-HAR-Truncated") != "false" {
t.Fatalf("expected non-truncated export, headers=%+v", wExp.Header())
}
if wExp.Body.Len() >= decodeMaxBytes {
t.Fatalf("exported HAR size=%d exceeds decodeMaxBytes=%d", wExp.Body.Len(), decodeMaxBytes)
}
// Re-import the exact exported bytes through the HTTP handler (including
// limitRequestBody 32 MiB middleware) — must answer 200, never 413.
wImp := request(http.MethodPost, "/api/traffic/import/har", wExp.Body.String())
if wImp.Code != http.StatusOK {
t.Fatalf("re-import near-cap export via HTTP status=%d body=%s, want 200", wImp.Code, wImp.Body.String())
}
}
+21
View File
@@ -0,0 +1,21 @@
package server
import (
"net/http"
"testing"
)
// apply_rules is a boolean in the API contract: a string or a number is a 400,
// not a silent false or true.
func TestAdversarialReplayRouteRejectsNonBoolApplyRules(t *testing.T) {
_, replay := replayServer(t, true, false)
for _, body := range []string{
`{"method":"GET","url":"http://example.com/","apply_rules":"yes"}`,
`{"method":"GET","url":"http://example.com/","apply_rules":1}`,
} {
w := replay(body)
if w.Code != http.StatusBadRequest {
t.Fatalf("%s = %d %s, want 400", body, w.Code, w.Body)
}
}
}
+78
View File
@@ -14,6 +14,7 @@ import (
"time"
"github.com/Autumn-27/artex/traffic"
"github.com/Autumn-27/artex/traffic/rewrite"
)
// replayServer wires a Server like ingestServer. With traffic off,
@@ -161,3 +162,80 @@ func TestTrafficReplayRouteEndToEnd(t *testing.T) {
time.Sleep(20 * time.Millisecond)
}
}
func TestGetTrafficExposesRewriteStats(t *testing.T) {
// Traffic disabled: rewrite object still present with zero counters.
sOff := &Server{m: &Manager{}}
wOff := httptest.NewRecorder()
sOff.getTraffic(wOff, httptest.NewRequest(http.MethodGet, "/api/traffic", nil))
if wOff.Code != http.StatusOK {
t.Fatalf("getTraffic(off) status = %d", wOff.Code)
}
var bodyOff map[string]any
if err := json.Unmarshal(wOff.Body.Bytes(), &bodyOff); err != nil {
t.Fatal(err)
}
rwOff, ok := bodyOff["rewrite"].(map[string]any)
if !ok {
t.Fatalf("getTraffic(off) missing rewrite object: %s", wOff.Body.String())
}
for _, k := range []string{"large_responses", "capacity_refusals", "tombstone_drops", "conversions", "sweeps"} {
if _, exists := rwOff[k]; !exists {
t.Fatalf("getTraffic(off) rewrite object missing key %q: %v", k, rwOff)
}
}
// Traffic store open: rewrite counters mirror tr.RewriteStats().
tr, err := traffic.Open(t.TempDir(), ":0")
if err != nil {
t.Fatal(err)
}
t.Cleanup(func() { tr.Close() })
sOn := &Server{m: &Manager{traffic: tr}}
wOn := httptest.NewRecorder()
sOn.getTraffic(wOn, httptest.NewRequest(http.MethodGet, "/api/traffic", nil))
if wOn.Code != http.StatusOK {
t.Fatalf("getTraffic(on) status = %d", wOn.Code)
}
var bodyOn struct {
Rewrite traffic.RewriteStats `json:"rewrite"`
}
if err := json.Unmarshal(wOn.Body.Bytes(), &bodyOn); err != nil {
t.Fatal(err)
}
if bodyOn.Rewrite != tr.RewriteStats() {
t.Fatalf("getTraffic rewrite = %+v, want %+v", bodyOn.Rewrite, tr.RewriteStats())
}
}
func TestTrafficReplayRouteApplyRulesToggle(t *testing.T) {
s, replay := replayServer(t, true, true)
target := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
w.Header().Set("X-Seen-Rule", r.Header.Get("X-Rule-Tag"))
_, _ = io.WriteString(w, "ok")
}))
t.Cleanup(target.Close)
set := rewrite.Compile([]rewrite.Rule{{
ID: 7, Name: "tag", Enabled: true, Phase: rewrite.PhaseRequest,
Actions: []rewrite.Action{{Type: rewrite.ActionSetHeader, Name: "X-Rule-Tag", Value: "on"}},
}}, rewrite.Options{})
if len(set.LoadErrors()) != 0 {
t.Fatalf("load errors: %v", set.LoadErrors())
}
s.m.Traffic().SetRewriteRules(set)
// Default (apply_rules absent -> false): bypasses rule.
reqOff, _ := json.Marshal(map[string]any{"method": "GET", "url": target.URL + "/toggle-off"})
wOff := replay(string(reqOff))
if wOff.Code != 200 || strings.Contains(wOff.Body.String(), `"X-Seen-Rule","value":"on"`) {
t.Fatalf("default replay applied rewrite rule: %d %s", wOff.Code, wOff.Body.String())
}
// Opt-in (apply_rules: true): applies rule.
reqOn, _ := json.Marshal(map[string]any{"method": "GET", "url": target.URL + "/toggle-on", "apply_rules": true})
wOn := replay(string(reqOn))
if wOn.Code != 200 || !strings.Contains(wOn.Body.String(), `"value":"on"`) {
t.Fatalf("apply_rules=true did not apply rewrite rule: %d %s", wOn.Code, wOn.Body.String())
}
}
+363
View File
@@ -0,0 +1,363 @@
package server
import (
"context"
"encoding/json"
"fmt"
"net/http"
"net/http/httptest"
"os"
"path/filepath"
"slices"
"strconv"
"strings"
"testing"
"github.com/Autumn-27/artex/db"
"github.com/Autumn-27/artex/traffic"
)
func TestGetTrafficRewrittenFlagAndFilter(t *testing.T) {
tr, err := traffic.Open(t.TempDir(), ":0")
if err != nil {
t.Fatal(err)
}
t.Cleanup(func() { tr.Close() })
insertEx := func(id string, ts int64, host string, status int) {
t.Helper()
if _, err := tr.DB().Exec(`INSERT INTO exchanges(id,ts,host,method,url_template,url,status,content_type,req_len,resp_len,path)
VALUES(?,?,?,?,?,?,?,?,?,?,?)`, id, ts, host, "GET", "/"+id, "https://"+host+"/"+id, status, "application/json", 0, 12, ""); err != nil {
t.Fatal(err)
}
}
insertEx("rw-alpha-200", 10, "alpha.example", 200)
insertEx("rw-beta-404", 20, "beta.example", 404)
insertEx("orig-alpha-200", 30, "alpha.example", 200)
insertEx("skip-alpha-404", 40, "alpha.example", 404)
if _, err := tr.DB().Exec(`INSERT INTO exchange_rewrites(exchange_id,seq,phase,rule_id,outcome) VALUES
('rw-alpha-200',0,'request',1,'applied'),
('rw-beta-404',0,'response',2,'applied'),
('skip-alpha-404',0,'request',3,'skipped')`); err != nil {
t.Fatal(err)
}
s := &Server{m: &Manager{traffic: tr}}
type respPayload struct {
Count int `json:"count"`
Total int `json:"total"`
Exchanges []TrafficExchangeDTO `json:"exchanges"`
}
fetch := func(query string) (respPayload, []map[string]any) {
t.Helper()
w := httptest.NewRecorder()
s.getTraffic(w, httptest.NewRequest(http.MethodGet, "/api/traffic"+query, nil))
if w.Code != http.StatusOK {
t.Fatalf("GET /api/traffic%s status=%d body=%s", query, w.Code, w.Body.String())
}
var out respPayload
if err := json.Unmarshal(w.Body.Bytes(), &out); err != nil {
t.Fatal(err)
}
var raw struct {
Exchanges []map[string]any `json:"exchanges"`
}
if err := json.Unmarshal(w.Body.Bytes(), &raw); err != nil {
t.Fatal(err)
}
return out, raw.Exchanges
}
// Default (absent) and rewritten=all return all 4 exchanges and each row carries explicit "rewritten" bool.
for _, q := range []string{"", "?rewritten=all"} {
res, rawRows := fetch(q)
if res.Count != 4 || res.Total != 4 || len(res.Exchanges) != 4 {
t.Fatalf("query %q count=%d total=%d len=%d, want 4/4/4", q, res.Count, res.Total, len(res.Exchanges))
}
flags := map[string]bool{}
for i, row := range res.Exchanges {
if _, ok := rawRows[i]["rewritten"].(bool); !ok {
t.Fatalf("query %q row %s missing boolean rewritten field: %v", q, row.ID, rawRows[i])
}
flags[row.ID] = row.Rewritten
}
if !flags["rw-alpha-200"] || !flags["rw-beta-404"] || flags["orig-alpha-200"] || flags["skip-alpha-404"] {
t.Fatalf("query %q flags=%v", q, flags)
}
}
// Filter rewritten (true and canonical "rewritten").
for _, q := range []string{"?rewritten=true", "?rewritten=rewritten"} {
res, _ := fetch(q)
if res.Count != 4 || res.Total != 2 || len(res.Exchanges) != 2 {
t.Fatalf("query %q count=%d total=%d len=%d, want 4/2/2", q, res.Count, res.Total, len(res.Exchanges))
}
ids := []string{res.Exchanges[0].ID, res.Exchanges[1].ID}
slices.Sort(ids)
if !slices.Equal(ids, []string{"rw-alpha-200", "rw-beta-404"}) {
t.Fatalf("query %q ids=%v", q, ids)
}
for _, row := range res.Exchanges {
if !row.Rewritten {
t.Fatalf("query %q returned non-rewritten row %s", q, row.ID)
}
}
}
// Filter original (false and canonical "original"): finds plain + skipped-only, excludes rewritten.
for _, q := range []string{"?rewritten=false", "?rewritten=original"} {
res, _ := fetch(q)
if res.Count != 4 || res.Total != 2 || len(res.Exchanges) != 2 {
t.Fatalf("query %q count=%d total=%d len=%d, want 4/2/2", q, res.Count, res.Total, len(res.Exchanges))
}
ids := []string{res.Exchanges[0].ID, res.Exchanges[1].ID}
slices.Sort(ids)
if !slices.Equal(ids, []string{"orig-alpha-200", "skip-alpha-404"}) {
t.Fatalf("query %q ids=%v", q, ids)
}
for _, row := range res.Exchanges {
if row.Rewritten {
t.Fatalf("query %q returned rewritten row %s", q, row.ID)
}
}
}
// Combined with host and status.
resHostRW, _ := fetch("?rewritten=true&host=alpha.example&status=200")
if resHostRW.Total != 1 || len(resHostRW.Exchanges) != 1 || resHostRW.Exchanges[0].ID != "rw-alpha-200" {
t.Fatalf("combined rewritten+host+status = %+v", resHostRW)
}
resHostOrig, _ := fetch("?rewritten=false&host=alpha.example&status=4xx")
if resHostOrig.Total != 1 || len(resHostOrig.Exchanges) != 1 || resHostOrig.Exchanges[0].ID != "skip-alpha-404" {
t.Fatalf("combined original+host+4xx = %+v", resHostOrig)
}
// Pagination stays consistent with total under rewritten filter.
p0, _ := fetch("?rewritten=rewritten&page=0&size=1")
p1, _ := fetch("?rewritten=rewritten&page=1&size=1")
if p0.Total != 2 || p1.Total != 2 || len(p0.Exchanges) != 1 || len(p1.Exchanges) != 1 || p0.Exchanges[0].ID == p1.Exchanges[0].ID {
t.Fatalf("paginated rewritten p0=%+v p1=%+v", p0, p1)
}
}
func TestTaskArchiveRestorePreservesRewriteProvenanceAndV1DoesNot(t *testing.T) {
s, f, req := trafficEvidenceServer(t)
ctx := context.Background()
finding, err := s.m.pg.GetFinding(f.FindingID)
if err != nil {
t.Fatal(err)
}
tid := *finding.TaskID
s.ctx = ctx
seedHostFlow := func(host, id, path string) {
t.Helper()
if _, err := s.m.traffic.DB().Exec(`INSERT INTO exchanges(id,ts,host,method,url_template,url,status,content_type,req_len,resp_len,path)
VALUES(?,100,?,'GET',?,?,200,'text/plain',0,4,'')`, id, host, path, "https://"+host+path); err != nil {
t.Fatal(err)
}
if _, err := s.m.traffic.DB().Exec(`INSERT INTO exchange_bodies(id,req_head,resp_head,req_body,resp_body)
VALUES(?,?, 'HTTP/1.1 200 OK', ?, ?)`, id, "GET "+path+" HTTP/1.1\nHost: "+host, []byte{}, []byte("body")); err != nil {
t.Fatal(err)
}
}
callTrafficGet := func(id string) string {
t.Helper()
for _, tool := range s.m.traffic.Tools() {
if tool.Name() != "traffic_get" {
continue
}
res, err := tool.Call(ctx, json.RawMessage(fmt.Sprintf(`{"id":%q}`, id)), nil)
if err != nil || res.IsError {
t.Fatalf("traffic_get(%s) err=%v res=%s", id, err, res.Flatten())
}
return res.Flatten()
}
t.Fatal("traffic_get tool not found")
return ""
}
bindAndFetchViewer := func(findingID int64, trafficID string) (rewritten bool, ruleIDs []int64) {
t.Helper()
list, err := s.evidenceStore().Bind(ctx, findingID, []db.TrafficRef{{TrafficID: trafficID, Role: "proof"}})
if err != nil {
t.Fatalf("Bind(%s): %v", trafficID, err)
}
var bindingID int64
for _, b := range list.Bindings {
if b.Snapshot.SourceTrafficID == trafficID {
bindingID = b.ID
break
}
}
if bindingID == 0 {
t.Fatalf("binding for %s not found in %+v", trafficID, list.Bindings)
}
w := req(http.MethodGet, fmt.Sprintf("/api/exploration/findings/%d/traffic/%d", findingID, bindingID), "")
if w.Code != http.StatusOK {
t.Fatalf("evidence viewer GET status=%d body=%s", w.Code, w.Body.String())
}
var detail struct {
Binding db.FindingTrafficBinding `json:"binding"`
}
if err := json.Unmarshal(w.Body.Bytes(), &detail); err != nil {
t.Fatal(err)
}
return detail.Binding.Snapshot.Rewritten, detail.Binding.Snapshot.RewriteRuleIDs
}
// ---- Part 1: v2 rewritten archive restored into task ----
const v2Host = "v2-archive-provenance.local"
var v2AssetID int64
if err := s.m.pg.QueryRow(`INSERT INTO assets(type,domain,root_domain,task_ids) VALUES('subdomain',$1,$1,ARRAY[$2]::bigint[]) RETURNING id`, v2Host, tid).Scan(&v2AssetID); err != nil {
t.Fatal(err)
}
t.Cleanup(func() { _, _ = s.m.pg.Exec(`DELETE FROM assets WHERE id=$1`, v2AssetID) })
seedHostFlow(v2Host, "ex-v2-rw", "/rw")
seedHostFlow(v2Host, "ex-v2-orig", "/orig")
if _, err := s.m.traffic.DB().Exec(`INSERT INTO exchange_rewrites(exchange_id,seq,phase,rule_id,outcome) VALUES
('ex-v2-rw',0,'request',7,'applied'),
('ex-v2-rw',1,'response',2,'applied'),
('ex-v2-rw',2,'response',9,'skipped')`); err != nil {
t.Fatal(err)
}
if err := s.m.SetTaskPaused(strconv.FormatInt(tid, 10), true); err != nil {
t.Fatal(err)
}
job, err := s.m.pg.QueueTaskArchive(tid)
if err != nil {
t.Fatal(err)
}
if err := s.runOneTaskArchiveJob(); err != nil {
t.Fatalf("archive v2 task: %v", err)
}
// Exclusive host traffic must have been compacted out of the hot SQLite store.
if _, _, err := s.m.traffic.Get("ex-v2-rw"); err == nil {
t.Fatal("expected hot traffic to be compacted by archive")
}
if _, err := s.m.pg.QueueTaskArchiveRestore(job.ID); err != nil {
t.Fatal(err)
}
if err := s.runOneTaskArchiveJob(); err != nil {
t.Fatalf("restore v2 task: %v", err)
}
if out := callTrafficGet("ex-v2-rw"); !strings.HasPrefix(out, "=== REWRITTEN === rewrite rules [2 7] ") {
t.Fatalf("restored v2 traffic_get(ex-v2-rw) = %q, want REWRITTEN line", out)
}
if out := callTrafficGet("ex-v2-orig"); strings.Contains(out, "REWRITTEN") {
t.Fatalf("restored v2 traffic_get(ex-v2-orig) unexpectedly had REWRITTEN: %q", out)
}
if rw, ids := bindAndFetchViewer(f.FindingID, "ex-v2-rw"); !rw || !slices.Equal(ids, []int64{2, 7}) {
t.Fatalf("restored v2 evidence viewer(ex-v2-rw) = (%v, %v), want (true, [2 7])", rw, ids)
}
if rw, ids := bindAndFetchViewer(f.FindingID, "ex-v2-orig"); rw || len(ids) != 0 {
t.Fatalf("restored v2 evidence viewer(ex-v2-orig) = (%v, %v), want (false, nil)", rw, ids)
}
// ---- Part 2: v1 traffic archive restored into a task must NOT gain a REWRITTEN line ----
taskV1, err := s.m.CreateTask("v1 traffic archive", "verify v1 restore", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
tidV1, _ := strconv.ParseInt(taskV1.ID, 10, 64)
t.Cleanup(func() {
_, _ = s.m.pg.Exec(`DELETE FROM task_archives WHERE task_id=$1`, tidV1)
_ = s.m.pg.DeleteTask(tidV1)
})
fV1, err := s.evidenceStore().Record(ctx, db.RecordFindingInput{
TaskID: tidV1, ExplorationID: taskV1.ExpID, Worker: "test",
VulnClass: "TEST", Summary: "v1 evidence", Name: "V1 Finding", Severity: "low",
}, nil)
if err != nil {
t.Fatal(err)
}
const v1Host = "v1-archive-legacy.local"
var v1AssetID int64
if err := s.m.pg.QueryRow(`INSERT INTO assets(type,domain,root_domain,task_ids) VALUES('subdomain',$1,$1,ARRAY[$2]::bigint[]) RETURNING id`, v1Host, tidV1).Scan(&v1AssetID); err != nil {
t.Fatal(err)
}
t.Cleanup(func() { _, _ = s.m.pg.Exec(`DELETE FROM assets WHERE id=$1`, v1AssetID) })
seedHostFlow(v1Host, "ex-v1-legacy", "/legacy")
if err := s.m.SetTaskPaused(taskV1.ID, true); err != nil {
t.Fatal(err)
}
jobV1, err := s.m.pg.QueueTaskArchive(tidV1)
if err != nil {
t.Fatal(err)
}
if err := s.runOneTaskArchiveJob(); err != nil {
t.Fatalf("archive v1 fixture task: %v", err)
}
// Repackage the task archive with a version-1 traffic.json (no rewrite provenance fields).
archivedV1, err := s.m.pg.GetTaskArchive(jobV1.ID)
if err != nil {
t.Fatal(err)
}
repackDir := t.TempDir()
if err := extractTaskArchivePackage(archivedV1.ArchivePath, archivedV1.SHA256, repackDir); err != nil {
t.Fatal(err)
}
manifestRaw, err := os.ReadFile(filepath.Join(repackDir, "manifest.json"))
if err != nil {
t.Fatal(err)
}
var taskSnap db.TaskArchiveSnapshot
if err := json.Unmarshal(manifestRaw, &taskSnap); err != nil {
t.Fatal(err)
}
trafficPath := filepath.Join(repackDir, "traffic", "traffic.json")
rawTraffic, err := os.ReadFile(trafficPath)
if err != nil {
t.Fatal(err)
}
var trafficSnap traffic.ArchiveSnapshot
if err := json.Unmarshal(rawTraffic, &trafficSnap); err != nil {
t.Fatal(err)
}
trafficSnap.Version = 1
for i := range trafficSnap.Exchanges {
trafficSnap.Exchanges[i].Rewritten = false
trafficSnap.Exchanges[i].RewriteRuleIDs = nil
}
v1Raw, err := json.Marshal(trafficSnap)
if err != nil {
t.Fatal(err)
}
if err := os.WriteFile(trafficPath, v1Raw, 0o600); err != nil {
t.Fatal(err)
}
if err := os.Remove(archivedV1.ArchivePath); err != nil {
t.Fatal(err)
}
_, _, newSHA, err := writeTaskArchivePackage(archivedV1.ArchivePath, repackDir, &taskSnap)
if err != nil {
t.Fatal(err)
}
if _, err := s.m.pg.Exec(`UPDATE task_archives SET sha256=$1 WHERE id=$2`, newSHA, jobV1.ID); err != nil {
t.Fatal(err)
}
if _, err := s.m.pg.QueueTaskArchiveRestore(jobV1.ID); err != nil {
t.Fatal(err)
}
if err := s.runOneTaskArchiveJob(); err != nil {
t.Fatalf("restore v1 task: %v", err)
}
if out := callTrafficGet("ex-v1-legacy"); strings.Contains(out, "REWRITTEN") {
t.Fatalf("restored v1 traffic_get(ex-v1-legacy) unexpectedly gained REWRITTEN: %q", out)
}
if rw, ids := bindAndFetchViewer(fV1.FindingID, "ex-v1-legacy"); rw || len(ids) != 0 {
t.Fatalf("restored v1 evidence viewer(ex-v1-legacy) = (%v, %v), want (false, nil)", rw, ids)
}
}
+98 -28
View File
@@ -9,10 +9,16 @@ import (
"fmt"
"os"
"path/filepath"
"slices"
"sort"
"strings"
)
const (
archiveVersionV1 = 1
archiveVersionV2 = 2
)
// ArchiveSnapshot is the portable traffic subset embedded in one task archive.
// Large content-addressed bodies are copied alongside this manifest rather than
// base64-encoded, preserving deduplication and allowing the tar writer to stream.
@@ -23,22 +29,24 @@ type ArchiveSnapshot struct {
}
type ArchiveExchange struct {
ID string `json:"id"`
TS int64 `json:"ts"`
Host string `json:"host"`
Method string `json:"method"`
URLTemplate string `json:"url_template"`
URL string `json:"url"`
Status int `json:"status"`
ContentType string `json:"content_type"`
ReqLen int `json:"req_len"`
RespLen int `json:"resp_len"`
ReqHead string `json:"req_head"`
ReqBody []byte `json:"req_body,omitempty"`
ReqBlob string `json:"req_blob,omitempty"`
RespHead string `json:"resp_head"`
RespBody []byte `json:"resp_body,omitempty"`
RespBlob string `json:"resp_blob,omitempty"`
ID string `json:"id"`
TS int64 `json:"ts"`
Host string `json:"host"`
Method string `json:"method"`
URLTemplate string `json:"url_template"`
URL string `json:"url"`
Status int `json:"status"`
ContentType string `json:"content_type"`
ReqLen int `json:"req_len"`
RespLen int `json:"resp_len"`
ReqHead string `json:"req_head"`
ReqBody []byte `json:"req_body,omitempty"`
ReqBlob string `json:"req_blob,omitempty"`
RespHead string `json:"resp_head"`
RespBody []byte `json:"resp_body,omitempty"`
RespBlob string `json:"resp_blob,omitempty"`
Rewritten bool `json:"rewritten,omitempty"`
RewriteRuleIDs []int64 `json:"rewrite_rule_ids,omitempty"`
}
// ExportHosts writes an exact-host snapshot to dir. It holds the traffic writer
@@ -66,7 +74,7 @@ FROM exchanges WHERE host IN (`+strings.Join(placeholders, ",")+`) ORDER BY ts,i
return 0, err
}
defer rows.Close()
snapshot := ArchiveSnapshot{Version: 1}
snapshot := ArchiveSnapshot{Version: archiveVersionV2}
blobs := map[string]struct{}{}
for rows.Next() {
var item ArchiveExchange
@@ -102,21 +110,37 @@ FROM exchange_bodies WHERE id=?`, item.ID).Scan(&item.ReqHead, &item.ReqBody, &r
if err := rows.Err(); err != nil {
return 0, err
}
for hash := range blobs {
source, err := t.blobPath(hash)
_ = rows.Close()
for i := range snapshot.Exchanges {
rewritten, ruleIDs, err := t.RewriteInfo(snapshot.Exchanges[i].ID)
if err != nil {
return 0, err
}
data, err := os.ReadFile(source)
if err != nil {
return 0, err
if rewritten {
snapshot.Exchanges[i].Rewritten = true
snapshot.Exchanges[i].RewriteRuleIDs = ruleIDs
}
}
if len(blobs) > 0 {
snapshot.Blobs = make([]string, 0, len(blobs))
for hash := range blobs {
snapshot.Blobs = append(snapshot.Blobs, hash)
}
sort.Strings(snapshot.Blobs)
for _, hash := range snapshot.Blobs {
source, err := t.blobPath(hash)
if err != nil {
return 0, err
}
data, err := os.ReadFile(source)
if err != nil {
return 0, err
}
if err := os.WriteFile(filepath.Join(dir, "blobs", hash+".bin"), data, 0o600); err != nil {
return 0, err
}
}
if err := os.WriteFile(filepath.Join(dir, "blobs", hash+".bin"), data, 0o600); err != nil {
return 0, err
}
snapshot.Blobs = append(snapshot.Blobs, hash)
}
sort.Strings(snapshot.Blobs)
raw, err := json.Marshal(snapshot)
if err != nil {
return 0, err
@@ -129,6 +153,11 @@ FROM exchange_bodies WHERE id=?`, item.ID).Scan(&item.ReqHead, &item.ReqBody, &r
// ImportArchive imports only missing exchange IDs. Current hot rows always win,
// and repeated restore attempts are safe after a partial external failure.
// Version 1 archives carry no rewrite provenance (imported as unrewritten).
// Version 2 archives carry per-exchange rewritten + rewrite_rule_ids and
// reconstruct minimal applied rows in exchange_rewrites (phase='archived',
// outcome='applied', rule_snapshot=NULL) so RewriteInfo and the rewritten list
// filter return the same provenance without inventing a rule snapshot.
func (t *Traffic) ImportArchive(dir string) (int64, error) {
if t == nil {
return 0, nil
@@ -144,7 +173,7 @@ func (t *Traffic) ImportArchive(dir string) (int64, error) {
if err := json.Unmarshal(raw, &snapshot); err != nil {
return 0, err
}
if snapshot.Version != 1 {
if snapshot.Version != archiveVersionV1 && snapshot.Version != archiveVersionV2 {
return 0, fmt.Errorf("unsupported traffic archive version %d", snapshot.Version)
}
t.wmu.Lock()
@@ -200,6 +229,11 @@ VALUES(?,?,?,?,?,?,?)`, item.ID, item.ReqHead, item.ReqBody, nullIfEmpty(item.Re
}
}
}
if snapshot.Version >= archiveVersionV2 && item.Rewritten {
if err := insertArchivedRewriteProvenance(tx, item.ID, item.RewriteRuleIDs); err != nil {
return imported, err
}
}
if t.fts {
var rowID int64
if err := tx.QueryRow(`SELECT rowid FROM exchanges WHERE id=?`, item.ID).Scan(&rowID); err != nil {
@@ -218,6 +252,42 @@ VALUES(?,?,?,?,?,?,?)`, item.ID, item.ReqHead, item.ReqBody, nullIfEmpty(item.Re
return imported, nil
}
// insertArchivedRewriteProvenance writes minimal applied rows so RewriteInfo
// and the traffic list's rewritten flag/filter return the exact archived
// provenance without fabricating a rule snapshot or request/response phase.
func insertArchivedRewriteProvenance(tx *sql.Tx, exchangeID string, ruleIDs []int64) error {
ids := normalizeArchiveRuleIDs(ruleIDs)
if len(ids) == 0 {
_, err := tx.Exec(`INSERT INTO exchange_rewrites(exchange_id,seq,phase,rule_id,outcome,reason)
VALUES(?,0,'archived',NULL,'applied','archived')`, exchangeID)
return err
}
for seq, rid := range ids {
if _, err := tx.Exec(`INSERT INTO exchange_rewrites(exchange_id,seq,phase,rule_id,outcome,reason)
VALUES(?,?,'archived',?,'applied','archived')`, exchangeID, seq, rid); err != nil {
return err
}
}
return nil
}
func normalizeArchiveRuleIDs(ids []int64) []int64 {
if len(ids) == 0 {
return nil
}
out := make([]int64, 0, len(ids))
for _, id := range ids {
if id > 0 {
out = append(out, id)
}
}
if len(out) == 0 {
return nil
}
slices.Sort(out)
return slices.Compact(out)
}
func uniqueArchiveHosts(hosts []string) []string {
seen := make(map[string]struct{}, len(hosts))
out := make([]string, 0, len(hosts))
+293
View File
@@ -1,8 +1,13 @@
package traffic
import (
"bytes"
"context"
"database/sql"
"encoding/json"
"os"
"path/filepath"
"slices"
"strings"
"testing"
)
@@ -156,3 +161,291 @@ func simulateTrafficStageProcessExit(t *testing.T, stage *HostDeleteStage) {
stage.done = true
stage.traffic.wmu.Unlock()
}
func callTrafficGetTool(t *testing.T, tr *Traffic, id string) string {
t.Helper()
for _, tool := range tr.Tools() {
if tool.Name() != "traffic_get" {
continue
}
res, err := tool.Call(context.Background(), json.RawMessage(`{"id":`+mustJSON(t, id)+`}`), nil)
if err != nil {
t.Fatalf("traffic_get(%s) error: %v", id, err)
}
if res.IsError {
t.Fatalf("traffic_get(%s) tool error: %s", id, res.Flatten())
}
return res.Flatten()
}
t.Fatal("traffic_get tool not found")
return ""
}
func mustJSON(t *testing.T, v any) string {
t.Helper()
b, err := json.Marshal(v)
if err != nil {
t.Fatal(err)
}
return string(b)
}
func TestTrafficArchiveV2RewriteProvenanceRoundTripAndByteIdentical(t *testing.T) {
src, err := Open(t.TempDir(), "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
defer src.Close()
const host = "v2-provenance.example"
insertEx := func(id string, ts int64, urlPath string) {
t.Helper()
if _, err := src.DB().Exec(`INSERT INTO exchanges(id,ts,host,method,url_template,url,status,content_type,req_len,resp_len,path)
VALUES(?,?,?,?,?,?,?,?,?,?,?)`, id, ts, host, "GET", urlPath, "https://"+host+urlPath, 200, "text/plain", 0, 2, ""); err != nil {
t.Fatal(err)
}
if _, err := src.DB().Exec(`INSERT INTO exchange_bodies(id,req_head,req_body,resp_head,resp_body)
VALUES(?,?,?,?,?)`, id, "GET "+urlPath+" HTTP/1.1\nHost: "+host, []byte{}, "HTTP/1.1 200 OK", []byte("ok")); err != nil {
t.Fatal(err)
}
}
// 1. Rewritten exchange with out-of-order + duplicate applied rule IDs plus skipped/error rows.
insertEx("ex-rewritten", 10, "/rewritten")
for seq, row := range []struct {
phase string
ruleID any
outcome string
}{
{"request", 7, "applied"},
{"request", 99, "skipped"},
{"response", 2, "applied"},
{"response", 7, "applied"},
{"response", 88, "error"},
} {
if _, err := src.DB().Exec(`INSERT INTO exchange_rewrites(exchange_id,seq,phase,rule_id,rule_snapshot,outcome,reason)
VALUES(?,?,?,?,?,?,?)`, "ex-rewritten", seq, row.phase, row.ruleID, `{"id":1}`, row.outcome, ""); err != nil {
t.Fatal(err)
}
}
// 2. Rewritten exchange with applied row carrying NULL rule_id.
insertEx("ex-rewritten-null-rule", 20, "/rewritten-null")
if _, err := src.DB().Exec(`INSERT INTO exchange_rewrites(exchange_id,seq,phase,rule_id,outcome,reason)
VALUES('ex-rewritten-null-rule',0,'request',NULL,'applied','')`); err != nil {
t.Fatal(err)
}
// 3. Exchange with only skipped/error rows -> NOT rewritten.
insertEx("ex-skipped-only", 30, "/skipped")
if _, err := src.DB().Exec(`INSERT INTO exchange_rewrites(exchange_id,seq,phase,rule_id,outcome,reason)
VALUES('ex-skipped-only',0,'request',5,'skipped','not rewritten: binary'),('ex-skipped-only',1,'response',6,'error','boom')`); err != nil {
t.Fatal(err)
}
// 4. Plain original exchange with no rewrite rows.
insertEx("ex-plain", 40, "/plain")
export1Dir := filepath.Join(t.TempDir(), "export1")
count, err := src.ExportHosts([]string{host}, export1Dir)
if err != nil || count != 4 {
t.Fatalf("ExportHosts = (%d, %v), want (4, nil)", count, err)
}
raw1, err := os.ReadFile(filepath.Join(export1Dir, "traffic.json"))
if err != nil {
t.Fatal(err)
}
var snap ArchiveSnapshot
if err := json.Unmarshal(raw1, &snap); err != nil {
t.Fatal(err)
}
if snap.Version != 2 {
t.Fatalf("archive version = %d, want 2", snap.Version)
}
if len(snap.Exchanges) != 4 {
t.Fatalf("archive exchanges = %d, want 4", len(snap.Exchanges))
}
if !snap.Exchanges[0].Rewritten || !slices.Equal(snap.Exchanges[0].RewriteRuleIDs, []int64{2, 7}) {
t.Fatalf("ex-rewritten exported = (%v, %v), want (true, [2 7])", snap.Exchanges[0].Rewritten, snap.Exchanges[0].RewriteRuleIDs)
}
if !snap.Exchanges[1].Rewritten || len(snap.Exchanges[1].RewriteRuleIDs) != 0 {
t.Fatalf("ex-rewritten-null-rule exported = (%v, %v), want (true, nil)", snap.Exchanges[1].Rewritten, snap.Exchanges[1].RewriteRuleIDs)
}
if snap.Exchanges[2].Rewritten || len(snap.Exchanges[2].RewriteRuleIDs) != 0 {
t.Fatalf("ex-skipped-only exported = (%v, %v), want (false, nil)", snap.Exchanges[2].Rewritten, snap.Exchanges[2].RewriteRuleIDs)
}
if snap.Exchanges[3].Rewritten || len(snap.Exchanges[3].RewriteRuleIDs) != 0 {
t.Fatalf("ex-plain exported = (%v, %v), want (false, nil)", snap.Exchanges[3].Rewritten, snap.Exchanges[3].RewriteRuleIDs)
}
// Import into a fresh store and verify RewriteInfo, reconstructed rows, traffic_get, and Page.
dst, err := Open(t.TempDir(), "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
defer dst.Close()
imported, err := dst.ImportArchive(export1Dir)
if err != nil || imported != 4 {
t.Fatalf("ImportArchive = (%d, %v), want (4, nil)", imported, err)
}
if rw, ids, err := dst.RewriteInfo("ex-rewritten"); err != nil || !rw || !slices.Equal(ids, []int64{2, 7}) {
t.Fatalf("dst RewriteInfo(ex-rewritten) = (%v, %v, %v), want (true, [2 7], nil)", rw, ids, err)
}
if rw, ids, err := dst.RewriteInfo("ex-rewritten-null-rule"); err != nil || !rw || len(ids) != 0 {
t.Fatalf("dst RewriteInfo(ex-rewritten-null-rule) = (%v, %v, %v), want (true, nil, nil)", rw, ids, err)
}
for _, id := range []string{"ex-skipped-only", "ex-plain"} {
if rw, ids, err := dst.RewriteInfo(id); err != nil || rw || len(ids) != 0 {
t.Fatalf("dst RewriteInfo(%s) = (%v, %v, %v), want (false, nil, nil)", id, rw, ids, err)
}
}
// Reconstructed rows must not fabricate a rule_snapshot.
var snapCount int
if err := dst.DB().QueryRow(`SELECT COUNT(*) FROM exchange_rewrites WHERE exchange_id='ex-rewritten' AND rule_snapshot IS NOT NULL`).Scan(&snapCount); err != nil || snapCount != 0 {
t.Fatalf("reconstructed rows should have NULL rule_snapshot, got count=%d err=%v", snapCount, err)
}
var phase, outcome string
var rid sql.NullInt64
if err := dst.DB().QueryRow(`SELECT phase, outcome, rule_id FROM exchange_rewrites WHERE exchange_id='ex-rewritten' AND seq=0`).Scan(&phase, &outcome, &rid); err != nil {
t.Fatal(err)
}
if phase != "archived" || outcome != "applied" || !rid.Valid || rid.Int64 != 2 {
t.Fatalf("reconstructed row 0 = (%q, %q, %+v), want ('archived', 'applied', 2)", phase, outcome, rid)
}
// traffic_get shows === REWRITTEN === for rewritten exchange and not for plain/skipped-only.
if out := callTrafficGetTool(t, dst, "ex-rewritten"); !strings.HasPrefix(out, "=== REWRITTEN === rewrite rules [2 7] ") {
t.Fatalf("traffic_get(ex-rewritten) = %q, want REWRITTEN header", out)
}
if out := callTrafficGetTool(t, dst, "ex-plain"); strings.Contains(out, "REWRITTEN") {
t.Fatalf("traffic_get(ex-plain) unexpectedly contained REWRITTEN: %q", out)
}
if out := callTrafficGetTool(t, dst, "ex-skipped-only"); strings.Contains(out, "REWRITTEN") {
t.Fatalf("traffic_get(ex-skipped-only) unexpectedly contained REWRITTEN: %q", out)
}
// Re-export from the imported store must be byte-identical.
export2Dir := filepath.Join(t.TempDir(), "export2")
if count2, err := dst.ExportHosts([]string{host}, export2Dir); err != nil || count2 != 4 {
t.Fatalf("re-export ExportHosts = (%d, %v), want (4, nil)", count2, err)
}
raw2, err := os.ReadFile(filepath.Join(export2Dir, "traffic.json"))
if err != nil {
t.Fatal(err)
}
if !bytes.Equal(raw1, raw2) {
t.Fatalf("re-exported traffic.json differs:\nfirst: %s\nsecond: %s", string(raw1), string(raw2))
}
// Re-importing over the original hot store must preserve the original live trace rows.
if importedAgain, err := src.ImportArchive(export1Dir); err != nil || importedAgain != 0 {
t.Fatalf("ImportArchive over hot store = (%d, %v), want (0, nil)", importedAgain, err)
}
var liveTraceRows int
if err := src.DB().QueryRow(`SELECT COUNT(*) FROM exchange_rewrites WHERE exchange_id='ex-rewritten'`).Scan(&liveTraceRows); err != nil || liveTraceRows != 5 {
t.Fatalf("hot store live trace rows = %d err=%v, want 5", liveTraceRows, err)
}
}
func TestTrafficArchiveV1FixtureImportsWithoutRewritten(t *testing.T) {
dir := filepath.Join(t.TempDir(), "traffic")
if err := os.MkdirAll(filepath.Join(dir, "blobs"), 0o700); err != nil {
t.Fatal(err)
}
// v1 fixture: first entry is standard v1 without rewrite fields; second entry
// tries to smuggle rewrite fields under version 1 and must still import as unrewritten.
v1JSON := `{
"version": 1,
"exchanges": [
{
"id": "v1-plain",
"ts": 100,
"host": "legacy.example",
"method": "GET",
"url_template": "/legacy",
"url": "https://legacy.example/legacy",
"status": 200,
"content_type": "text/plain",
"req_len": 0,
"resp_len": 2,
"req_head": "GET /legacy HTTP/1.1\nHost: legacy.example",
"resp_head": "HTTP/1.1 200 OK",
"resp_body": "b2s="
},
{
"id": "v1-ignored-fields",
"ts": 101,
"host": "legacy.example",
"method": "GET",
"url_template": "/legacy2",
"url": "https://legacy.example/legacy2",
"status": 200,
"content_type": "text/plain",
"req_len": 0,
"resp_len": 2,
"req_head": "GET /legacy2 HTTP/1.1\nHost: legacy.example",
"resp_head": "HTTP/1.1 200 OK",
"resp_body": "b2s=",
"rewritten": true,
"rewrite_rule_ids": [3, 9]
}
],
"blobs": []
}`
if err := os.WriteFile(filepath.Join(dir, "traffic.json"), []byte(v1JSON), 0o600); err != nil {
t.Fatal(err)
}
tr, err := Open(t.TempDir(), "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
defer tr.Close()
imported, err := tr.ImportArchive(dir)
if err != nil || imported != 2 {
t.Fatalf("ImportArchive(v1) = (%d, %v), want (2, nil)", imported, err)
}
for _, id := range []string{"v1-plain", "v1-ignored-fields"} {
if rw, ids, err := tr.RewriteInfo(id); err != nil || rw || len(ids) != 0 {
t.Fatalf("v1 RewriteInfo(%s) = (%v, %v, %v), want (false, nil, nil)", id, rw, ids, err)
}
if out := callTrafficGetTool(t, tr, id); strings.Contains(out, "REWRITTEN") {
t.Fatalf("v1 traffic_get(%s) unexpectedly has REWRITTEN: %q", id, out)
}
}
rows, total, err := tr.Page(PageQuery{RespMin: -1, RespMax: -1}, 0, 10)
if err != nil || total != 2 || len(rows) != 2 {
t.Fatalf("Page after v1 import = (%d, %d, %v), want (2, 2, nil)", len(rows), total, err)
}
for _, r := range rows {
if r.Rewritten {
t.Fatalf("v1 row %s marked rewritten", r.ID)
}
}
}
func TestTrafficArchiveUnknownVersionRefused(t *testing.T) {
tr, err := Open(t.TempDir(), "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
defer tr.Close()
for _, ver := range []int{-1, 0, 3, 99} {
dir := filepath.Join(t.TempDir(), "traffic")
if err := os.MkdirAll(dir, 0o700); err != nil {
t.Fatal(err)
}
payload := mustJSON(t, ArchiveSnapshot{Version: ver})
if err := os.WriteFile(filepath.Join(dir, "traffic.json"), []byte(payload), 0o600); err != nil {
t.Fatal(err)
}
if _, err := tr.ImportArchive(dir); err == nil || !strings.Contains(err.Error(), "unsupported traffic archive version") {
t.Fatalf("ImportArchive(version=%d) err = %v, want unsupported version error", ver, err)
}
}
}
+966
View File
@@ -0,0 +1,966 @@
package traffic
import (
"bytes"
"crypto/sha256"
"database/sql"
"encoding/base64"
"encoding/hex"
"encoding/json"
"errors"
"fmt"
"io"
"net/url"
"os"
"path/filepath"
"strings"
"time"
"unicode/utf8"
"github.com/Autumn-27/artex/db"
)
const (
// HARVersion is the HAR specification version produced and accepted.
HARVersion = "1.2"
// DefaultHARMaxImportEntries caps how many entries a single HAR import accepts.
DefaultHARMaxImportEntries = 2000
// DefaultHARMaxImportBytes caps total decoded request+response body bytes
// imported in a single HAR document.
DefaultHARMaxImportBytes int64 = 32 << 20
// DefaultHARMaxExportEntries caps how many exchanges a single HAR export emits.
DefaultHARMaxExportEntries = 1000
// DefaultHARMaxExportBytes caps total raw header + body bytes emitted in a
// single HAR export. At 20 MiB raw, even pure binary payloads expand via
// base64 (4/3x ≈ 26.7 MiB) plus HAR JSON overhead safely below the 32 MiB
// HTTP import limit.
DefaultHARMaxExportBytes int64 = 20 << 20
// DefaultHARMaxExportJSONBytes backstops the marshaled JSON size of an export
// (31 MiB, leaving 1 MiB margin below the 32 MiB import HTTP body cap) so
// even pathological JSON-escaped control-character payloads cannot exceed
// the import limit.
DefaultHARMaxExportJSONBytes int64 = 31 << 20
maxHARImportErrors = 50
maxArtexIDLen = 128
)
// ErrInvalidHAR marks a malformed or unsupported HAR document (mapped to HTTP 400).
var ErrInvalidHAR = errors.New("invalid HAR")
// HAR is a HAR 1.2 envelope. Required HAR 1.2 fields that the traffic store
// does not record use explicit unknown placeholders (time: 0, cache: {},
// timings: {send:-1, wait:-1, receive:-1}, cookies: [], queryString: [],
// headersSize: -1, redirectURL: "") rather than fabricated measurements.
type HAR struct {
Log *HARLog `json:"log"`
}
type HARLog struct {
Version string `json:"version"`
Creator HARCreator `json:"creator"`
Entries []HAREntry `json:"entries"`
Truncated bool `json:"_truncated,omitempty"`
TruncatedReason string `json:"_truncatedReason,omitempty"`
TotalMatched int `json:"_totalMatched,omitempty"`
}
type HARCreator struct {
Name string `json:"name"`
Version string `json:"version"`
}
type HARCache struct{}
type HARTimings struct {
Send float64 `json:"send"`
Wait float64 `json:"wait"`
Receive float64 `json:"receive"`
}
type HAREntry struct {
StartedDateTime string `json:"startedDateTime"`
Time float64 `json:"time"`
Request HARRequest `json:"request"`
Response HARResponse `json:"response"`
Cache HARCache `json:"cache"`
Timings HARTimings `json:"timings"`
ArtexExchangeID string `json:"_artexExchangeId,omitempty"`
}
type HARRequest struct {
Method string `json:"method"`
URL string `json:"url"`
HTTPVersion string `json:"httpVersion"`
Cookies []HARCookie `json:"cookies"`
Headers []HARHeader `json:"headers"`
QueryString []HARParam `json:"queryString"`
PostData *HARPostData `json:"postData,omitempty"`
HeadersSize int `json:"headersSize"`
BodySize int `json:"bodySize"`
}
type HARPostData struct {
MimeType string `json:"mimeType,omitempty"`
Params []HARParam `json:"params,omitempty"`
Text string `json:"text,omitempty"`
Encoding string `json:"encoding,omitempty"`
}
type HARParam struct {
Name string `json:"name"`
Value string `json:"value"`
FileName string `json:"fileName,omitempty"`
ContentType string `json:"contentType,omitempty"`
}
type HARCookie struct {
Name string `json:"name"`
Value string `json:"value"`
Path string `json:"path,omitempty"`
Domain string `json:"domain,omitempty"`
Expires string `json:"expires,omitempty"`
HTTPOnly bool `json:"httpOnly,omitempty"`
Secure bool `json:"secure,omitempty"`
}
type HARResponse struct {
Status int `json:"status"`
StatusText string `json:"statusText"`
HTTPVersion string `json:"httpVersion"`
Cookies []HARCookie `json:"cookies"`
Headers []HARHeader `json:"headers"`
Content HARContent `json:"content"`
RedirectURL string `json:"redirectURL"`
HeadersSize int `json:"headersSize"`
BodySize int `json:"bodySize"`
}
type HARContent struct {
Size int `json:"size"`
MimeType string `json:"mimeType"`
Text string `json:"text,omitempty"`
Encoding string `json:"encoding,omitempty"`
}
type HARHeader struct {
Name string `json:"name"`
Value string `json:"value"`
}
// HARImportOptions configures safety caps for ImportHAR. Zero values use the
// package defaults.
type HARImportOptions struct {
MaxEntries int
MaxTotalBytes int64
}
// HARImportError describes one entry that could not be imported.
type HARImportError struct {
Index int `json:"index"`
URL string `json:"url,omitempty"`
Reason string `json:"reason"`
}
// HARImportResult summarizes one ImportHAR run.
type HARImportResult struct {
Imported int `json:"imported"`
Skipped int `json:"skipped"`
Failed int `json:"failed"`
Truncated bool `json:"truncated,omitempty"`
Errors []HARImportError `json:"errors,omitempty"`
}
// HARExportOptions configures ExportHAR. Zero caps use package defaults.
type HARExportOptions struct {
CreatorVersion string
MaxEntries int
MaxTotalBytes int64 // raw header + body bytes cap
MaxJSONBytes int64 // marshaled JSON cap (defaults to DefaultHARMaxExportJSONBytes)
}
// HARExportMeta reports export counts and whether a cap stopped the export.
type HARExportMeta struct {
Exported int
TotalMatched int
TotalBytes int64 // total raw header + body bytes exported
Truncated bool
TruncatedReason string // "" | "max_entries" | "max_bytes"
}
// ImportHAR parses a HAR 1.2 JSON document and inserts valid entries into the
// traffic store. ARTEX-exported entries carrying a valid _artexExchangeId
// preserve that ID so distinct exchanges within the same second round-trip
// cleanly; external HAR entries receive a deterministic content-based har- ID
// so re-importing the same HAR is idempotent and skips existing rows. URL
// userinfo is stripped before storing, indexing, or reporting errors. Imported
// traffic never passes through capture rewrite rules and never gains
// exchange_rewrites rows.
func (t *Traffic) ImportHAR(raw []byte, opts HARImportOptions) (HARImportResult, error) {
if t == nil {
return HARImportResult{}, errors.New("traffic store is not available")
}
if opts.MaxEntries <= 0 {
opts.MaxEntries = DefaultHARMaxImportEntries
}
if opts.MaxTotalBytes <= 0 {
opts.MaxTotalBytes = DefaultHARMaxImportBytes
}
doc, err := decodeHARDocument(raw)
if err != nil {
return HARImportResult{}, err
}
if len(doc.Log.Entries) > opts.MaxEntries {
return HARImportResult{}, fmt.Errorf("%w: HAR contains %d entries, exceeding limit of %d", ErrInvalidHAR, len(doc.Log.Entries), opts.MaxEntries)
}
var res HARImportResult
addErr := func(idx int, rawURL, reason string) {
if len(res.Errors) < maxHARImportErrors {
res.Errors = append(res.Errors, HARImportError{
Index: idx,
URL: redactURLUserinfo(rawURL),
Reason: reason,
})
}
}
var totalBytes int64
for i, entry := range doc.Log.Entries {
parsed, err := parseHAREntry(entry)
if err != nil {
res.Failed++
addErr(i, entry.Request.URL, err.Error())
continue
}
entryBytes := int64(len(parsed.reqBody) + len(parsed.respBody))
if totalBytes+entryBytes > opts.MaxTotalBytes {
remaining := len(doc.Log.Entries) - i
res.Truncated = true
res.Skipped += remaining
addErr(i, entry.Request.URL, fmt.Sprintf("decoded body byte cap (%d bytes) reached; %d remaining entries skipped", opts.MaxTotalBytes, remaining))
break
}
inserted, err := t.insertHARExchange(parsed)
if err != nil {
res.Failed++
addErr(i, entry.Request.URL, err.Error())
continue
}
if !inserted {
res.Skipped++
continue
}
totalBytes += entryBytes
res.Imported++
}
return res, nil
}
func decodeHARDocument(raw []byte) (HAR, error) {
if len(bytes.TrimSpace(raw)) == 0 {
return HAR{}, fmt.Errorf("%w: empty HAR document", ErrInvalidHAR)
}
dec := json.NewDecoder(bytes.NewReader(raw))
var doc HAR
if err := dec.Decode(&doc); err != nil {
return HAR{}, fmt.Errorf("%w: malformed JSON: %v", ErrInvalidHAR, err)
}
// Reject trailing non-whitespace tokens after the root object.
if err := dec.Decode(&struct{}{}); err != io.EOF {
return HAR{}, fmt.Errorf("%w: malformed JSON: trailing content after HAR object", ErrInvalidHAR)
}
if doc.Log == nil {
return HAR{}, fmt.Errorf("%w: missing log object", ErrInvalidHAR)
}
if strings.TrimSpace(doc.Log.Version) != HARVersion {
return HAR{}, fmt.Errorf("%w: unsupported HAR version %q (expected %s)", ErrInvalidHAR, doc.Log.Version, HARVersion)
}
if doc.Log.Entries == nil {
return HAR{}, fmt.Errorf("%w: missing log.entries", ErrInvalidHAR)
}
return doc, nil
}
type parsedHARExchange struct {
id string
ts int64
host string
method string
urlTemplate string
rawURL string
status int
reqCT string
respCT string
reqHead string
reqBody []byte
respHead string
respBody []byte
}
func parseHAREntry(entry HAREntry) (parsedHARExchange, error) {
tsStr := strings.TrimSpace(entry.StartedDateTime)
if tsStr == "" {
return parsedHARExchange{}, errors.New("missing startedDateTime")
}
parsedTS, err := time.Parse(time.RFC3339Nano, tsStr)
if err != nil {
return parsedHARExchange{}, fmt.Errorf("invalid startedDateTime %q", entry.StartedDateTime)
}
ts := parsedTS.UTC().Unix()
rawURL := strings.TrimSpace(entry.Request.URL)
u, err := url.Parse(rawURL)
if err != nil || u.Hostname() == "" || (u.Scheme != "http" && u.Scheme != "https") {
return parsedHARExchange{}, fmt.Errorf("invalid request url %q", redactURLUserinfo(entry.Request.URL))
}
// Strip URL userinfo (user:pass@) before storing, indexing, or exporting,
// matching the asset store.
u.User = nil
method := strings.ToUpper(strings.TrimSpace(entry.Request.Method))
if method == "" || !validHeaderName(method) {
return parsedHARExchange{}, fmt.Errorf("invalid request method %q", entry.Request.Method)
}
if entry.Response.Status < 0 || entry.Response.Status > 999 {
return parsedHARExchange{}, fmt.Errorf("invalid response status %d", entry.Response.Status)
}
reqHeaderText, reqCT, hasReqCookie, err := formatHARHeaders(entry.Request.Headers, "Cookie")
if err != nil {
return parsedHARExchange{}, fmt.Errorf("invalid request headers: %w", err)
}
if !hasReqCookie && len(entry.Request.Cookies) > 0 {
cookieHeader, err := formatHARRequestCookies(entry.Request.Cookies)
if err != nil {
return parsedHARExchange{}, fmt.Errorf("invalid request cookies: %w", err)
}
reqHeaderText += cookieHeader
}
respHeaderText, respHeaderCT, hasRespSetCookie, err := formatHARHeaders(entry.Response.Headers, "Set-Cookie")
if err != nil {
return parsedHARExchange{}, fmt.Errorf("invalid response headers: %w", err)
}
if !hasRespSetCookie && len(entry.Response.Cookies) > 0 {
setCookies, err := formatHARResponseCookies(entry.Response.Cookies)
if err != nil {
return parsedHARExchange{}, fmt.Errorf("invalid response cookies: %w", err)
}
respHeaderText += setCookies
}
var reqBody []byte
if entry.Request.PostData != nil {
if entry.Request.PostData.Text != "" || len(entry.Request.PostData.Params) == 0 {
reqBody, err = decodeHARBody(entry.Request.PostData.Text, entry.Request.PostData.Encoding)
if err != nil {
return parsedHARExchange{}, fmt.Errorf("invalid request postData: %w", err)
}
if reqCT == "" {
reqCT = strings.TrimSpace(entry.Request.PostData.MimeType)
}
} else {
reqBody = encodeHARFormParams(entry.Request.PostData.Params)
if reqCT == "" {
reqCT = strings.TrimSpace(entry.Request.PostData.MimeType)
if reqCT == "" {
reqCT = "application/x-www-form-urlencoded"
}
}
}
}
respBody, err := decodeHARBody(entry.Response.Content.Text, entry.Response.Content.Encoding)
if err != nil {
return parsedHARExchange{}, fmt.Errorf("invalid response content: %w", err)
}
respCT := strings.TrimSpace(entry.Response.Content.MimeType)
if respCT == "" {
respCT = respHeaderCT
}
reqProto := sanitizeHARProto(entry.Request.HTTPVersion)
respProto := sanitizeHARProto(entry.Response.HTTPVersion)
reqHead := fmt.Sprintf("%s %s %s\n%s", method, u.RequestURI(), reqProto, reqHeaderText)
statusText := strings.TrimSpace(entry.Response.StatusText)
if strings.ContainsAny(statusText, "\r\n\x00") {
return parsedHARExchange{}, errors.New("response statusText contains control characters")
}
respHead := fmt.Sprintf("%s %d\n%s", respProto, entry.Response.Status, respHeaderText)
if statusText != "" {
respHead = fmt.Sprintf("%s %d %s\n%s", respProto, entry.Response.Status, statusText, respHeaderText)
}
canonicalURL := u.String()
id := harExchangeID(ts, method, canonicalURL, entry.Response.Status, reqBody, respBody)
if preserved := strings.TrimSpace(entry.ArtexExchangeID); validArtexExchangeID(preserved) {
id = preserved
}
return parsedHARExchange{
id: id,
ts: ts,
host: u.Hostname(),
method: method,
urlTemplate: db.TemplatePath(u.EscapedPath()),
rawURL: canonicalURL,
status: entry.Response.Status,
reqCT: reqCT,
respCT: respCT,
reqHead: reqHead,
reqBody: reqBody,
respHead: respHead,
respBody: respBody,
}, nil
}
// harExchangeID derives a deterministic ID from the exchange's timestamp (Unix
// seconds), method, canonical URL, status code, and SHA-256 digests of both
// bodies. Using Unix seconds matches exchanges.ts so export -> re-import
// produces the exact same ID for external HARs.
func harExchangeID(ts int64, method, rawURL string, status int, reqBody, respBody []byte) string {
reqSum := sha256.Sum256(reqBody)
respSum := sha256.Sum256(respBody)
h := sha256.New()
fmt.Fprintf(h, "%d\x00%s\x00%s\x00%d\x00%x\x00%x", ts, method, rawURL, status, reqSum, respSum)
return "har-" + hex.EncodeToString(h.Sum(nil))
}
func validArtexExchangeID(id string) bool {
if id == "" || len(id) > maxArtexIDLen {
return false
}
for i := 0; i < len(id); i++ {
c := id[i]
if (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z') || (c >= '0' && c <= '9') ||
c == '-' || c == '_' || c == '.' || c == ':' {
continue
}
return false
}
return true
}
// redactURLUserinfo strips user:pass@ from both parseable and malformed URLs so
// credentials never leak into stored URLs or import error details.
func redactURLUserinfo(raw string) string {
s := strings.TrimSpace(raw)
if s == "" {
return ""
}
if u, err := url.Parse(s); err == nil && u.User != nil {
u.User = nil
return u.String()
}
schemeEnd := strings.Index(s, "://")
if schemeEnd < 0 {
return s
}
rest := s[schemeEnd+3:]
authEnd := strings.IndexAny(rest, "/?#")
auth := rest
tail := ""
if authEnd >= 0 {
auth = rest[:authEnd]
tail = rest[authEnd:]
}
if at := strings.LastIndexByte(auth, '@'); at >= 0 {
return s[:schemeEnd+3] + auth[at+1:] + tail
}
return s
}
func formatHARHeaders(headers []HARHeader, checkHeader string) (formatted string, contentType string, hasChecked bool, err error) {
var b strings.Builder
for _, h := range headers {
name := strings.TrimSpace(h.Name)
if name == "" {
return "", "", false, errors.New("empty header name")
}
// Skip HTTP/2 pseudo-headers (:method, :authority, :scheme, :path, :status)
// emitted by browser DevTools HAR exports; their semantics live in the
// request/status line.
if strings.HasPrefix(name, ":") {
continue
}
if !validHeaderName(name) {
return "", "", false, fmt.Errorf("invalid header name %q", h.Name)
}
if strings.ContainsAny(h.Value, "\r\n\x00") {
return "", "", false, fmt.Errorf("header %s contains a line break or NUL byte", name)
}
if contentType == "" && strings.EqualFold(name, "Content-Type") {
contentType = strings.TrimSpace(h.Value)
}
if checkHeader != "" && strings.EqualFold(name, checkHeader) {
hasChecked = true
}
b.WriteString(name)
b.WriteString(": ")
b.WriteString(h.Value)
b.WriteByte('\n')
}
return b.String(), contentType, hasChecked, nil
}
func formatHARRequestCookies(cookies []HARCookie) (string, error) {
parts := make([]string, 0, len(cookies))
for _, c := range cookies {
name := strings.TrimSpace(c.Name)
if name == "" {
return "", errors.New("empty cookie name")
}
if strings.ContainsAny(name, " \t;=\r\n\x00") || strings.ContainsAny(c.Value, "\r\n\x00") {
return "", fmt.Errorf("invalid cookie %q", name)
}
parts = append(parts, name+"="+c.Value)
}
if len(parts) == 0 {
return "", nil
}
return "Cookie: " + strings.Join(parts, "; ") + "\n", nil
}
func formatHARResponseCookies(cookies []HARCookie) (string, error) {
var b strings.Builder
for _, c := range cookies {
name := strings.TrimSpace(c.Name)
if name == "" {
return "", errors.New("empty cookie name")
}
if strings.ContainsAny(name, " \t;=\r\n\x00") ||
strings.ContainsAny(c.Value, "\r\n\x00") ||
strings.ContainsAny(c.Path, ";\r\n\x00") ||
strings.ContainsAny(c.Domain, ";\r\n\x00") ||
strings.ContainsAny(c.Expires, ";\r\n\x00") {
return "", fmt.Errorf("invalid response cookie %q", name)
}
b.WriteString("Set-Cookie: ")
b.WriteString(name)
b.WriteByte('=')
b.WriteString(c.Value)
if c.Path != "" {
b.WriteString("; Path=")
b.WriteString(c.Path)
}
if c.Domain != "" {
b.WriteString("; Domain=")
b.WriteString(c.Domain)
}
if c.Expires != "" {
b.WriteString("; Expires=")
b.WriteString(c.Expires)
}
if c.Secure {
b.WriteString("; Secure")
}
if c.HTTPOnly {
b.WriteString("; HttpOnly")
}
b.WriteByte('\n')
}
return b.String(), nil
}
func encodeHARFormParams(params []HARParam) []byte {
parts := make([]string, 0, len(params))
for _, p := range params {
parts = append(parts, url.QueryEscape(p.Name)+"="+url.QueryEscape(p.Value))
}
return []byte(strings.Join(parts, "&"))
}
func decodeHARBody(text, encoding string) ([]byte, error) {
enc := strings.ToLower(strings.TrimSpace(encoding))
switch enc {
case "":
if text == "" {
return nil, nil
}
return []byte(text), nil
case "base64":
if text == "" {
return nil, nil
}
decoded, err := base64.StdEncoding.DecodeString(text)
if err != nil {
return nil, fmt.Errorf("invalid base64 body: %w", err)
}
return decoded, nil
default:
return nil, fmt.Errorf("unsupported body encoding %q", encoding)
}
}
func sanitizeHARProto(v string) string {
v = strings.TrimSpace(v)
if v == "" || strings.ContainsAny(v, " \t\r\n\x00") || !strings.HasPrefix(strings.ToUpper(v), "HTTP/") {
return "HTTP/1.1"
}
return v
}
// insertHARExchange inserts one HAR exchange idempotently (INSERT OR IGNORE)
// through the same spill/blob_refs/FTS path as the recorder and ImportArchive.
// It never writes to exchange_rewrites.
func (t *Traffic) insertHARExchange(ex parsedHARExchange) (bool, error) {
t.wmu.Lock()
defer t.wmu.Unlock()
var exists int
if err := t.db.QueryRow(`SELECT 1 FROM exchanges WHERE id=?`, ex.id).Scan(&exists); err == nil {
return false, nil
} else if !errors.Is(err, sql.ErrNoRows) {
return false, err
}
reqB := t.spill(ex.reqBody, ex.reqCT)
respB := t.spill(ex.respBody, ex.respCT)
tx, err := t.db.Begin()
if err != nil {
return false, fmt.Errorf("beginning transaction: %w", err)
}
defer tx.Rollback() //nolint:errcheck
res, err := tx.Exec(`INSERT OR IGNORE INTO exchanges(id,ts,host,method,url_template,url,status,content_type,req_len,resp_len,path)
VALUES(?,?,?,?,?,?,?,?,?,?,'')`,
ex.id, ex.ts, ex.host, ex.method, ex.urlTemplate, ex.rawURL, ex.status, ex.respCT, len(ex.reqBody), len(ex.respBody))
if err != nil {
return false, fmt.Errorf("writing index: %w", err)
}
n, _ := res.RowsAffected()
if n == 0 {
return false, nil
}
var rowid int64
if err := tx.QueryRow(`SELECT rowid FROM exchanges WHERE id=?`, ex.id).Scan(&rowid); err != nil {
return false, fmt.Errorf("fetching rowid: %w", err)
}
if _, err := tx.Exec(`INSERT INTO exchange_bodies(id,req_head,req_body,req_blob,resp_head,resp_body,resp_blob)
VALUES(?,?,?,?,?,?,?)`,
ex.id, ex.reqHead, reqB.inline, nullIfEmpty(reqB.hash), ex.respHead, respB.inline, nullIfEmpty(respB.hash)); err != nil {
return false, fmt.Errorf("writing body: %w", err)
}
for _, h := range []string{reqB.hash, respB.hash} {
if h == "" {
continue
}
if _, err := tx.Exec(`INSERT OR IGNORE INTO blob_refs(hash,exchange_id) VALUES(?,?)`, h, ex.id); err != nil {
return false, fmt.Errorf("registering blob reference: %w", err)
}
}
if t.fts {
idx := strings.Join([]string{ex.rawURL, ex.reqHead, reqB.index, ex.respHead, respB.index}, "\n")
if _, err := tx.Exec(`INSERT INTO ex_fts(rowid,content) VALUES(?,?)`, rowid, idx); err != nil {
return false, fmt.Errorf("writing full-text index: %w", err)
}
}
if err := tx.Commit(); err != nil {
return false, fmt.Errorf("committing: %w", err)
}
return true, nil
}
// ExportHAR exports exchanges matching f as a valid HAR 1.2 document in stable
// chronological order (ts ASC, id ASC), capped by opts.MaxEntries,
// opts.MaxTotalBytes (raw headers + bodies), and opts.MaxJSONBytes (marshaled
// JSON size). Truncation is reported in both HARExportMeta and the HAR 1.2
// custom fields _truncated / _truncatedReason / _totalMatched.
func (t *Traffic) ExportHAR(f PageQuery, opts HARExportOptions) (HAR, HARExportMeta, error) {
if t == nil {
return HAR{}, HARExportMeta{}, errors.New("traffic store is not available")
}
if opts.MaxEntries <= 0 {
opts.MaxEntries = DefaultHARMaxExportEntries
}
if opts.MaxTotalBytes <= 0 {
opts.MaxTotalBytes = DefaultHARMaxExportBytes
}
if opts.MaxJSONBytes <= 0 {
opts.MaxJSONBytes = DefaultHARMaxExportJSONBytes
}
creatorVer := strings.TrimSpace(opts.CreatorVersion)
if creatorVer == "" {
creatorVer = "dev"
}
where, args := t.pageWhere(f)
var totalMatched int
if err := t.db.QueryRow(`SELECT COUNT(*) FROM exchanges`+where, args...).Scan(&totalMatched); err != nil {
return HAR{}, HARExportMeta{}, err
}
type exRow struct {
id string
ts int64
host string
method string
rawURL string
status int
contentType string
legacyPath string
}
// Fetch at most MaxEntries+1 rows so we know whether the entry cap fired.
sel := `SELECT id,ts,host,method,url,status,content_type,path FROM exchanges` +
where + ` ORDER BY ts ASC, id ASC LIMIT ?`
qargs := append(append([]any{}, args...), opts.MaxEntries+1)
rows, err := t.db.Query(sel, qargs...)
if err != nil {
return HAR{}, HARExportMeta{}, err
}
var matched []exRow
for rows.Next() {
var r exRow
if err := rows.Scan(&r.id, &r.ts, &r.host, &r.method, &r.rawURL, &r.status, &r.contentType, &r.legacyPath); err != nil {
rows.Close()
return HAR{}, HARExportMeta{}, err
}
matched = append(matched, r)
}
if err := rows.Err(); err != nil {
rows.Close()
return HAR{}, HARExportMeta{}, err
}
rows.Close()
meta := HARExportMeta{TotalMatched: totalMatched}
hitEntryCap := len(matched) > opts.MaxEntries
if hitEntryCap {
matched = matched[:opts.MaxEntries]
}
// Reserve envelope overhead (~512 bytes) for the outer log object + metadata.
const harEnvelopeMargin int64 = 512
var encodedJSONBytes int64 = harEnvelopeMargin
entries := make([]HAREntry, 0, len(matched))
for _, r := range matched {
reqHead, reqBytes, respHead, respBytes, err := t.loadFullExchange(r.id, r.legacyPath)
if err != nil {
return HAR{}, HARExportMeta{}, err
}
entryRawBytes := int64(len(reqHead) + len(reqBytes) + len(respHead) + len(respBytes))
if meta.TotalBytes+entryRawBytes > opts.MaxTotalBytes {
meta.Truncated = true
meta.TruncatedReason = "max_bytes"
break
}
reqProto, reqHeaders := parseStoredRequestHead(reqHead)
respProto, respStatusText, respHeaders := parseStoredResponseHead(respHead)
harReq := HARRequest{
Method: r.method,
URL: redactURLUserinfo(r.rawURL),
HTTPVersion: reqProto,
Cookies: []HARCookie{},
Headers: reqHeaders,
QueryString: []HARParam{},
HeadersSize: -1,
BodySize: len(reqBytes),
}
if len(reqBytes) > 0 {
text, enc := encodeHARBody(reqBytes)
harReq.PostData = &HARPostData{
MimeType: findHeaderValue(reqHeaders, "Content-Type"),
Text: text,
Encoding: enc,
}
}
harContent := HARContent{
Size: len(respBytes),
MimeType: r.contentType,
}
if len(respBytes) > 0 {
harContent.Text, harContent.Encoding = encodeHARBody(respBytes)
}
harResp := HARResponse{
Status: r.status,
StatusText: respStatusText,
HTTPVersion: respProto,
Cookies: []HARCookie{},
Headers: respHeaders,
Content: harContent,
RedirectURL: "",
HeadersSize: -1,
BodySize: len(respBytes),
}
entry := HAREntry{
StartedDateTime: time.Unix(r.ts, 0).UTC().Format(time.RFC3339),
Time: 0,
Request: harReq,
Response: harResp,
Cache: HARCache{},
Timings: HARTimings{Send: -1, Wait: -1, Receive: -1},
ArtexExchangeID: r.id,
}
entryJSON, err := json.Marshal(entry)
if err != nil {
return HAR{}, HARExportMeta{}, err
}
// +1 for the separating comma inside the JSON array.
if encodedJSONBytes+int64(len(entryJSON))+1 > opts.MaxJSONBytes {
meta.Truncated = true
meta.TruncatedReason = "max_bytes"
break
}
encodedJSONBytes += int64(len(entryJSON)) + 1
meta.TotalBytes += entryRawBytes
entries = append(entries, entry)
}
if !meta.Truncated && hitEntryCap {
meta.Truncated = true
meta.TruncatedReason = "max_entries"
}
meta.Exported = len(entries)
doc := HAR{
Log: &HARLog{
Version: HARVersion,
Creator: HARCreator{Name: "ARTEX", Version: creatorVer},
Entries: entries,
Truncated: meta.Truncated,
TruncatedReason: meta.TruncatedReason,
},
}
if meta.Truncated {
doc.Log.TotalMatched = meta.TotalMatched
}
return doc, meta, nil
}
func (t *Traffic) loadFullExchange(id, legacyPath string) (reqHead string, reqBody []byte, respHead string, respBody []byte, err error) {
var reqBlob, respBlob sql.NullString
err = t.db.QueryRow(`SELECT req_head,req_body,req_blob,resp_head,resp_body,resp_blob
FROM exchange_bodies WHERE id=?`, id).Scan(&reqHead, &reqBody, &reqBlob, &respHead, &respBody, &respBlob)
if errors.Is(err, sql.ErrNoRows) && strings.TrimSpace(legacyPath) != "" {
rb, readErr := os.ReadFile(filepath.Join(t.dir, legacyPath, "request.http"))
if readErr != nil && !os.IsNotExist(readErr) {
return "", nil, "", nil, readErr
}
pb, readErr := os.ReadFile(filepath.Join(t.dir, legacyPath, "response.http"))
if readErr != nil && !os.IsNotExist(readErr) {
return "", nil, "", nil, readErr
}
reqHead, reqBody = splitLegacyHTTP(rb)
respHead, respBody = splitLegacyHTTP(pb)
return reqHead, reqBody, respHead, respBody, nil
}
if err != nil {
return "", nil, "", nil, err
}
if reqBlob.Valid && reqBlob.String != "" {
p, err := t.blobPath(reqBlob.String)
if err != nil {
return "", nil, "", nil, err
}
reqBody, err = os.ReadFile(p)
if err != nil {
return "", nil, "", nil, err
}
}
if respBlob.Valid && respBlob.String != "" {
p, err := t.blobPath(respBlob.String)
if err != nil {
return "", nil, "", nil, err
}
respBody, err = os.ReadFile(p)
if err != nil {
return "", nil, "", nil, err
}
}
return reqHead, reqBody, respHead, respBody, nil
}
func splitLegacyHTTP(raw []byte) (string, []byte) {
if idx := bytes.Index(raw, []byte("\r\n\r\n")); idx >= 0 {
return string(raw[:idx]), raw[idx+4:]
}
if idx := bytes.Index(raw, []byte("\n\n")); idx >= 0 {
return string(raw[:idx]), raw[idx+2:]
}
return string(raw), nil
}
// encodeHARBody returns plain text for valid NUL-free UTF-8 bodies, and base64
// otherwise so binary payloads survive export and re-import bit-for-bit.
func encodeHARBody(body []byte) (text, encoding string) {
if utf8.Valid(body) && bytes.IndexByte(body, 0) < 0 {
return string(body), ""
}
return base64.StdEncoding.EncodeToString(body), "base64"
}
func parseStoredRequestHead(raw string) (httpVersion string, headers []HARHeader) {
headers = []HARHeader{}
lines := strings.Split(strings.ReplaceAll(raw, "\r\n", "\n"), "\n")
if len(lines) == 0 {
return "", headers
}
first := strings.TrimSpace(lines[0])
if parts := strings.SplitN(first, " ", 3); len(parts) == 3 && strings.HasPrefix(strings.ToUpper(parts[2]), "HTTP/") {
httpVersion = strings.TrimSpace(parts[2])
}
headers = appendParsedHeaders(headers, lines[1:])
return httpVersion, headers
}
func parseStoredResponseHead(raw string) (httpVersion, statusText string, headers []HARHeader) {
headers = []HARHeader{}
lines := strings.Split(strings.ReplaceAll(raw, "\r\n", "\n"), "\n")
if len(lines) == 0 {
return "", "", headers
}
first := strings.TrimSpace(lines[0])
if parts := strings.SplitN(first, " ", 3); len(parts) >= 2 && strings.HasPrefix(strings.ToUpper(parts[0]), "HTTP") {
if strings.HasPrefix(strings.ToUpper(parts[0]), "HTTP/") {
httpVersion = strings.TrimSpace(parts[0])
}
if len(parts) == 3 {
statusText = strings.TrimSpace(parts[2])
}
}
headers = appendParsedHeaders(headers, lines[1:])
return httpVersion, statusText, headers
}
func appendParsedHeaders(dst []HARHeader, lines []string) []HARHeader {
for _, line := range lines {
line = strings.TrimRight(line, "\r")
if strings.TrimSpace(line) == "" {
continue
}
colon := strings.IndexByte(line, ':')
if colon <= 0 {
continue
}
name := strings.TrimSpace(line[:colon])
if name == "" || !validHeaderName(name) {
continue
}
val := strings.TrimSpace(line[colon+1:])
dst = append(dst, HARHeader{Name: name, Value: val})
}
return dst
}
func findHeaderValue(headers []HARHeader, name string) string {
for _, h := range headers {
if strings.EqualFold(h.Name, name) {
return h.Value
}
}
return ""
}
+817
View File
@@ -0,0 +1,817 @@
package traffic
import (
"bytes"
"encoding/base64"
"encoding/json"
"errors"
"strings"
"testing"
"github.com/Autumn-27/artex/traffic/rewrite"
)
func TestHARRoundTripPlainBase64BinaryEmptyAndLargeBlob(t *testing.T) {
src, err := Open(t.TempDir(), "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
defer src.Close()
// Install an enabled capture rewrite rule to prove HAR import NEVER runs rules
// or writes exchange_rewrites rows.
set := rewrite.Compile([]rewrite.Rule{{
ID: 42, Name: "should-not-apply-to-har", Enabled: true, Phase: rewrite.PhaseRequest,
Actions: []rewrite.Action{{Type: rewrite.ActionSetHeader, Name: "X-Injected", Value: "1"}},
}}, rewrite.Options{})
if len(set.LoadErrors()) != 0 {
t.Fatalf("compile rewrite rules: %v", set.LoadErrors())
}
src.SetRewriteRules(set)
binaryResp := []byte{0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a, 0x00, 0xff, 0xfe, 0x01}
binaryReq := []byte{0x00, 0x01, 0x02, 0x80, 0xff}
largeTextBody := "LARGE-HAR-NEEDLE-" + strings.Repeat("abcdefghij", (maxInlineBody/10)+50) // > 256 KiB -> blob!
inputHAR := HAR{
Log: &HARLog{
Version: "1.2",
Creator: HARCreator{Name: "TestBrowser", Version: "1.0"},
Entries: []HAREntry{
{
StartedDateTime: "2026-05-01T10:00:00.789Z",
Request: HARRequest{
Method: "POST",
URL: "https://app.example.com/api/v1/items/42?mode=full",
HTTPVersion: "HTTP/1.1",
Headers: []HARHeader{
{Name: ":method", Value: "POST"},
{Name: ":authority", Value: "app.example.com"},
{Name: "Host", Value: "app.example.com"},
{Name: "Content-Type", Value: "application/json"},
{Name: "X-Trace", Value: "one"},
{Name: "X-Trace", Value: "two"},
},
PostData: &HARPostData{
MimeType: "application/json",
Text: `{"action":"create","needle":"plain-req-token"}`,
},
},
Response: HARResponse{
Status: 201,
StatusText: "Created",
HTTPVersion: "HTTP/1.1",
Headers: []HARHeader{
{Name: "Content-Type", Value: "application/json; charset=utf-8"},
{Name: "X-Resp", Value: "ok"},
},
Content: HARContent{
MimeType: "application/json; charset=utf-8",
Text: `{"id":42,"msg":"plain-resp-marker"}`,
},
},
},
{
StartedDateTime: "2026-05-01T10:00:05Z",
Request: HARRequest{
Method: "POST",
URL: "https://cdn.example.com/upload",
HTTPVersion: "HTTP/2.0",
Headers: []HARHeader{{Name: "Content-Type", Value: "application/octet-stream"}},
PostData: &HARPostData{
MimeType: "application/octet-stream",
Text: base64.StdEncoding.EncodeToString(binaryReq),
Encoding: "base64",
},
},
Response: HARResponse{
Status: 200,
HTTPVersion: "HTTP/2.0",
Headers: []HARHeader{{Name: "Content-Type", Value: "image/png"}},
Content: HARContent{
MimeType: "image/png",
Text: base64.StdEncoding.EncodeToString(binaryResp),
Encoding: "base64",
},
},
},
{
StartedDateTime: "2026-05-01T10:00:10Z",
Request: HARRequest{
Method: "GET",
URL: "https://app.example.com/empty",
Headers: []HARHeader{{Name: "Accept", Value: "*/*"}},
},
Response: HARResponse{
Status: 204,
Headers: []HARHeader{},
Content: HARContent{},
},
},
{
StartedDateTime: "2026-05-01T10:00:15Z",
Request: HARRequest{
Method: "GET",
URL: "https://blob.example.com/big.txt",
Headers: []HARHeader{},
},
Response: HARResponse{
Status: 200,
Headers: []HARHeader{{Name: "Content-Type", Value: "text/plain"}},
Content: HARContent{
MimeType: "text/plain",
Text: largeTextBody,
},
},
},
},
},
}
rawIn, err := json.Marshal(inputHAR)
if err != nil {
t.Fatal(err)
}
res, err := src.ImportHAR(rawIn, HARImportOptions{})
if err != nil {
t.Fatalf("ImportHAR: %v", err)
}
if res.Imported != 4 || res.Skipped != 0 || res.Failed != 0 || res.Truncated {
t.Fatalf("ImportHAR result = %+v, want 4/0/0/false", res)
}
// Re-importing the exact same HAR must be idempotent (all skipped).
res2, err := src.ImportHAR(rawIn, HARImportOptions{})
if err != nil {
t.Fatalf("re-import HAR: %v", err)
}
if res2.Imported != 0 || res2.Skipped != 4 || res2.Failed != 0 {
t.Fatalf("re-import result = %+v, want 0/4/0", res2)
}
// Verify none of the imported exchanges are flagged rewritten or have exchange_rewrites rows.
var rwRows int
if err := src.DB().QueryRow(`SELECT COUNT(*) FROM exchange_rewrites`).Scan(&rwRows); err != nil || rwRows != 0 {
t.Fatalf("exchange_rewrites count = %d err=%v, want 0", rwRows, err)
}
allRows, total, err := src.Page(PageQuery{RespMin: -1, RespMax: -1}, 0, 20)
if err != nil || total != 4 || len(allRows) != 4 {
t.Fatalf("Page after import = (%d, %d, %v), want (4, 4, nil)", len(allRows), total, err)
}
for _, r := range allRows {
if !strings.HasPrefix(r.ID, "har-") {
t.Fatalf("expected har- prefix on id %q", r.ID)
}
if r.Rewritten {
t.Fatalf("imported exchange %s unexpectedly flagged rewritten", r.ID)
}
if rw, ids, err := src.RewriteInfo(r.ID); err != nil || rw || len(ids) != 0 {
t.Fatalf("RewriteInfo(%s) = (%v, %v, %v), want (false, nil, nil)", r.ID, rw, ids, err)
}
}
// Verify searchable via FTS (both inline body and spilled large text body).
if rows, tot, err := src.Page(PageQuery{Body: "plain-resp-marker", RespMin: -1, RespMax: -1}, 0, 10); err != nil || tot != 1 || len(rows) != 1 {
t.Fatalf("FTS search inline body = (%d, %d, %v), want 1", len(rows), tot, err)
}
if rows, tot, err := src.Page(PageQuery{Body: "LARGE-HAR-NEEDLE", RespMin: -1, RespMax: -1}, 0, 10); err != nil || tot != 1 || len(rows) != 1 {
t.Fatalf("FTS search blob body = (%d, %d, %v), want 1", len(rows), tot, err)
}
// Verify large body actually spilled to a blob and registered in blob_refs.
var blobHash string
var blobRefCount int
if err := src.DB().QueryRow(`SELECT resp_blob FROM exchange_bodies b JOIN exchanges e ON e.id=b.id WHERE e.host='blob.example.com'`).Scan(&blobHash); err != nil || blobHash == "" {
t.Fatalf("large body did not spill to blob: hash=%q err=%v", blobHash, err)
}
if err := src.DB().QueryRow(`SELECT COUNT(*) FROM blob_refs WHERE hash=?`, blobHash).Scan(&blobRefCount); err != nil || blobRefCount != 1 {
t.Fatalf("blob_refs count for %s = %d err=%v, want 1", blobHash, blobRefCount, err)
}
// Export HAR from src.
exported, meta, err := src.ExportHAR(PageQuery{RespMin: -1, RespMax: -1}, HARExportOptions{CreatorVersion: "2.3.0"})
if err != nil {
t.Fatalf("ExportHAR: %v", err)
}
if meta.Exported != 4 || meta.TotalMatched != 4 || meta.Truncated || meta.TruncatedReason != "" {
t.Fatalf("ExportHAR meta = %+v", meta)
}
if exported.Log.Version != "1.2" || exported.Log.Creator.Name != "ARTEX" || exported.Log.Creator.Version != "2.3.0" {
t.Fatalf("ExportHAR header = %+v", exported.Log)
}
if len(exported.Log.Entries) != 4 {
t.Fatalf("ExportHAR entries = %d, want 4", len(exported.Log.Entries))
}
// Verify HAR 1.2 required keys are present with explicit unknown placeholders
// and serverIPAddress is not fabricated.
rawExported, err := json.Marshal(exported)
if err != nil {
t.Fatal(err)
}
if strings.Contains(string(rawExported), `"serverIPAddress"`) {
t.Fatalf("exported HAR invented serverIPAddress: %s", string(rawExported))
}
var rawDoc map[string]any
if err := json.Unmarshal(rawExported, &rawDoc); err != nil {
t.Fatal(err)
}
rawLog := rawDoc["log"].(map[string]any)
rawEntries := rawLog["entries"].([]any)
firstEntry := rawEntries[0].(map[string]any)
if got, ok := firstEntry["time"].(float64); !ok || got != 0 {
t.Fatalf("entry.time = %#v, want 0", firstEntry["time"])
}
if cacheObj, ok := firstEntry["cache"].(map[string]any); !ok || len(cacheObj) != 0 {
t.Fatalf("entry.cache = %#v, want {}", firstEntry["cache"])
}
timingsObj, ok := firstEntry["timings"].(map[string]any)
if !ok || timingsObj["send"] != float64(-1) || timingsObj["wait"] != float64(-1) || timingsObj["receive"] != float64(-1) {
t.Fatalf("entry.timings = %#v, want {send:-1 wait:-1 receive:-1}", firstEntry["timings"])
}
reqObj := firstEntry["request"].(map[string]any)
if cookies, ok := reqObj["cookies"].([]any); !ok || len(cookies) != 0 {
t.Fatalf("request.cookies = %#v, want []", reqObj["cookies"])
}
if qs, ok := reqObj["queryString"].([]any); !ok || len(qs) != 0 {
t.Fatalf("request.queryString = %#v, want []", reqObj["queryString"])
}
if hs, ok := reqObj["headersSize"].(float64); !ok || hs != -1 {
t.Fatalf("request.headersSize = %#v, want -1", reqObj["headersSize"])
}
if _, ok := reqObj["httpVersion"].(string); !ok {
t.Fatalf("request.httpVersion missing: %#v", reqObj)
}
respObj := firstEntry["response"].(map[string]any)
if cookies, ok := respObj["cookies"].([]any); !ok || len(cookies) != 0 {
t.Fatalf("response.cookies = %#v, want []", respObj["cookies"])
}
if redir, ok := respObj["redirectURL"].(string); !ok || redir != "" {
t.Fatalf("response.redirectURL = %#v, want \"\"", respObj["redirectURL"])
}
if hs, ok := respObj["headersSize"].(float64); !ok || hs != -1 {
t.Fatalf("response.headersSize = %#v, want -1", respObj["headersSize"])
}
if _, ok := respObj["httpVersion"].(string); !ok {
t.Fatalf("response.httpVersion missing: %#v", respObj)
}
if _, ok := respObj["statusText"].(string); !ok {
t.Fatalf("response.statusText missing: %#v", respObj)
}
contentObj := respObj["content"].(map[string]any)
if _, ok := contentObj["mimeType"].(string); !ok {
t.Fatalf("response.content.mimeType missing: %#v", contentObj)
}
// Verify stable chronological ordering & exact body encodings.
e0 := exported.Log.Entries[0]
if e0.Request.URL != "https://app.example.com/api/v1/items/42?mode=full" || e0.Response.Content.Encoding != "" || e0.Response.Content.Text != `{"id":42,"msg":"plain-resp-marker"}` {
t.Fatalf("entry 0 mismatch: %+v", e0)
}
e1 := exported.Log.Entries[1]
if e1.Response.Content.Encoding != "base64" || e1.Request.PostData == nil || e1.Request.PostData.Encoding != "base64" {
t.Fatalf("entry 1 binary encodings wrong: req=%+v resp=%+v", e1.Request.PostData, e1.Response.Content)
}
gotBinResp, err := base64.StdEncoding.DecodeString(e1.Response.Content.Text)
if err != nil || !bytes.Equal(gotBinResp, binaryResp) {
t.Fatalf("entry 1 binary resp mismatch: got=%x err=%v", gotBinResp, err)
}
gotBinReq, err := base64.StdEncoding.DecodeString(e1.Request.PostData.Text)
if err != nil || !bytes.Equal(gotBinReq, binaryReq) {
t.Fatalf("entry 1 binary req mismatch: got=%x err=%v", gotBinReq, err)
}
e2 := exported.Log.Entries[2]
if e2.Request.PostData != nil || e2.Response.Content.Size != 0 || e2.Response.Content.Text != "" {
t.Fatalf("entry 2 empty body mismatch: %+v", e2)
}
e3 := exported.Log.Entries[3]
if e3.Response.Content.Size != len(largeTextBody) || e3.Response.Content.Text != largeTextBody || e3.Response.Content.Encoding != "" {
t.Fatalf("entry 3 large blob body truncated or altered: size=%d len=%d", e3.Response.Content.Size, len(e3.Response.Content.Text))
}
// Re-importing the exported HAR into src must be 100% idempotent (same deterministic IDs!).
reimpSame, err := src.ImportHAR(rawExported, HARImportOptions{})
if err != nil {
t.Fatalf("re-import exported HAR into src: %v", err)
}
if reimpSame.Imported != 0 || reimpSame.Skipped != 4 || reimpSame.Failed != 0 {
t.Fatalf("re-import exported HAR into src = %+v, want 0/4/0", reimpSame)
}
// Re-importing the exported HAR into a fresh store must yield identical IDs, metadata, and bodies.
dst, err := Open(t.TempDir(), "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
defer dst.Close()
reimpFresh, err := dst.ImportHAR(rawExported, HARImportOptions{})
if err != nil {
t.Fatalf("import exported HAR into dst: %v", err)
}
if reimpFresh.Imported != 4 || reimpFresh.Skipped != 0 || reimpFresh.Failed != 0 {
t.Fatalf("dst import = %+v, want 4/0/0", reimpFresh)
}
exportedDst, _, err := dst.ExportHAR(PageQuery{RespMin: -1, RespMax: -1}, HARExportOptions{CreatorVersion: "2.3.0"})
if err != nil {
t.Fatal(err)
}
rawDst, err := json.Marshal(exportedDst)
if err != nil {
t.Fatal(err)
}
if !bytes.Equal(rawExported, rawDst) {
t.Fatalf("dst round-trip HAR differs from src export")
}
}
func TestHARImportValidationDuplicatesAndPartialErrors(t *testing.T) {
tr, err := Open(t.TempDir(), "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
defer tr.Close()
badDocs := []struct {
name string
raw string
want string
}{
{"empty", "", "empty HAR"},
{"malformed json", `{"log":`, "malformed JSON"},
{"trailing garbage", `{"log":{"version":"1.2","entries":[]}} extra`, "trailing content"},
{"missing log", `{}`, "missing log"},
{"missing entries", `{"log":{"version":"1.2"}}`, "missing log.entries"},
{"null entries", `{"log":{"version":"1.2","entries":null}}`, "missing log.entries"},
{"unsupported version 1.1", `{"log":{"version":"1.1","entries":[]}}`, "unsupported HAR version"},
{"unsupported version 2.0", `{"log":{"version":"2.0","entries":[]}}`, "unsupported HAR version"},
{"empty version", `{"log":{"version":"","entries":[]}}`, "unsupported HAR version"},
}
for _, tc := range badDocs {
t.Run(tc.name, func(t *testing.T) {
_, err := tr.ImportHAR([]byte(tc.raw), HARImportOptions{})
if !errors.Is(err, ErrInvalidHAR) || !strings.Contains(err.Error(), tc.want) {
t.Fatalf("ImportHAR(%s) err = %v, want ErrInvalidHAR containing %q", tc.name, err, tc.want)
}
})
}
// Duplicate entries inside one HAR + bad entries mixed with valid ones.
mixed := `{
"log": {
"version": "1.2",
"entries": [
{
"startedDateTime": "2026-05-01T12:00:00Z",
"request": {"method": "GET", "url": "https://dup.example/a", "headers": []},
"response": {"status": 200, "headers": [], "content": {"text": "ok"}}
},
{
"startedDateTime": "2026-05-01T12:00:00Z",
"request": {"method": "GET", "url": "https://dup.example/a", "headers": []},
"response": {"status": 200, "headers": [], "content": {"text": "ok"}}
},
{
"startedDateTime": "not-a-timestamp",
"request": {"method": "GET", "url": "https://bad.example/ts", "headers": []},
"response": {"status": 200, "headers": [], "content": {}}
},
{
"startedDateTime": "2026-05-01T12:00:01Z",
"request": {"method": "GET", "url": "https://bad.example/hdr", "headers": [{"name": "X-Bad", "value": "a\r\nb"}]},
"response": {"status": 200, "headers": [], "content": {}}
},
{
"startedDateTime": "2026-05-01T12:00:02Z",
"request": {"method": "GET", "url": "https://bad.example/b64", "headers": []},
"response": {"status": 200, "headers": [], "content": {"text": "!!!not-base64!!!", "encoding": "base64"}}
},
{
"startedDateTime": "2026-05-01T12:00:03Z",
"request": {"method": "GET", "url": "https://good.example/b", "headers": []},
"response": {"status": 200, "headers": [], "content": {"text": "second"}}
}
]
}
}`
res, err := tr.ImportHAR([]byte(mixed), HARImportOptions{})
if err != nil {
t.Fatalf("mixed ImportHAR: %v", err)
}
if res.Imported != 2 || res.Skipped != 1 || res.Failed != 3 || len(res.Errors) != 3 {
t.Fatalf("mixed ImportHAR result = %+v, want imported=2 skipped=1 failed=3 errors=3", res)
}
// Entry cap returns ErrInvalidHAR.
if _, err := tr.ImportHAR([]byte(mixed), HARImportOptions{MaxEntries: 2}); !errors.Is(err, ErrInvalidHAR) {
t.Fatalf("MaxEntries exceeded err = %v, want ErrInvalidHAR", err)
}
// Byte cap truncates cleanly and reports remaining skipped.
byteCapHAR := `{
"log": {
"version": "1.2",
"entries": [
{
"startedDateTime": "2026-05-01T12:01:00Z",
"request": {"method": "GET", "url": "https://cap.example/1", "headers": []},
"response": {"status": 200, "headers": [], "content": {"text": "1234567890"}}
},
{
"startedDateTime": "2026-05-01T12:01:01Z",
"request": {"method": "GET", "url": "https://cap.example/2", "headers": []},
"response": {"status": 200, "headers": [], "content": {"text": "1234567890"}}
}
]
}
}`
capRes, err := tr.ImportHAR([]byte(byteCapHAR), HARImportOptions{MaxTotalBytes: 15})
if err != nil {
t.Fatalf("byte-capped ImportHAR: %v", err)
}
if capRes.Imported != 1 || capRes.Skipped != 1 || !capRes.Truncated || len(capRes.Errors) != 1 {
t.Fatalf("byte-capped ImportHAR = %+v, want imported=1 skipped=1 truncated=true", capRes)
}
}
func TestHARExportFiltersAndTruncationCaps(t *testing.T) {
tr, err := Open(t.TempDir(), "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
defer tr.Close()
insert := func(id string, ts int64, host string, status int, body string) {
t.Helper()
if _, err := tr.DB().Exec(`INSERT INTO exchanges(id,ts,host,method,url_template,url,status,content_type,req_len,resp_len,path)
VALUES(?,?,?,?,?,?,?,?,?,?,?)`, id, ts, host, "GET", "/"+id, "https://"+host+"/"+id, status, "text/plain", 0, len(body), ""); err != nil {
t.Fatal(err)
}
if _, err := tr.DB().Exec(`INSERT INTO exchange_bodies(id,req_head,req_body,resp_head,resp_body)
VALUES(?,?,?,?,?)`, id, "GET /"+id+" HTTP/1.1\nHost: "+host, []byte{}, "HTTP/1.1 200 OK", []byte(body)); err != nil {
t.Fatal(err)
}
}
insert("e-2", 20, "alpha.example", 200, "bbbbbbbbbb")
insert("e-1", 10, "alpha.example", 404, "aaaaaaaaaa")
insert("e-3", 30, "beta.example", 200, "cccccccccc")
// Filter by host + status, verify chronological order.
doc, meta, err := tr.ExportHAR(PageQuery{Host: "alpha.example", RespMin: -1, RespMax: -1}, HARExportOptions{})
if err != nil {
t.Fatal(err)
}
if meta.Exported != 2 || meta.TotalMatched != 2 || meta.Truncated {
t.Fatalf("filtered export meta = %+v", meta)
}
if doc.Log.Entries[0].Request.URL != "https://alpha.example/e-1" || doc.Log.Entries[1].Request.URL != "https://alpha.example/e-2" {
t.Fatalf("export not in stable (ts,id) order: %+v", doc.Log.Entries)
}
// Truncation by MaxEntries.
docEntryCap, metaEntryCap, err := tr.ExportHAR(PageQuery{RespMin: -1, RespMax: -1}, HARExportOptions{MaxEntries: 2})
if err != nil {
t.Fatal(err)
}
if !metaEntryCap.Truncated || metaEntryCap.TruncatedReason != "max_entries" || metaEntryCap.Exported != 2 || metaEntryCap.TotalMatched != 3 {
t.Fatalf("entry-capped meta = %+v", metaEntryCap)
}
if !docEntryCap.Log.Truncated || docEntryCap.Log.TruncatedReason != "max_entries" || docEntryCap.Log.TotalMatched != 3 {
t.Fatalf("entry-capped log fields = %+v", docEntryCap.Log)
}
// Truncation by MaxTotalBytes (note: raw cap counts both headers and bodies).
// Each row here has req_head (~46B) + resp_head (15B) + body (10B) ≈ 71B.
docByteCap, metaByteCap, err := tr.ExportHAR(PageQuery{RespMin: -1, RespMax: -1}, HARExportOptions{MaxTotalBytes: 100})
if err != nil {
t.Fatal(err)
}
if !metaByteCap.Truncated || metaByteCap.TruncatedReason != "max_bytes" || metaByteCap.Exported != 1 || metaByteCap.TotalMatched != 3 {
t.Fatalf("byte-capped meta = %+v", metaByteCap)
}
if !docByteCap.Log.Truncated || docByteCap.Log.TruncatedReason != "max_bytes" || docByteCap.Log.TotalMatched != 3 {
t.Fatalf("byte-capped log fields = %+v", docByteCap.Log)
}
}
func TestHARTwoIdenticalExchangesRoundTripPreservesBoth(t *testing.T) {
src, err := Open(t.TempDir(), "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
defer src.Close()
// Two distinct ARTEX exchanges recorded in the same second with identical
// method, URL, status, headers, and bodies.
for _, id := range []string{"dup-a", "dup-b"} {
if _, err := src.DB().Exec(`INSERT INTO exchanges(id,ts,host,method,url_template,url,status,content_type,req_len,resp_len,path)
VALUES(?,1700000000,'same.example','POST','/ping','https://same.example/ping',200,'application/json',2,2,'')`, id); err != nil {
t.Fatal(err)
}
if _, err := src.DB().Exec(`INSERT INTO exchange_bodies(id,req_head,req_body,resp_head,resp_body)
VALUES(?,'POST /ping HTTP/1.1\nHost: same.example\n',?,'HTTP/1.1 200 OK\n',?)`, id, []byte("{}"), []byte("{}")); err != nil {
t.Fatal(err)
}
}
exported, meta, err := src.ExportHAR(PageQuery{RespMin: -1, RespMax: -1}, HARExportOptions{})
if err != nil {
t.Fatal(err)
}
if meta.Exported != 2 || len(exported.Log.Entries) != 2 {
t.Fatalf("export = %+v, want 2 entries", meta)
}
if exported.Log.Entries[0].ArtexExchangeID != "dup-a" || exported.Log.Entries[1].ArtexExchangeID != "dup-b" {
t.Fatalf("exported _artexExchangeId = [%q %q], want [dup-a dup-b]",
exported.Log.Entries[0].ArtexExchangeID, exported.Log.Entries[1].ArtexExchangeID)
}
raw, err := json.Marshal(exported)
if err != nil {
t.Fatal(err)
}
// Re-importing into src is idempotent (both skipped).
reimpSrc, err := src.ImportHAR(raw, HARImportOptions{})
if err != nil {
t.Fatal(err)
}
if reimpSrc.Imported != 0 || reimpSrc.Skipped != 2 || reimpSrc.Failed != 0 {
t.Fatalf("re-import into src = %+v, want 0/2/0", reimpSrc)
}
// Importing into a fresh store preserves BOTH dup-a and dup-b (no collapse!).
dst, err := Open(t.TempDir(), "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
defer dst.Close()
impDst, err := dst.ImportHAR(raw, HARImportOptions{})
if err != nil {
t.Fatal(err)
}
if impDst.Imported != 2 || impDst.Skipped != 0 || impDst.Failed != 0 {
t.Fatalf("import into fresh dst = %+v, want 2/0/0", impDst)
}
if count, err := dst.Count(); err != nil || count != 2 {
t.Fatalf("dst.Count() = (%d, %v), want 2", count, err)
}
if _, _, err := dst.Get("dup-a"); err != nil {
t.Fatalf("dup-a missing in dst: %v", err)
}
if _, _, err := dst.Get("dup-b"); err != nil {
t.Fatalf("dup-b missing in dst: %v", err)
}
// Re-importing into dst is idempotent.
reimpDst, err := dst.ImportHAR(raw, HARImportOptions{})
if err != nil {
t.Fatal(err)
}
if reimpDst.Imported != 0 || reimpDst.Skipped != 2 || reimpDst.Failed != 0 {
t.Fatalf("re-import into dst = %+v, want 0/2/0", reimpDst)
}
}
func TestHARExportBinaryNearCapFitsUnderImportLimit(t *testing.T) {
src, err := Open(t.TempDir(), "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
defer src.Close()
// Binary body near the 20 MiB raw export cap (~19.5 MiB). After base64
// expansion (4/3x ≈ 26 MiB) plus headers and JSON envelope, it must still
// marshal strictly under the 32 MiB HTTP import limit.
const bodySize = 19*1024*1024 + 512*1024
binBody := make([]byte, bodySize)
for i := range binBody {
binBody[i] = byte((i*31 + 0x80) & 0xff)
}
binBody[0] = 0x00 // guarantee binary/base64 encoding
respB := src.spill(binBody, "application/octet-stream")
reqHead := "GET /big.bin HTTP/1.1\nHost: cap.example\n" + strings.Repeat("X-Pad: 0123456789\n", 10)
respHead := "HTTP/1.1 200 OK\nContent-Type: application/octet-stream\n"
if err := src.insertExchange("bin-near-cap", 1700000100, "cap.example", "GET", "/big.bin",
"https://cap.example/big.bin", 200, "application/octet-stream", 0, len(binBody), reqHead, storedBody{}, respHead, respB); err != nil {
t.Fatal(err)
}
doc, meta, err := src.ExportHAR(PageQuery{RespMin: -1, RespMax: -1}, HARExportOptions{})
if err != nil {
t.Fatalf("ExportHAR near cap: %v", err)
}
if meta.Truncated || meta.Exported != 1 {
t.Fatalf("near-cap export meta = %+v, want Truncated=false Exported=1", meta)
}
// Verify headers were counted in TotalBytes alongside the body.
wantRaw := int64(len(reqHead) + len(respHead) + len(binBody))
if meta.TotalBytes != wantRaw {
t.Fatalf("meta.TotalBytes = %d, want %d (headers + body)", meta.TotalBytes, wantRaw)
}
rawJSON, err := json.Marshal(doc)
if err != nil {
t.Fatal(err)
}
const httpImportLimit = 32 << 20
if len(rawJSON) >= httpImportLimit {
t.Fatalf("marshaled near-cap binary HAR = %d bytes, exceeds 32 MiB import limit (%d)", len(rawJSON), httpImportLimit)
}
// Re-importing into a fresh store succeeds with exact bytes.
dst, err := Open(t.TempDir(), "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
defer dst.Close()
imp, err := dst.ImportHAR(rawJSON, HARImportOptions{})
if err != nil || imp.Imported != 1 || imp.Failed != 0 || imp.Truncated {
t.Fatalf("re-import near-cap binary HAR = (%+v, %v)", imp, err)
}
_, _, _, gotResp, err := dst.loadFullExchange("bin-near-cap", "")
if err != nil || !bytes.Equal(gotResp, binBody) {
t.Fatalf("near-cap binary round-trip mismatch: len=%d err=%v", len(gotResp), err)
}
}
func TestHARImportPostDataParamsAndCookies(t *testing.T) {
tr, err := Open(t.TempDir(), "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
defer tr.Close()
// Entry 0: form body ONLY in postData.params, cookies ONLY in cookies arrays.
// Entry 1: Cookie and Set-Cookie already present in headers -> must NOT duplicate.
harJSON := `{
"log": {
"version": "1.2",
"entries": [
{
"startedDateTime": "2026-05-02T09:00:00Z",
"request": {
"method": "POST",
"url": "https://form.example/login",
"headers": [{"name": "Host", "value": "form.example"}],
"cookies": [
{"name": "sid", "value": "abc123"},
{"name": "theme", "value": "dark"}
],
"postData": {
"mimeType": "application/x-www-form-urlencoded",
"params": [
{"name": "user", "value": "alice"},
{"name": "msg", "value": "hello world"}
]
}
},
"response": {
"status": 302,
"headers": [{"name": "Location", "value": "/home"}],
"cookies": [
{"name": "session", "value": "xyz", "path": "/", "httpOnly": true, "secure": true}
],
"content": {"size": 0, "mimeType": ""}
}
},
{
"startedDateTime": "2026-05-02T09:00:05Z",
"request": {
"method": "GET",
"url": "https://form.example/home",
"headers": [{"name": "Cookie", "value": "existing=1"}],
"cookies": [{"name": "existing", "value": "1"}]
},
"response": {
"status": 200,
"headers": [{"name": "Set-Cookie", "value": "kept=2; Path=/"}],
"cookies": [{"name": "kept", "value": "2", "path": "/"}],
"content": {"size": 2, "mimeType": "text/plain", "text": "ok"}
}
}
]
}
}`
res, err := tr.ImportHAR([]byte(harJSON), HARImportOptions{})
if err != nil || res.Imported != 2 || res.Failed != 0 {
t.Fatalf("ImportHAR(params+cookies) = (%+v, %v)", res, err)
}
rows, _, err := tr.Page(PageQuery{Host: "form.example", RespMin: -1, RespMax: -1, Sort: "ts", Order: "asc"}, 0, 10)
if err != nil || len(rows) != 2 {
t.Fatalf("Page = (%d, %v)", len(rows), err)
}
req0, resp0, err := tr.Get(rows[0].ID)
if err != nil {
t.Fatal(err)
}
if !strings.Contains(req0, "Cookie: sid=abc123; theme=dark\n") {
t.Fatalf("entry 0 missing synthesized Cookie header: %q", req0)
}
if !strings.Contains(req0, "user=alice&msg=hello+world") {
t.Fatalf("entry 0 missing postData.params form body: %q", req0)
}
if !strings.Contains(resp0, "Set-Cookie: session=xyz; Path=/; Secure; HttpOnly\n") {
t.Fatalf("entry 0 missing synthesized Set-Cookie header: %q", resp0)
}
req1, resp1, err := tr.Get(rows[1].ID)
if err != nil {
t.Fatal(err)
}
if strings.Count(req1, "Cookie:") != 1 {
t.Fatalf("entry 1 duplicated Cookie header: %q", req1)
}
if strings.Count(resp1, "Set-Cookie:") != 1 {
t.Fatalf("entry 1 duplicated Set-Cookie header: %q", resp1)
}
}
func TestHARImportStripsAndRedactsURLUserinfo(t *testing.T) {
tr, err := Open(t.TempDir(), "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
defer tr.Close()
harJSON := `{
"log": {
"version": "1.2",
"entries": [
{
"startedDateTime": "2026-05-03T10:00:00Z",
"request": {
"method": "GET",
"url": "https://alice:supersecret@api.example.com/v1/items?k=1",
"headers": []
},
"response": {
"status": 200,
"headers": [],
"content": {"mimeType": "text/plain", "text": "ok"}
}
},
{
"startedDateTime": "2026-05-03T10:00:01Z",
"request": {
"method": "BAD METHOD",
"url": "https://bob:topsecret@bad.example.com/fail",
"headers": []
},
"response": {"status": 200, "headers": [], "content": {}}
},
{
"startedDateTime": "2026-05-03T10:00:02Z",
"request": {
"method": "GET",
"url": "https://carol:hiddenpass@bad host/x",
"headers": []
},
"response": {"status": 200, "headers": [], "content": {}}
}
]
}
}`
res, err := tr.ImportHAR([]byte(harJSON), HARImportOptions{})
if err != nil {
t.Fatal(err)
}
if res.Imported != 1 || res.Failed != 2 || len(res.Errors) != 2 {
t.Fatalf("ImportHAR userinfo result = %+v, want 1 imported / 2 failed", res)
}
for _, e := range res.Errors {
for _, secret := range []string{"bob", "topsecret", "carol", "hiddenpass"} {
if strings.Contains(e.URL, secret) || strings.Contains(e.Reason, secret) {
t.Fatalf("import error leaked userinfo %q: %+v", secret, e)
}
}
}
// Stored URL must have userinfo stripped.
rows, total, err := tr.Page(PageQuery{RespMin: -1, RespMax: -1}, 0, 10)
if err != nil || total != 1 || len(rows) != 1 {
t.Fatalf("Page = (%d, %d, %v)", len(rows), total, err)
}
if rows[0].URL != "https://api.example.com/v1/items?k=1" {
t.Fatalf("stored URL = %q, want userinfo stripped", rows[0].URL)
}
// FTS must NOT find the stripped password.
if found, tot, err := tr.Page(PageQuery{Query: "supersecret", RespMin: -1, RespMax: -1}, 0, 10); err != nil || tot != 0 || len(found) != 0 {
t.Fatalf("FTS leaked userinfo secret: %+v", found)
}
// Exported HAR must NOT contain the password.
doc, _, err := tr.ExportHAR(PageQuery{RespMin: -1, RespMax: -1}, HARExportOptions{})
if err != nil {
t.Fatal(err)
}
raw, _ := json.Marshal(doc)
if strings.Contains(string(raw), "supersecret") || strings.Contains(string(raw), "alice") {
t.Fatalf("exported HAR leaked userinfo: %s", string(raw))
}
}
+71 -4
View File
@@ -3,8 +3,10 @@ package traffic
import (
"bytes"
"context"
"crypto/rand"
"crypto/tls"
"crypto/x509"
"encoding/hex"
"errors"
"fmt"
"io"
@@ -13,6 +15,7 @@ import (
"os"
"sort"
"strings"
"sync"
"time"
)
@@ -28,8 +31,61 @@ const (
maxReplayResponse = 2 * 1024 * 1024
// maxReplayHeaders bounds the number of request header lines.
maxReplayHeaders = 200
// replayNonceHeader is the internal header Replay attaches when ApplyRules
// is false (the default). The capture proxy strips it unconditionally
// before forwarding and bypasses rewrite rules only when the value matches
// a live registered nonce.
replayNonceHeader = "X-ARTEX-Replay-Nonce"
)
// replayNonces holds live single-use nonces for in-flight Replay calls that
// bypass capture rewrite rules (ApplyRules == false).
var replayNonces struct {
mu sync.Mutex
live map[string]struct{}
}
func registerReplayNonce() (string, error) {
var raw [32]byte
if _, err := rand.Read(raw[:]); err != nil {
return "", fmt.Errorf("replay nonce: %w", err)
}
nonce := hex.EncodeToString(raw[:])
replayNonces.mu.Lock()
if replayNonces.live == nil {
replayNonces.live = map[string]struct{}{}
}
replayNonces.live[nonce] = struct{}{}
replayNonces.mu.Unlock()
return nonce, nil
}
func deleteReplayNonce(nonce string) {
replayNonces.mu.Lock()
delete(replayNonces.live, nonce)
replayNonces.mu.Unlock()
}
func consumeReplayNonce(nonce string) bool {
if nonce == "" {
return false
}
replayNonces.mu.Lock()
_, ok := replayNonces.live[nonce]
if ok {
delete(replayNonces.live, nonce)
}
replayNonces.mu.Unlock()
return ok
}
func replayNonceCount() int {
replayNonces.mu.Lock()
n := len(replayNonces.live)
replayNonces.mu.Unlock()
return n
}
// ErrReplayInvalid marks a replay request rejected before anything is sent.
var ErrReplayInvalid = errors.New("invalid replay request")
@@ -45,10 +101,11 @@ type ReplayHeader struct {
// ReplayRequest is the edited request to send.
type ReplayRequest struct {
Method string
URL string
Headers []ReplayHeader
Body []byte
Method string
URL string
Headers []ReplayHeader
Body []byte
ApplyRules bool
}
// ReplayResult is the response as received by the client side of the proxy.
@@ -78,6 +135,7 @@ var replayDropped = map[string]bool{
"Te": true,
"Trailer": true,
"Expect": true,
http.CanonicalHeaderKey(replayNonceHeader): true,
}
// replayRecorded reports whether the proxy records a request to scheme://host:
@@ -217,6 +275,15 @@ func (t *Traffic) Replay(ctx context.Context, req ReplayRequest) (ReplayResult,
httpReq.Host = host
}
if !req.ApplyRules && t.replayRecorded(target.Scheme, target.Host) {
nonce, err := registerReplayNonce()
if err != nil {
return ReplayResult{}, err
}
defer deleteReplayNonce(nonce)
httpReq.Header.Set(replayNonceHeader, nonce)
}
start := time.Now()
resp, err := client.Do(httpReq)
if err != nil {
+334
View File
@@ -0,0 +1,334 @@
package traffic
import (
"context"
"io"
"net/http"
"net/http/httptest"
"strings"
"sync"
"testing"
"time"
"github.com/Autumn-27/artex/traffic/rewrite"
)
func advHeaderValue(hs []ReplayHeader, name string) string {
for _, h := range hs {
if strings.EqualFold(h.Name, name) {
return h.Value
}
}
return ""
}
// The internal header is stripped whatever its casing and however many
// occurrences are sent, and it never bypasses rules without a live nonce.
func TestAdversarialReplayNonceHeaderDuplicatesStripped(t *testing.T) {
var mu sync.Mutex
var seen []string
target := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
mu.Lock()
seen = append([]string{}, r.Header.Values(replayNonceHeader)...)
mu.Unlock()
w.Header().Set("X-Seen-Rule", r.Header.Get("X-Rule"))
_, _ = io.WriteString(w, "ok")
}))
t.Cleanup(target.Close)
tr, _, proxyURL := startProxy(t)
tr.SetRewriteRules(rules(t, nil, nil,
rewrite.Rule{ID: 1, Phase: rewrite.PhaseRequest, Actions: []rewrite.Action{
{Type: rewrite.ActionSetHeader, Name: "X-Rule", Value: "applied"},
}},
rewrite.Rule{ID: 2, Phase: rewrite.PhaseResponse, Actions: []rewrite.Action{
{Type: rewrite.ActionSetHeader, Name: "X-Rule-Resp", Value: "applied"},
}},
))
h := http.Header{}
h[replayNonceHeader] = []string{"aaaa", "bbbb"}
h["x-artex-replay-nonce"] = []string{"cccc"}
resp, _ := sendThrough(t, tr, proxyURL, http.MethodGet, target.URL+"/dup", "", h)
mu.Lock()
got := append([]string{}, seen...)
mu.Unlock()
if len(got) != 0 {
t.Fatalf("target saw nonce header values %v", got)
}
if resp.Header.Get("X-Rule-Resp") != "applied" {
t.Fatalf("duplicate forged nonce bypassed the response rule: %v", resp.Header)
}
if resp.Header.Get("X-Seen-Rule") != "applied" {
t.Fatalf("duplicate forged nonce bypassed the request rule: %v", resp.Header)
}
id := waitExchangeIDBySuffix(t, tr, "/dup")
reqText, _, err := tr.Get(id)
if err != nil {
t.Fatal(err)
}
if strings.Contains(strings.ToLower(reqText), strings.ToLower(replayNonceHeader)) {
t.Fatalf("recorded request kept a nonce header:\n%s", reqText)
}
}
// A nonce is single use, and takeReplayNonce refuses to return a value when
// more than one occurrence is present.
func TestAdversarialReplayNonceRegistrySingleUse(t *testing.T) {
n, err := registerReplayNonce()
if err != nil {
t.Fatal(err)
}
if len(n) != 64 {
t.Fatalf("nonce length = %d, want 64 hex chars", len(n))
}
if !consumeReplayNonce(n) {
t.Fatal("first consume did not match")
}
if consumeReplayNonce(n) {
t.Fatal("second consume matched: nonce is not single use")
}
if consumeReplayNonce("") {
t.Fatal("empty nonce matched")
}
h := http.Header{}
h.Set(replayNonceHeader, n)
if got, present := takeReplayNonce(h); !present || got != n || len(h) != 0 {
t.Fatalf("single occurrence = (%q, %v, %d), want (%q, true, 0)", got, present, len(h), n)
}
h2 := http.Header{replayNonceHeader: {"a", "b"}, "x-artex-replay-nonce": {"c"}}
if got, present := takeReplayNonce(h2); !present || got != "" || len(h2) != 0 {
t.Fatalf("duplicates = (%q, %v, %d), want (\"\", true, 0)", got, present, len(h2))
}
h3 := http.Header{replayNonceHeader: {}}
if got, present := takeReplayNonce(h3); !present || got != "" || len(h3) != 0 {
t.Fatalf("empty values = (%q, %v, %d), want (\"\", true, 0)", got, present, len(h3))
}
}
// A caller-supplied nonce header is dropped in both modes and never wins over
// the real nonce: default still bypasses, apply_rules still applies.
func TestAdversarialReplayForgedNonceHeaderCannotSteerReplay(t *testing.T) {
var mu sync.Mutex
var seen []string
var rule string
target := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
mu.Lock()
rule = r.Header.Get("X-Rule")
seen = append([]string{}, r.Header.Values(replayNonceHeader)...)
mu.Unlock()
w.Header().Set("X-Seen-Rule", r.Header.Get("X-Rule"))
_, _ = io.WriteString(w, "ok")
}))
t.Cleanup(target.Close)
tr, _, _ := startProxy(t)
tr.SetRewriteRules(rules(t, nil, nil,
rewrite.Rule{ID: 1, Phase: rewrite.PhaseRequest, Actions: []rewrite.Action{
{Type: rewrite.ActionSetHeader, Name: "X-Rule", Value: "applied"},
}},
))
forged := ReplayHeader{Name: replayNonceHeader, Value: strings.Repeat("ab", 32)}
res, err := tr.Replay(context.Background(), ReplayRequest{
Method: http.MethodGet,
URL: target.URL + "/forged-bypass",
Headers: []ReplayHeader{forged},
})
if err != nil {
t.Fatalf("default replay: %v", err)
}
mu.Lock()
if rule != "" || len(seen) != 0 {
t.Fatalf("default replay: target saw X-Rule=%q nonce=%v", rule, seen)
}
mu.Unlock()
if advHeaderValue(res.Headers, "X-Seen-Rule") != "" {
t.Fatalf("default replay with forged header did not bypass: %v", res.Headers)
}
res2, err := tr.Replay(context.Background(), ReplayRequest{
Method: http.MethodGet,
URL: target.URL + "/forged-apply",
Headers: []ReplayHeader{forged},
ApplyRules: true,
})
if err != nil {
t.Fatalf("apply replay: %v", err)
}
mu.Lock()
if rule != "applied" || len(seen) != 0 {
t.Fatalf("apply replay: target saw X-Rule=%q nonce=%v", rule, seen)
}
mu.Unlock()
if advHeaderValue(res2.Headers, "X-Seen-Rule") != "applied" {
t.Fatalf("apply replay did not apply the rule: %v", res2.Headers)
}
if replayNonceCount() != 0 {
t.Fatalf("nonce registry not empty: %d", replayNonceCount())
}
}
// A bypassing replay and an applying replay to the same target at the same
// time must not share state, and a normal request after a bypass must still be
// rewritten.
func TestAdversarialReplayConcurrentBypassAndApply(t *testing.T) {
var mu sync.Mutex
seen := map[string]string{}
target := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
mu.Lock()
seen[r.URL.Path] = r.Header.Get("X-Rule")
mu.Unlock()
w.Header().Set("X-Seen-Rule", r.Header.Get("X-Rule"))
_, _ = io.WriteString(w, "ok")
}))
t.Cleanup(target.Close)
tr, _, pxy := startProxy(t)
tr.SetRewriteRules(rules(t, nil, nil,
rewrite.Rule{ID: 1, Phase: rewrite.PhaseRequest, Actions: []rewrite.Action{
{Type: rewrite.ActionSetHeader, Name: "X-Rule", Value: "applied"},
}},
))
var wg sync.WaitGroup
var offRule, onRule string
wg.Add(2)
go func() {
defer wg.Done()
res, err := tr.Replay(context.Background(), ReplayRequest{Method: http.MethodGet, URL: target.URL + "/conc-off"})
if err != nil {
t.Errorf("bypass replay: %v", err)
return
}
offRule = advHeaderValue(res.Headers, "X-Seen-Rule")
}()
go func() {
defer wg.Done()
res, err := tr.Replay(context.Background(), ReplayRequest{Method: http.MethodGet, URL: target.URL + "/conc-on", ApplyRules: true})
if err != nil {
t.Errorf("apply replay: %v", err)
return
}
onRule = advHeaderValue(res.Headers, "X-Seen-Rule")
}()
wg.Wait()
if offRule != "" {
t.Fatalf("bypass replay saw X-Rule=%q", offRule)
}
if onRule != "applied" {
t.Fatalf("apply replay saw X-Rule=%q", onRule)
}
if replayNonceCount() != 0 {
t.Fatalf("nonce registry not empty after concurrent replays: %d", replayNonceCount())
}
// A plain request after the bypass is still rewritten: the marker is per flow.
resp, _ := sendThrough(t, tr, pxy, http.MethodGet, target.URL+"/after", "", nil)
if resp.Header.Get("X-Seen-Rule") != "applied" {
t.Fatalf("normal request after a bypass was not rewritten: %v", resp.Header)
}
mu.Lock()
defer mu.Unlock()
if seen["/conc-off"] != "" || seen["/conc-on"] != "applied" || seen["/after"] != "applied" {
t.Fatalf("target saw %v", seen)
}
}
// An SSE response is bypassed by default and rewritten when opted in: the
// stream phase is the path the plain response test cannot reach.
func TestAdversarialReplayBypassCoversSSE(t *testing.T) {
tr, _, _ := startProxy(t)
srv := httptest.NewTLSServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
w.Header().Set("Content-Type", "text/event-stream")
w.WriteHeader(200)
fl := w.(http.Flusher)
for _, ev := range []string{"data: one\n\n", "data: two\n\n"} {
_, _ = io.WriteString(w, ev)
fl.Flush()
time.Sleep(10 * time.Millisecond)
}
}))
t.Cleanup(srv.Close)
tr.SetRewriteRules(rules(t, nil, nil,
rewrite.Rule{ID: 1, Phase: rewrite.PhaseResponse, Actions: []rewrite.Action{
{Type: rewrite.ActionSetHeader, Name: "X-Rule-Resp", Value: "applied"},
}},
))
res, err := tr.Replay(context.Background(), ReplayRequest{
Method: http.MethodGet,
URL: localhostURL(srv.URL) + "/sse-off",
})
if err != nil {
t.Fatalf("bypass SSE replay: %v", err)
}
if got := advHeaderValue(res.Headers, "X-Rule-Resp"); got != "" {
t.Fatalf("bypass SSE replay applied header rule: %q", got)
}
if !strings.Contains(res.Raw, "data: two") {
t.Fatalf("bypass SSE body changed: %s", res.Raw)
}
idOff := waitExchangeIDBySuffix(t, tr, "/sse-off")
if rewritten, ids, err := tr.RewriteInfo(idOff); err != nil || rewritten || len(ids) != 0 {
t.Fatalf("bypass SSE RewriteInfo = (%v, %v, %v)", rewritten, ids, err)
}
res2, err := tr.Replay(context.Background(), ReplayRequest{
Method: http.MethodGet,
URL: localhostURL(srv.URL) + "/sse-on",
ApplyRules: true,
})
if err != nil {
t.Fatalf("apply SSE replay: %v", err)
}
if got := advHeaderValue(res2.Headers, "X-Rule-Resp"); got != "applied" {
t.Fatalf("apply SSE replay header rule = %q, want applied", got)
}
idOn := waitExchangeIDBySuffix(t, tr, "/sse-on")
if rows := traceRows(t, tr, idOn); len(rows) == 0 {
t.Fatal("apply SSE replay recorded no rewrite trace")
}
}
// A cancelled replay leaves no live nonce and no permanent bypass marker.
func TestAdversarialReplayCancelCleansUp(t *testing.T) {
block := make(chan struct{})
target := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
<-block
_, _ = io.WriteString(w, "late")
}))
t.Cleanup(target.Close)
tr, s, _ := startProxy(t)
tr.SetRewriteRules(rules(t, nil, nil,
rewrite.Rule{ID: 1, Phase: rewrite.PhaseRequest, Actions: []rewrite.Action{
{Type: rewrite.ActionSetHeader, Name: "X-Rule", Value: "applied"},
}},
))
ctx, cancel := context.WithTimeout(context.Background(), 100*time.Millisecond)
defer cancel()
if _, err := tr.Replay(ctx, ReplayRequest{Method: http.MethodGet, URL: target.URL + "/cancel"}); err == nil {
t.Fatal("cancelled replay returned no error")
}
close(block)
if replayNonceCount() != 0 {
t.Fatalf("nonce registry leaked after cancel: %d", replayNonceCount())
}
waitFor(t, "bypass marker cleanup", func() bool {
empty := true
s.bypass.Range(func(_, _ any) bool {
empty = false
return false
})
return empty
})
}
+271
View File
@@ -0,0 +1,271 @@
package traffic
import (
"context"
"io"
"net/http"
"net/http/httptest"
"slices"
"strings"
"sync"
"testing"
"time"
"github.com/Autumn-27/artex/traffic/rewrite"
mproxy "github.com/lqqyt2423/go-mitmproxy/proxy"
)
func hasNonceHeader(h http.Header) bool {
for k := range h {
if strings.EqualFold(k, replayNonceHeader) {
return true
}
}
return false
}
func waitExchangeIDBySuffix(t *testing.T, tr *Traffic, suffix string) string {
t.Helper()
deadline := time.Now().Add(5 * time.Second)
for {
var id string
err := tr.DB().QueryRow(`SELECT id FROM exchanges WHERE url LIKE ?`, "%"+suffix).Scan(&id)
if err == nil {
return id
}
if time.Now().After(deadline) {
t.Fatalf("exchange %s not recorded: %v", suffix, err)
}
time.Sleep(15 * time.Millisecond)
}
}
func TestReplayBypassesRewriteRulesByDefault(t *testing.T) {
var mu sync.Mutex
var gotMethod, gotBody, gotOrig, gotRuleReq string
var sawNonce bool
target := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
b, _ := io.ReadAll(r.Body)
mu.Lock()
gotMethod = r.Method
gotBody = string(b)
gotOrig = r.Header.Get("X-Orig")
gotRuleReq = r.Header.Get("X-Rule-Req")
sawNonce = hasNonceHeader(r.Header)
mu.Unlock()
w.Header().Set("Content-Type", "text/plain")
_, _ = io.WriteString(w, "target-resp")
}))
t.Cleanup(target.Close)
tr, s, _ := startProxy(t)
tr.SetRewriteRules(rules(t, nil, nil,
rewrite.Rule{ID: 1, Phase: rewrite.PhaseRequest, Actions: []rewrite.Action{
{Type: rewrite.ActionSetHeader, Name: "X-Rule-Req", Value: "applied"},
{Type: rewrite.ActionSetBody, Value: "rewritten-req"},
}},
rewrite.Rule{ID: 2, Phase: rewrite.PhaseResponse, Actions: []rewrite.Action{
{Type: rewrite.ActionSetHeader, Name: "X-Rule-Resp", Value: "applied"},
{Type: rewrite.ActionReplaceBody, Regex: "target-resp", Value: "rewritten-resp"},
}},
))
// ApplyRules omitted (false by default) -> bypass both phases.
res, err := tr.Replay(context.Background(), ReplayRequest{
Method: http.MethodPost,
URL: target.URL + "/default-bypass",
Headers: []ReplayHeader{{Name: "X-Orig", Value: "1"}},
Body: []byte("orig-body"),
})
if err != nil {
t.Fatalf("Replay: %v", err)
}
if replayNonceCount() != 0 {
t.Fatalf("nonce registry not empty after Replay: %d", replayNonceCount())
}
mu.Lock()
if gotMethod != http.MethodPost || gotBody != "orig-body" || gotOrig != "1" || gotRuleReq != "" || sawNonce {
t.Fatalf("target received method=%q body=%q X-Orig=%q X-Rule-Req=%q sawNonce=%v", gotMethod, gotBody, gotOrig, gotRuleReq, sawNonce)
}
mu.Unlock()
if res.Status != 200 || !res.Recorded || strings.Contains(res.Raw, "X-Rule-Resp") || !strings.HasSuffix(res.Raw, "target-resp") {
t.Fatalf("client received modified response: %+v", res)
}
id := waitExchangeIDBySuffix(t, tr, "/default-bypass")
if rewritten, ids, err := tr.RewriteInfo(id); err != nil || rewritten || len(ids) != 0 {
t.Fatalf("RewriteInfo = (%v, %v, %v), want (false, nil, nil)", rewritten, ids, err)
}
var traceCount int
if err := tr.DB().QueryRow(`SELECT COUNT(*) FROM exchange_rewrites WHERE exchange_id=?`, id).Scan(&traceCount); err != nil {
t.Fatal(err)
}
if traceCount != 0 {
t.Fatalf("exchange_rewrites rows = %d, want 0", traceCount)
}
reqText, _, err := tr.Get(id)
if err != nil {
t.Fatal(err)
}
if strings.Contains(strings.ToLower(reqText), strings.ToLower(replayNonceHeader)) {
t.Fatalf("recorded request leaked internal nonce header:\n%s", reqText)
}
// Response-only rules must also be bypassed and the nonce header stripped.
tr.SetRewriteRules(rules(t, nil, nil,
rewrite.Rule{ID: 3, Phase: rewrite.PhaseResponse, Actions: []rewrite.Action{
{Type: rewrite.ActionSetHeader, Name: "X-Rule-Resp", Value: "applied"},
}},
))
res2, err := tr.Replay(context.Background(), ReplayRequest{
Method: http.MethodGet,
URL: target.URL + "/resp-only-bypass",
})
if err != nil {
t.Fatalf("Replay response-only: %v", err)
}
mu.Lock()
if sawNonce {
t.Fatal("target saw nonce header when only response rules were active")
}
mu.Unlock()
if strings.Contains(res2.Raw, "X-Rule-Resp") {
t.Fatalf("response-only rule was not bypassed: %s", res2.Raw)
}
id2 := waitExchangeIDBySuffix(t, tr, "/resp-only-bypass")
if rewritten, _, err := tr.RewriteInfo(id2); err != nil || rewritten {
t.Fatalf("response-only bypass RewriteInfo = %v %v", rewritten, err)
}
// RequestError clears the bypass marker.
f := withID(&mproxy.Flow{}, 99)
s.bypass.Store(keyOf(f), struct{}{})
s.RequestError(f, io.EOF)
if _, ok := s.bypass.Load(keyOf(f)); ok {
t.Fatal("RequestError did not clear bypass marker")
}
// Failed Replay (unreachable target) still leaves nonce registry empty.
_, _ = tr.Replay(context.Background(), ReplayRequest{
Method: http.MethodGet,
URL: "http://127.0.0.1:1/unreachable",
})
if replayNonceCount() != 0 {
t.Fatalf("nonce registry leaked after failed Replay: %d", replayNonceCount())
}
}
func TestReplayApplyRulesTrue(t *testing.T) {
var mu sync.Mutex
var gotBody, gotRuleReq string
var sawNonce bool
target := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
b, _ := io.ReadAll(r.Body)
mu.Lock()
gotBody = string(b)
gotRuleReq = r.Header.Get("X-Rule-Req")
sawNonce = hasNonceHeader(r.Header)
mu.Unlock()
w.Header().Set("Content-Type", "text/plain")
_, _ = io.WriteString(w, "target-resp")
}))
t.Cleanup(target.Close)
tr, _, _ := startProxy(t)
tr.SetRewriteRules(rules(t, nil, nil,
rewrite.Rule{ID: 1, Phase: rewrite.PhaseRequest, Actions: []rewrite.Action{
{Type: rewrite.ActionSetHeader, Name: "X-Rule-Req", Value: "applied"},
{Type: rewrite.ActionSetBody, Value: "rewritten-req"},
}},
rewrite.Rule{ID: 2, Phase: rewrite.PhaseResponse, Actions: []rewrite.Action{
{Type: rewrite.ActionSetHeader, Name: "X-Rule-Resp", Value: "applied"},
{Type: rewrite.ActionReplaceBody, Regex: "target-resp", Value: "rewritten-resp"},
}},
))
res, err := tr.Replay(context.Background(), ReplayRequest{
Method: http.MethodPost,
URL: target.URL + "/apply-rules",
Body: []byte("orig-body"),
ApplyRules: true,
})
if err != nil {
t.Fatalf("Replay(ApplyRules=true): %v", err)
}
if replayNonceCount() != 0 {
t.Fatalf("nonce registry not empty after Replay(ApplyRules=true): %d", replayNonceCount())
}
mu.Lock()
if gotBody != "rewritten-req" || gotRuleReq != "applied" || sawNonce {
t.Fatalf("target received body=%q X-Rule-Req=%q sawNonce=%v", gotBody, gotRuleReq, sawNonce)
}
mu.Unlock()
if !strings.Contains(res.Raw, "X-Rule-Resp: applied") || !strings.HasSuffix(res.Raw, "rewritten-resp") {
t.Fatalf("response rule did not apply on Replay(ApplyRules=true): %s", res.Raw)
}
id := waitExchangeIDBySuffix(t, tr, "/apply-rules")
rewritten, ids, err := tr.RewriteInfo(id)
if err != nil || !rewritten || !slices.Equal(ids, []int64{1, 2}) {
t.Fatalf("RewriteInfo = (%v, %v, %v), want (true, [1 2], nil)", rewritten, ids, err)
}
if rows := traceRows(t, tr, id); len(rows) == 0 {
t.Fatal("expected exchange_rewrites trace rows when ApplyRules=true")
}
}
func TestForgedReplayNonceHeaderStrippedAndIgnored(t *testing.T) {
var mu sync.Mutex
var gotRuleReq string
var sawNonce bool
target := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
mu.Lock()
gotRuleReq = r.Header.Get("X-Rule-Req")
sawNonce = hasNonceHeader(r.Header)
mu.Unlock()
w.Header().Set("Content-Type", "text/plain")
_, _ = io.WriteString(w, "target-resp")
}))
t.Cleanup(target.Close)
tr, _, proxyURL := startProxy(t)
tr.SetRewriteRules(rules(t, nil, nil,
rewrite.Rule{ID: 1, Phase: rewrite.PhaseRequest, Actions: []rewrite.Action{
{Type: rewrite.ActionSetHeader, Name: "X-Rule-Req", Value: "applied"},
}},
rewrite.Rule{ID: 2, Phase: rewrite.PhaseResponse, Actions: []rewrite.Action{
{Type: rewrite.ActionSetHeader, Name: "X-Rule-Resp", Value: "applied"},
}},
))
resp, _ := sendThrough(t, tr, proxyURL, http.MethodGet, target.URL+"/forged", "", http.Header{
replayNonceHeader: {"0000000000000000000000000000000000000000000000000000000000000000"},
})
mu.Lock()
if sawNonce {
t.Fatal("forged replay nonce header reached target")
}
if gotRuleReq != "applied" {
t.Fatalf("forged nonce bypassed request rule: X-Rule-Req=%q", gotRuleReq)
}
mu.Unlock()
if resp.Header.Get("X-Rule-Resp") != "applied" {
t.Fatalf("forged nonce bypassed response rule: X-Rule-Resp=%q", resp.Header.Get("X-Rule-Resp"))
}
id := waitExchangeIDBySuffix(t, tr, "/forged")
rewritten, ids, err := tr.RewriteInfo(id)
if err != nil || !rewritten || !slices.Equal(ids, []int64{1, 2}) {
t.Fatalf("forged nonce RewriteInfo = (%v, %v, %v), want (true, [1 2], nil)", rewritten, ids, err)
}
reqText, _, err := tr.Get(id)
if err != nil {
t.Fatal(err)
}
if strings.Contains(strings.ToLower(reqText), strings.ToLower(replayNonceHeader)) {
t.Fatalf("recorded request kept forged nonce header:\n%s", reqText)
}
}
+46
View File
@@ -6,6 +6,7 @@ import (
"log"
"net/http"
"net/url"
"strings"
"sync"
"sync/atomic"
"time"
@@ -420,6 +421,32 @@ func engineRequest(r *mproxy.Request) rewrite.Request {
return rewrite.Request{Method: r.Method, URL: r.URL, Header: r.Header, Body: r.Body}
}
// takeReplayNonce removes every occurrence of X-ARTEX-Replay-Nonce from h
// unconditionally and returns its value only when exactly one header value was
// present.
func takeReplayNonce(h http.Header) (nonce string, present bool) {
if h == nil {
return "", false
}
count := 0
for k, vs := range h {
if strings.EqualFold(k, replayNonceHeader) {
present = true
for _, v := range vs {
count++
if count == 1 {
nonce = strings.TrimSpace(v)
}
}
delete(h, k)
}
}
if count != 1 {
return "", present
}
return nonce, present
}
// Request runs the request phase. The result goes on the flow only once the
// request's snapshot slot is reserved; without one the request leaves as the
// client sent it. Rules fail open: a panic leaves the request unchanged.
@@ -427,6 +454,12 @@ func (s *sink) Request(f *mproxy.Flow) {
if s.t == nil || f.Request == nil || f.Request.URL == nil {
return
}
// Strip the internal replay bypass header unconditionally before forwarding
// or running rules; honour it only when its value matches a registered nonce.
if nonce, present := takeReplayNonce(f.Request.Header); present && consumeReplayNonce(nonce) {
s.bypass.Store(keyOf(f), struct{}{})
return
}
set := s.t.rewriteRules.Load()
if set.Len(rewrite.PhaseRequest) == 0 {
return
@@ -459,6 +492,10 @@ func (s *sink) Response(f *mproxy.Flow) {
if f.Request == nil || f.Response == nil {
return
}
if _, bypassed := s.bypass.LoadAndDelete(keyOf(f)); bypassed {
s.t.record(f)
return
}
snap := s.t.snaps.take(keyOf(f))
set := s.t.rewriteRules.Load()
if snap != nil {
@@ -506,6 +543,13 @@ func (s *sink) applyResponse(f *mproxy.Flow, set *rewrite.Set, in rewrite.Input)
// upgrade, with no response on the flow: nothing of the handshake can change
// there, so every candidate rule is traced "not rewritten: stream".
func (s *sink) rewriteStream(f *mproxy.Flow, kind string) (rq requestView, rs responseView, lines []rewrite.TraceLine, res rewrite.Result) {
if _, bypassed := s.bypass.LoadAndDelete(keyOf(f)); bypassed {
rq = flowRequestView(f.Request)
if kind == "sse" {
rs = flowResponseView(f.Response)
}
return rq, rs, nil, rewrite.Result{}
}
snap := s.t.snaps.take(keyOf(f))
set := s.t.rewriteRules.Load()
if snap != nil {
@@ -568,6 +612,7 @@ func (s *sink) applyStreamResponse(f *mproxy.Flow, res rewrite.Result) {
// and the rules, never the URL or the error text.
func (s *sink) RequestError(f *mproxy.Flow, err error) {
if f != nil {
s.bypass.Delete(keyOf(f))
if snap := s.t.snaps.take(keyOf(f)); snap != nil && snap.requestRewritten() {
ids := snap.appliedRuleIDs()
for _, id := range ids {
@@ -593,6 +638,7 @@ func (s *sink) StreamResponseModifier(f *mproxy.Flow, in io.Reader) (out io.Read
if s == nil || s.t == nil || f == nil || !f.Stream || f.SSE != nil {
return in
}
s.bypass.Delete(keyOf(f))
snap := s.t.snaps.take(keyOf(f))
if snap == nil {
return in
+64 -31
View File
@@ -469,6 +469,7 @@ type sink struct {
t *Traffic
paced sync.Map // ClientConn.Id of a marked connection -> struct{}
streams sync.Map // *mproxy.Flow of a live WebSocket/SSE stream -> *streamState (stream.go)
bypass sync.Map // flowKey of a replay flow bypassing rewrite rules -> struct{}
}
// maybePassthrough marks a host to be tunneled transparently on the next
@@ -979,7 +980,7 @@ func clampBlobRange(offset, length, size int64) (off, n int64, empty bool) {
return offset, length, false
}
// ExchangeMeta is one row of the index (returned by Search).
// ExchangeMeta is one row of the index (returned by Search and Page).
type ExchangeMeta struct {
ID string `json:"id"`
TS int64 `json:"ts"`
@@ -991,14 +992,17 @@ type ExchangeMeta struct {
ContentType string `json:"content_type"`
RespLen int `json:"resp_len"`
Path string `json:"path"`
Rewritten bool `json:"rewritten"`
}
const rewrittenExistsSQL = `EXISTS(SELECT 1 FROM exchange_rewrites r WHERE r.exchange_id=exchanges.id AND r.outcome='applied')`
// Search returns paged exchange metadata (never bodies).
func (t *Traffic) Search(host string, page, size int) ([]ExchangeMeta, error) {
if size <= 0 || size > 500 {
size = 100
}
q := `SELECT id,ts,host,method,url_template,url,status,content_type,resp_len,path FROM exchanges`
q := `SELECT id,ts,host,method,url_template,url,status,content_type,resp_len,path,` + rewrittenExistsSQL + ` FROM exchanges`
args := []any{}
if host != "" {
q += ` WHERE host=?`
@@ -1014,7 +1018,7 @@ func (t *Traffic) Search(host string, page, size int) ([]ExchangeMeta, error) {
var out []ExchangeMeta
for rows.Next() {
var m ExchangeMeta
if err := rows.Scan(&m.ID, &m.TS, &m.Host, &m.Method, &m.URLTemplate, &m.URL, &m.Status, &m.ContentType, &m.RespLen, &m.Path); err != nil {
if err := rows.Scan(&m.ID, &m.TS, &m.Host, &m.Method, &m.URLTemplate, &m.URL, &m.Status, &m.ContentType, &m.RespLen, &m.Path, &m.Rewritten); err != nil {
return nil, err
}
out = append(out, m)
@@ -1041,21 +1045,41 @@ func ftsQuote(s string) string {
return `"` + strings.ReplaceAll(s, `"`, `""`) + `"`
}
const (
RewriteFilterAll = "all"
RewriteFilterRewritten = "rewritten"
RewriteFilterOriginal = "original"
)
// PageQuery bundles the optional filters and sort order for Page. Every filter
// is combined with AND; an empty string skips that filter, and RespMin/RespMax
// of -1 skip the response-size bounds. An empty Sort/Order defaults to
// newest-first by timestamp.
type PageQuery struct {
Host string // host substring
Method string // exact method (case-insensitive)
Query string // broad search box: metadata columns OR captured text
Body string // content box: captured request/response text only (full-text)
Path string // url_template substring — the vulnerability's path
Status string // exact code ("404") or class bucket ("4xx")
RespMin int64 // minimum resp_len, or -1 when unset
RespMax int64 // maximum resp_len, or -1 when unset
Sort string // ts | status | resp_len (default ts)
Order string // asc | desc (default desc)
Host string // host substring
Method string // exact method (case-insensitive)
Query string // broad search box: metadata columns OR captured text
Body string // content box: captured request/response text only (full-text)
Path string // url_template substring — the vulnerability's path
Status string // exact code ("404") or class bucket ("4xx")
Rewritten string // all (default) | rewritten (or true) | original (or false)
RespMin int64 // minimum resp_len, or -1 when unset
RespMax int64 // maximum resp_len, or -1 when unset
Sort string // ts | status | resp_len (default ts)
Order string // asc | desc (default desc)
}
// rewriteFilterSQL turns a rewritten filter token into a SQL condition over
// exchange_rewrites. Empty, "all", or unrecognized values default to no filter.
func rewriteFilterSQL(s string) (cond string, ok bool) {
switch strings.ToLower(strings.TrimSpace(s)) {
case RewriteFilterRewritten, "true", "1":
return rewrittenExistsSQL, true
case RewriteFilterOriginal, "false", "0":
return "NOT " + rewrittenExistsSQL, true
default:
return "", false
}
}
// sortColumns whitelists the columns Page may order by, so the caller-supplied
@@ -1080,22 +1104,10 @@ func statusFilter(s string) (cond string, args []any, ok bool) {
return "", nil, false
}
// Page returns one page of exchange metadata for the traffic list, filtered by
// the criteria in f (all optional, combined with AND) and ordered by the
// requested column. Query is the broad search box (metadata columns OR, when the
// term is long enough for the trigram index, captured request/response text);
// Body narrows to exchanges whose captured text matches, via that same index.
// Also returns the total number of rows matching the filter (for the UI's
// pagination). Bodies are never included in the rows.
func (t *Traffic) Page(f PageQuery, page, size int) (rows []ExchangeMeta, total int, err error) {
if size <= 0 || size > 500 {
size = 100
}
if page < 0 {
page = 0
}
where := ""
var args []any
// pageWhere builds the SQL WHERE clause and bound arguments for PageQuery.
// Page and ExportHAR share this so list queries and HAR exports honor the exact
// same filter semantics.
func (t *Traffic) pageWhere(f PageQuery) (where string, args []any) {
add := func(cond string, vs ...any) {
if where == "" {
where = " WHERE "
@@ -1117,6 +1129,9 @@ func (t *Traffic) Page(f PageQuery, page, size int) (rows []ExchangeMeta, total
if cond, sargs, ok := statusFilter(f.Status); ok {
add(cond, sargs...)
}
if cond, ok := rewriteFilterSQL(f.Rewritten); ok {
add(cond)
}
if f.RespMin >= 0 {
add("resp_len>=?", f.RespMin)
}
@@ -1143,6 +1158,24 @@ func (t *Traffic) Page(f PageQuery, page, size int) (rows []ExchangeMeta, total
add(cond, arg)
}
}
return where, args
}
// Page returns one page of exchange metadata for the traffic list, filtered by
// the criteria in f (all optional, combined with AND) and ordered by the
// requested column. Query is the broad search box (metadata columns OR, when the
// term is long enough for the trigram index, captured request/response text);
// Body narrows to exchanges whose captured text matches, via that same index.
// Also returns the total number of rows matching the filter (for the UI's
// pagination). Bodies are never included in the rows.
func (t *Traffic) Page(f PageQuery, page, size int) (rows []ExchangeMeta, total int, err error) {
if size <= 0 || size > 500 {
size = 100
}
if page < 0 {
page = 0
}
where, args := t.pageWhere(f)
if err = t.db.QueryRow(`SELECT COUNT(*) FROM exchanges`+where, args...).Scan(&total); err != nil {
return nil, 0, err
}
@@ -1156,7 +1189,7 @@ func (t *Traffic) Page(f PageQuery, page, size int) (rows []ExchangeMeta, total
}
// id embeds capture timestamp + sequence, so a trailing id DESC makes the order
// total and pagination stable even when the sort column has many ties.
sel := `SELECT id,ts,host,method,url_template,url,status,content_type,resp_len,path FROM exchanges` +
sel := `SELECT id,ts,host,method,url_template,url,status,content_type,resp_len,path,` + rewrittenExistsSQL + ` FROM exchanges` +
where + ` ORDER BY ` + col + ` ` + dir + `, id DESC LIMIT ? OFFSET ?`
qargs := append(append([]any{}, args...), size, page*size)
rs, err := t.db.Query(sel, qargs...)
@@ -1166,7 +1199,7 @@ func (t *Traffic) Page(f PageQuery, page, size int) (rows []ExchangeMeta, total
defer rs.Close()
for rs.Next() {
var m ExchangeMeta
if err := rs.Scan(&m.ID, &m.TS, &m.Host, &m.Method, &m.URLTemplate, &m.URL, &m.Status, &m.ContentType, &m.RespLen, &m.Path); err != nil {
if err := rs.Scan(&m.ID, &m.TS, &m.Host, &m.Method, &m.URLTemplate, &m.URL, &m.Status, &m.ContentType, &m.RespLen, &m.Path, &m.Rewritten); err != nil {
return nil, 0, err
}
rows = append(rows, m)
+90
View File
@@ -526,3 +526,93 @@ func TestRecordConcurrent(t *testing.T) {
t.Fatalf("after concurrent writes exchanges=%d bodies=%d, want %d each", exchanges, bodies, n)
}
}
func TestTrafficPageRewrittenFlagAndFilter(t *testing.T) {
tr, _ := openTraffic(t)
insertEx := func(id string, ts int64, host string, status int) {
t.Helper()
if _, err := tr.DB().Exec(`INSERT INTO exchanges(id,ts,host,method,url_template,url,status,content_type,req_len,resp_len,path)
VALUES(?,?,?,?,?,?,?,?,?,?,?)`, id, ts, host, "GET", "/p/"+id, "https://"+host+"/p/"+id, status, "text/plain", 0, 10, ""); err != nil {
t.Fatal(err)
}
}
insertEx("rw-a-200", 1, "alpha.example", 200)
insertEx("rw-b-404", 2, "beta.example", 404)
insertEx("orig-a-200", 3, "alpha.example", 200)
insertEx("skip-a-404", 4, "alpha.example", 404)
insertEx("err-b-200", 5, "beta.example", 200)
if _, err := tr.DB().Exec(`INSERT INTO exchange_rewrites(exchange_id,seq,phase,rule_id,outcome) VALUES
('rw-a-200',0,'request',1,'applied'),
('rw-b-404',0,'response',2,'applied'),
('skip-a-404',0,'request',3,'skipped'),
('err-b-200',0,'response',4,'error')`); err != nil {
t.Fatal(err)
}
// Default ("") and "all" return all 5 with accurate Rewritten flags.
for _, filter := range []string{"", RewriteFilterAll} {
rows, total, err := tr.Page(PageQuery{Rewritten: filter, RespMin: -1, RespMax: -1}, 0, 20)
if err != nil || total != 5 || len(rows) != 5 {
t.Fatalf("filter=%q got (%d, total=%d, err=%v), want (5, 5, nil)", filter, len(rows), total, err)
}
got := map[string]bool{}
for _, r := range rows {
got[r.ID] = r.Rewritten
}
if !got["rw-a-200"] || !got["rw-b-404"] || got["orig-a-200"] || got["skip-a-404"] || got["err-b-200"] {
t.Fatalf("filter=%q flags = %v", filter, got)
}
}
// Rewritten filter (canonical + true alias).
for _, filter := range []string{RewriteFilterRewritten, "true"} {
rows, total, err := tr.Page(PageQuery{Rewritten: filter, RespMin: -1, RespMax: -1}, 0, 20)
if err != nil || total != 2 || len(rows) != 2 {
t.Fatalf("filter=%q got (%d, total=%d, err=%v), want (2, 2, nil)", filter, len(rows), total, err)
}
for _, r := range rows {
if !r.Rewritten {
t.Fatalf("filter=%q returned non-rewritten row %s", filter, r.ID)
}
}
}
// Original filter (canonical + false alias) includes plain, skipped-only, and error-only.
for _, filter := range []string{RewriteFilterOriginal, "false"} {
rows, total, err := tr.Page(PageQuery{Rewritten: filter, RespMin: -1, RespMax: -1}, 0, 20)
if err != nil || total != 3 || len(rows) != 3 {
t.Fatalf("filter=%q got (%d, total=%d, err=%v), want (3, 3, nil)", filter, len(rows), total, err)
}
for _, r := range rows {
if r.Rewritten {
t.Fatalf("filter=%q returned rewritten row %s", filter, r.ID)
}
}
}
// Combined with host + status and pagination.
rows, total, err := tr.Page(PageQuery{Host: "alpha", Status: "200", Rewritten: RewriteFilterRewritten, RespMin: -1, RespMax: -1}, 0, 10)
if err != nil || total != 1 || len(rows) != 1 || rows[0].ID != "rw-a-200" {
t.Fatalf("combined host+status+rewritten = (%+v, total=%d, err=%v)", rows, total, err)
}
p0, total0, err := tr.Page(PageQuery{Rewritten: RewriteFilterOriginal, RespMin: -1, RespMax: -1}, 0, 2)
p1, total1, err2 := tr.Page(PageQuery{Rewritten: RewriteFilterOriginal, RespMin: -1, RespMax: -1}, 1, 2)
if err != nil || err2 != nil || total0 != 3 || total1 != 3 || len(p0) != 2 || len(p1) != 1 {
t.Fatalf("pagination original p0=(%d/%d,%v) p1=(%d/%d,%v)", len(p0), total0, err, len(p1), total1, err2)
}
// Search also carries Rewritten.
srows, err := tr.Search("alpha.example", 0, 10)
if err != nil || len(srows) != 3 {
t.Fatalf("Search = (%d, %v), want (3, nil)", len(srows), err)
}
found := map[string]bool{}
for _, r := range srows {
found[r.ID] = r.Rewritten
}
if !found["rw-a-200"] || found["orig-a-200"] || found["skip-a-404"] {
t.Fatalf("Search rewritten flags = %v", found)
}
}
@@ -7,6 +7,7 @@ import { Loader2Icon, PlusIcon, SendIcon, XIcon } from "lucide-react";
import { HttpCodeBlock, statusTone } from "@/components/http-code-block";
import { Badge } from "@/components/ui/badge";
import { Button } from "@/components/ui/button";
import { Checkbox } from "@/components/ui/checkbox";
import { Input } from "@/components/ui/input";
import { Table, TableBody, TableCell, TableHead, TableHeader, TableRow } from "@/components/ui/table";
import { Textarea } from "@/components/ui/textarea";
@@ -23,6 +24,7 @@ type ReplayForm = {
url: string;
rows: HeaderRow[];
body: string;
applyRules: boolean;
// Headers of the recorded request that the server would not send.
dropped: string[];
binary: boolean;
@@ -40,6 +42,7 @@ function initialForm(exchange: TrafficExchange, raw: string): ReplayForm {
url: exchange.url,
rows: toRows(sent),
body: parsed.binary ? "" : parsed.body,
applyRules: false,
dropped: notSent.filter((n) => n.toLowerCase() !== "host"),
binary: parsed.binary,
bodyPreviewOnly: parsed.bodyPreviewOnly,
@@ -83,6 +86,7 @@ export function ReplayPanel({
// Answers for an older exchange are dropped.
const generation = React.useRef(0);
const methodListId = React.useId();
const applyRulesId = React.useId();
React.useEffect(() => {
generation.current++;
@@ -111,6 +115,7 @@ export function ReplayPanel({
headers: splitReplayHeaders(form.rows).sent,
body: form.body,
body_base64: false,
apply_rules: form.applyRules,
})
.then((r) => {
if (gen !== generation.current) return;
@@ -230,10 +235,20 @@ export function ReplayPanel({
)}
</div>
<Button onClick={send} disabled={!canSend}>
{sending ? <Loader2Icon className="animate-spin" /> : <SendIcon />}
{t("Send")}
</Button>
<div className="flex flex-wrap items-center gap-4">
<Button onClick={send} disabled={!canSend}>
{sending ? <Loader2Icon className="animate-spin" /> : <SendIcon />}
{t("Send")}
</Button>
<label htmlFor={applyRulesId} className="flex cursor-pointer items-center gap-2 text-xs select-none">
<Checkbox
id={applyRulesId}
checked={form.applyRules}
onCheckedChange={(checked) => update({ applyRules: checked === true })}
/>
<span>{t("Apply capture rewrite rules")}</span>
</label>
</div>
{error && (
<div className="space-y-1 rounded-md border border-red-500/40 bg-red-500/10 p-3 text-xs">
+199 -5
View File
@@ -7,6 +7,7 @@ import {
ArrowUpNarrowWideIcon,
ChevronLeftIcon,
ChevronRightIcon,
DownloadIcon,
EraserIcon,
FilterXIcon,
ListChecksIcon,
@@ -14,11 +15,14 @@ import {
RadioTowerIcon,
SearchIcon,
Trash2Icon,
UploadIcon,
XIcon,
} from "lucide-react";
import { toast } from "sonner";
import { HttpCodeBlock } from "@/components/http-code-block";
import { LinkTrafficDialog } from "@/components/link-traffic-dialog";
import { Alert, AlertDescription } from "@/components/ui/alert";
import {
AlertDialog,
AlertDialogAction,
@@ -44,7 +48,7 @@ import { Tooltip, TooltipContent, TooltipTrigger } from "@/components/ui/tooltip
import { api } from "@/lib/api";
import { setVisibleInterval } from "@/lib/polling";
import { useStoredSortPreference } from "@/lib/sort-preference";
import type { TrafficDetail, TrafficExchange, TrafficHost, TrafficResp } from "@/lib/types";
import type { TrafficDetail, TrafficExchange, TrafficHARImportResult, TrafficHost, TrafficResp } from "@/lib/types";
import { cn } from "@/lib/utils";
import { intlLocale, t } from "@/i18n";
@@ -135,6 +139,7 @@ export default function TrafficPage() {
const [path, setPath] = React.useState("");
const [pathQ, setPathQ] = React.useState("");
const [statusFilter, setStatusFilter] = React.useState("all");
const [rewrittenFilter, setRewrittenFilter] = React.useState("all");
const [respMin, setRespMin] = React.useState("");
const [respMinQ, setRespMinQ] = React.useState("");
const [respMax, setRespMax] = React.useState("");
@@ -158,6 +163,13 @@ export default function TrafficPage() {
const [deleting, setDeleting] = React.useState(false);
const [reloadTick, setReloadTick] = React.useState(0); // manual refetch trigger
// HAR 1.2 import/export state
const harInputRef = React.useRef<HTMLInputElement | null>(null);
const [harImporting, setHarImporting] = React.useState(false);
const [harExporting, setHarExporting] = React.useState(false);
const [harSummary, setHarSummary] = React.useState<TrafficHARImportResult | null>(null);
const [harError, setHarError] = React.useState<string | null>(null);
// Debounce both filters so we don't refetch on every keystroke.
React.useEffect(() => {
const t = setTimeout(() => setHostQ(host.trim()), 300);
@@ -184,13 +196,15 @@ export default function TrafficPage() {
return () => clearTimeout(t);
}, [respMax]);
const hasAdvancedFilter = Boolean(bodyQ || pathQ || respMinQ || respMaxQ) || statusFilter !== "all";
const hasAdvancedFilter =
Boolean(bodyQ || pathQ || respMinQ || respMaxQ) || statusFilter !== "all" || rewrittenFilter !== "all";
const resetAdvancedFilters = () => {
setBody("");
setBodyQ("");
setPath("");
setPathQ("");
setStatusFilter("all");
setRewrittenFilter("all");
setRespMin("");
setRespMinQ("");
setRespMax("");
@@ -201,7 +215,7 @@ export default function TrafficPage() {
// biome-ignore lint/correctness/useExhaustiveDependencies: these values intentionally trigger a page reset.
React.useEffect(() => {
setPage(0);
}, [hostQ, queryQ, method, size, bodyQ, pathQ, statusFilter, respMinQ, respMaxQ, sort]);
}, [hostQ, queryQ, method, size, bodyQ, pathQ, statusFilter, rewrittenFilter, respMinQ, respMaxQ, sort]);
// Load the current page. Auto-refresh only on page 0 (newest) so paging back
// through history isn't yanked out from under the user.
@@ -214,6 +228,7 @@ export default function TrafficPage() {
body: bodyQ,
path: pathQ,
status: statusFilter,
rewritten: rewrittenFilter,
respMin: respMinQ,
respMax: respMaxQ,
sort: sort.field,
@@ -243,7 +258,21 @@ export default function TrafficPage() {
alive = false;
stopPolling();
};
}, [page, size, hostQ, method, queryQ, bodyQ, pathQ, statusFilter, respMinQ, respMaxQ, sort, reloadTick]);
}, [
page,
size,
hostQ,
method,
queryQ,
bodyQ,
pathQ,
statusFilter,
rewrittenFilter,
respMinQ,
respMaxQ,
sort,
reloadTick,
]);
// Delete traffic for the current host filter (substring) or the checked
// hosts (exact batch), then refetch.
@@ -349,6 +378,76 @@ export default function TrafficPage() {
: { field, direction: "desc" },
);
const handleHarFileChange = async (event: React.ChangeEvent<HTMLInputElement>) => {
const file = event.target.files?.[0];
event.target.value = "";
if (!file) return;
setHarError(null);
setHarSummary(null);
if (file.size > 32 * 1024 * 1024) {
const msg = t("HAR file exceeds the 32 MiB limit");
setHarError(msg);
toast.error(msg);
return;
}
setHarImporting(true);
try {
const raw = await file.text();
const res = await api.trafficImportHAR(raw);
setHarSummary(res);
const summary = t("HAR import complete: {imported} imported, {skipped} skipped, {failed} failed", {
imported: res.imported,
skipped: res.skipped,
failed: res.failed,
});
if (res.failed > 0 || res.truncated) {
toast.warning(summary);
} else {
toast.success(summary);
}
setPage(0);
setReloadTick((n) => n + 1);
} catch (e) {
const msg = t("HAR import failed: {error}", { error: (e as Error).message });
setHarError(msg);
toast.error(msg);
} finally {
setHarImporting(false);
}
};
const handleHarExport = async () => {
setHarExporting(true);
try {
const meta = await api.trafficExportHAR({
host: hostQ,
method,
q: queryQ,
body: bodyQ,
path: pathQ,
status: statusFilter,
rewritten: rewrittenFilter,
respMin: respMinQ,
respMax: respMaxQ,
});
if (meta.truncated) {
toast.warning(
t("Exported {exported} of {matched} matching exchanges (truncated: {reason})", {
exported: meta.exported,
matched: meta.matched,
reason: meta.truncatedReason ?? "limit",
}),
);
} else {
toast.success(t("Exported {count} exchanges to HAR", { count: meta.exported }));
}
} catch (e) {
toast.error((e as Error).message);
} finally {
setHarExporting(false);
}
};
const exchanges = React.useMemo(() => traffic?.exchanges ?? [], [traffic]);
const total = traffic?.total ?? exchanges.length;
const pageCount = Math.max(1, Math.ceil(total / size));
@@ -378,9 +477,76 @@ export default function TrafficPage() {
<span className="text-xs text-muted-foreground tabular-nums">
{t("{count} records", { count: traffic?.count ?? 0 })}
</span>
<input
ref={harInputRef}
type="file"
accept=".har,application/json"
className="hidden"
onChange={(e) => void handleHarFileChange(e)}
/>
<Button
variant="outline"
size="sm"
className="h-8"
disabled={harImporting}
onClick={() => harInputRef.current?.click()}
>
{harImporting ? <Loader2Icon className="size-3.5 animate-spin" /> : <UploadIcon className="size-3.5" />}
{harImporting ? t("Importing HAR…") : t("Import HAR")}
</Button>
<Button
variant="outline"
size="sm"
className="h-8"
disabled={harExporting || !traffic?.count}
onClick={() => void handleHarExport()}
>
{harExporting ? <Loader2Icon className="size-3.5 animate-spin" /> : <DownloadIcon className="size-3.5" />}
{harExporting ? t("Exporting HAR…") : t("Export HAR")}
</Button>
</div>
</div>
{harError ? (
<Alert variant="destructive" className="flex items-center justify-between gap-2 py-2">
<AlertDescription className="text-xs">{harError}</AlertDescription>
<Button variant="ghost" size="icon-xs" aria-label={t("Dismiss")} onClick={() => setHarError(null)}>
<XIcon className="size-3.5" />
</Button>
</Alert>
) : null}
{harSummary ? (
<Alert className="flex items-start justify-between gap-2 py-2">
<AlertDescription className="space-y-1 text-xs">
<p>
{t("HAR import complete: {imported} imported, {skipped} skipped, {failed} failed", {
imported: harSummary.imported,
skipped: harSummary.skipped,
failed: harSummary.failed,
})}
</p>
{harSummary.truncated ? (
<p className="text-amber-600 dark:text-amber-400">
{t("HAR import stopped early at the 32 MiB decoded body cap.")}
</p>
) : null}
{harSummary.errors && harSummary.errors.length > 0 ? (
<ul className="list-disc pl-4 text-muted-foreground">
{harSummary.errors.slice(0, 5).map((item) => (
<li key={`${item.index}-${item.reason}`} className="font-mono">
#{item.index}: {item.reason}
</li>
))}
</ul>
) : null}
</AlertDescription>
<Button variant="ghost" size="icon-xs" aria-label={t("Dismiss")} onClick={() => setHarSummary(null)}>
<XIcon className="size-3.5" />
</Button>
</Alert>
) : null}
{/* Toolbar */}
<div className="flex flex-wrap items-center gap-2">
<Popover open={pickerOpen} onOpenChange={setPickerOpen}>
@@ -598,6 +764,16 @@ export default function TrafficPage() {
))}
</SelectContent>
</Select>
<Select value={rewrittenFilter} onValueChange={setRewrittenFilter}>
<SelectTrigger size="sm" className="w-32">
<SelectValue placeholder={t("Rewritten")} />
</SelectTrigger>
<SelectContent>
<SelectItem value="all">{t("All traffic")}</SelectItem>
<SelectItem value="rewritten">{t("Rewritten")}</SelectItem>
<SelectItem value="original">{t("Original")}</SelectItem>
</SelectContent>
</Select>
<div className="flex items-center gap-1 text-xs text-muted-foreground">
<span>{t("Response size")}</span>
<Input
@@ -721,7 +897,17 @@ export default function TrafficPage() {
<MethodBadge method={e.method} />
</TableCell>
<TableCell className="max-w-0">
<span className="block truncate font-mono text-xs">{e.url}</span>
<div className="flex items-center gap-1.5">
<span className="min-w-0 truncate font-mono text-xs">{e.url}</span>
{e.rewritten ? (
<Badge
variant="outline"
className="shrink-0 border-amber-500/30 bg-amber-500/10 px-1.5 py-0 text-[10px] font-medium text-amber-600 dark:text-amber-400"
>
{t("Rewritten")}
</Badge>
) : null}
</div>
</TableCell>
<TableCell>
<span className={cn("font-mono text-xs font-semibold tabular-nums", statusTone(e.status))}>
@@ -762,6 +948,14 @@ export default function TrafficPage() {
<Badge variant="secondary" className={cn("font-mono tabular-nums", statusTone(selected.status))}>
{selected.status}
</Badge>
{selected.rewritten ? (
<Badge
variant="outline"
className="border-amber-500/30 bg-amber-500/10 text-amber-600 dark:text-amber-400"
>
{t("Rewritten")}
</Badge>
) : null}
<span className="ml-auto text-xs text-muted-foreground tabular-nums">{fmtTime(selected.ts)}</span>
</div>
<SheetTitle className="break-all font-mono">{selected.host}</SheetTitle>
+18
View File
@@ -245,6 +245,7 @@
"All statuses": "Tüm durumlar",
"All tasks": "Tüm görevler",
"All tool execution records": "Tüm araç çalıştırma kayıtları",
"All traffic": "Tüm trafik",
"All types": "Tüm türler",
"all used": "tümü kullanıldı",
"All verification states": "Tüm doğrulama durumları",
@@ -275,6 +276,7 @@
"Appended {count} scopes": "{count} kapsam eklendi",
"Applies only to this profile, overriding the global defaults from \"Retry & backoff\". Each field left empty = follow the global; attempts -1 = disable this layer; a filled interval replaces the exponential backoff with a fixed one. The breaker and intent rerun are process-level and can only be tuned on the global page.": "Yalnızca bu profile uygulanır ve \"Yeniden deneme ve geri çekilme\" sekmesindeki genel varsayılanları geçersiz kılar. Boş bırakılan her alan = genel ayarı izle; deneme sayısı -1 = bu katmanı devre dışı bırak; doldurulan aralık üstel geri çekilmeyi sabit bir aralıkla değiştirir. Devre kesici ve niyet yeniden çalıştırma süreç düzeyindedir ve yalnızca genel sayfada ayarlanabilir.",
"Applies to this task only and is not written to global rules. Evaluation order: \"Block\" rules (including global ones) match first, and a hit forbids testing; with no hit and \"Allow\" rules configured for this task, an asset must match an allow rule to be permitted, otherwise testing is likewise forbidden; with no allow rules configured, no whitelist is active.": "Yalnızca bu göreve uygulanır ve genel kurallara yazılmaz. Değerlendirme sırası: önce \"Engelle\" kuralları (genel olanlar dahil) eşleştirilir ve eşleşme testi yasaklar; eşleşme yoksa ve bu görev için \"İzin ver\" kuralları yapılandırılmışsa, bir varlığa izin verilmesi için bir izin kuralıyla eşleşmesi gerekir, aksi halde test yine yasaktır; izin kuralı yapılandırılmamışsa beyaz liste etkin değildir.",
"Apply capture rewrite rules": "Yakalama yeniden yazma kurallarını uygula",
"Approval calls": "Onay çağrıları",
"Approval decision": "Onay kararı",
"Approval details #{id}": "Onay ayrıntıları #{id}",
@@ -765,6 +767,7 @@
"Discovery time": "Keşif zamanı",
"Discovery time currently ascending, click to toggle sort direction": "Keşif zamanı şu anda artan sırada, sıralama yönünü değiştirmek için tıklayın",
"Discovery time currently descending, click to toggle sort direction": "Keşif zamanı şu anda azalan sırada, sıralama yönünü değiştirmek için tıklayın",
"Dismiss": "Kapat",
"Dismiss update notice": "Güncelleme bildirimini kapat",
"dismissed": "yok sayıldı",
"Dismissed": "Yok sayıldı",
@@ -923,9 +926,13 @@
"Export failed: {error}": "Dışa aktarma başarısız: {error}",
"Export findings": "Bulguları dışa aktar",
"Export format": "Dışa aktarma biçimi",
"Export HAR": "HAR dışa aktar",
"Export scope": "Dışa aktarma kapsamı",
"Export the {count} selected items": "Seçili {count} öğeyi dışa aktar",
"Export the current filter results ({count} items)": "Mevcut süzgeç sonuçlarını dışa aktar ({count} öğe)",
"Exported {count} exchanges to HAR": "{count} değişim HAR olarak dışa aktarıldı",
"Exported {exported} of {matched} matching exchanges (truncated: {reason})": "Eşleşen {matched} değişimin {exported} tanesi dışa aktarıldı (kısaltıldı: {reason})",
"Exporting HAR…": "HAR dışa aktarılıyor…",
"Exporting…": "Dışa aktarılıyor…",
"External MCP tool servers · visibility granted per agent": "Harici MCP araç sunucuları · görünürlük ajan bazında verilir",
"Fact": "Olgu",
@@ -1068,6 +1075,13 @@
"greater than": "büyüktür",
"greater than or equal": "büyük veya eşittir",
"Handle automatically after the timeout instead of waiting": "Beklemek yerine zaman aşımından sonra otomatik olarak işle",
"HAR export failed: {error}": "HAR dışa aktarma başarısız: {error}",
"HAR export is not available in mock mode": "HAR dışa aktarma sahte modda kullanılamaz",
"HAR file exceeds the 32 MiB limit": "HAR dosyası 32 MiB sınırını aşıyor",
"HAR import complete: {imported} imported, {skipped} skipped, {failed} failed": "HAR içe aktarma tamamlandı: {imported} içe aktarıldı, {skipped} atlandı, {failed} başarısız",
"HAR import failed: {error}": "HAR içe aktarma başarısız: {error}",
"HAR import is not available in mock mode": "HAR içe aktarma sahte modda kullanılamaz",
"HAR import stopped early at the 32 MiB decoded body cap.": "HAR içe aktarma 32 MiB çözülmüş gövde sınırında erken durdu.",
"Hard delete": "Kalıcı silme",
"Header": "Başlık",
"Header name": "Başlık adı",
@@ -1110,6 +1124,8 @@
"Idle": "Boşta",
"Ignored": "Yok sayıldı",
"Implicit TLS": "Örtük TLS",
"Import HAR": "HAR içe aktar",
"Importing HAR…": "HAR içe aktarılıyor…",
"In": "Girdi",
"In {input} (incl. cache {cache}) · Out {output}": "Girdi {input} (önbellek {cache} dahil) · Çıktı {output}",
"In flight": "Devam eden",
@@ -1598,6 +1614,7 @@
"OR (either one matches)": "OR (herhangi biri eşleşir)",
"Ordered failover: workers share the current profile and switch after quota exhaustion or repeated provider failures.": "Sıralı yedeğe geçiş: çalışanlar geçerli profili paylaşır ve kota tükendiğinde veya sağlayıcı tekrar tekrar başarısız olduğunda geçiş yapar.",
"origin": "başlangıç",
"Original": "Orijinal",
"Original status": "Önceki durum",
"Other": "Diğer",
"Out": "Çıktı",
@@ -1904,6 +1921,7 @@
"Revoked Skill \"{name}\" visibility": "\"{name}\" becerisinin görünürlüğü kaldırıldı",
"Revoked visibility from \"{agent}\"": "\"{agent}\" için görünürlük kaldırıldı",
"Revoked visibility to \"{name}\"": "\"{name}\" için görünürlük kaldırıldı",
"Rewritten": "Yeniden yazılmış",
"Rewritten by capture rules": "Yakalama kuralları tarafından yeniden yazıldı",
"Rewritten by capture rules: {ids}": "Yakalama kuralları tarafından yeniden yazıldı: {ids}",
"Risk": "Risk",
+82 -3
View File
@@ -102,6 +102,8 @@ import type {
TrafficDetail,
TrafficEvidenceRef,
TrafficEvidenceRole,
TrafficHARExportMeta,
TrafficHARImportResult,
TrafficHost,
TrafficReplayRequest,
TrafficReplayResult,
@@ -840,6 +842,7 @@ export const api = {
body?: string;
path?: string;
status?: string;
rewritten?: string;
respMin?: string;
respMax?: string;
sort?: string;
@@ -854,6 +857,7 @@ export const api = {
(opts.body ? `&body=${encodeURIComponent(opts.body)}` : "") +
(opts.path ? `&path=${encodeURIComponent(opts.path)}` : "") +
(opts.status && opts.status !== "all" ? `&status=${encodeURIComponent(opts.status)}` : "") +
(opts.rewritten && opts.rewritten !== "all" ? `&rewritten=${encodeURIComponent(opts.rewritten)}` : "") +
(opts.respMin ? `&resp_min=${encodeURIComponent(opts.respMin)}` : "") +
(opts.respMax ? `&resp_max=${encodeURIComponent(opts.respMax)}` : "") +
(opts.sort ? `&sort=${encodeURIComponent(opts.sort)}` : "") +
@@ -874,11 +878,86 @@ export const api = {
if (!r.ok) throw new Error(t("Blob download failed: {error}", { error: r.status }));
return new Uint8Array(await r.arrayBuffer());
},
// Re-sends an edited request through the proxy. Errors carry the HTTP status
// (ApiError): 413 body too large, 502 proxy/upstream, 503 capture off, 504 timeout.
// Re-sends an edited request through the proxy (bypasses capture rewrite rules
// unless req.apply_rules is true). Errors carry the HTTP status (ApiError):
// 413 body too large, 502 proxy/upstream, 503 capture off, 504 timeout.
trafficReplay: async (req: TrafficReplayRequest) => {
if (MOCK) throw new Error(t("Replay is not available in mock mode"));
return post<TrafficReplayResult>(`/traffic/replay`, req);
return post<TrafficReplayResult>(`/traffic/replay`, {
...req,
apply_rules: Boolean(req.apply_rules),
});
},
// Imports a raw HAR 1.2 JSON document into the traffic store.
trafficImportHAR: async (rawJSON: string): Promise<TrafficHARImportResult> => {
if (MOCK) throw new Error(t("HAR import is not available in mock mode"));
return http<TrafficHARImportResult>(`/traffic/import/har`, {
method: "POST",
body: rawJSON,
});
},
// Exports filtered traffic as a HAR 1.2 file download and returns truncation metadata.
trafficExportHAR: async (
opts: {
host?: string;
method?: string;
q?: string;
body?: string;
path?: string;
status?: string;
rewritten?: string;
respMin?: string;
respMax?: string;
} = {},
): Promise<TrafficHARExportMeta> => {
if (MOCK) throw new Error(t("HAR export is not available in mock mode"));
const qs = new URLSearchParams();
if (opts.host) qs.set("host", opts.host);
if (opts.method && opts.method !== "all") qs.set("method", opts.method);
if (opts.q) qs.set("q", opts.q);
if (opts.body) qs.set("body", opts.body);
if (opts.path) qs.set("path", opts.path);
if (opts.status && opts.status !== "all") qs.set("status", opts.status);
if (opts.rewritten && opts.rewritten !== "all") qs.set("rewritten", opts.rewritten);
if (opts.respMin) qs.set("resp_min", opts.respMin);
if (opts.respMax) qs.set("resp_max", opts.respMax);
const suffix = qs.toString();
const token = getToken();
const r = await fetch(`/api/traffic/export/har${suffix ? `?${suffix}` : ""}`, {
headers: { ...(token ? { Authorization: `Bearer ${token}` } : {}) },
});
if (r.status === 401) {
if (typeof window !== "undefined") {
localStorage.removeItem("artex_token");
document.cookie = "artex_token=; path=/; max-age=0";
window.location.href = "/login";
}
throw new Error(t("Unauthorized"));
}
if (!r.ok) {
let msg = String(r.status);
try {
const payload = (await r.json()) as { error?: unknown };
if (typeof payload.error === "string" && payload.error.trim()) msg = payload.error.trim();
} catch {
// Keep status fallback.
}
throw new ApiError(t("HAR export failed: {error}", { error: msg }), r.status);
}
const blob = await r.blob();
const exported = Number(r.headers.get("X-ARTEX-HAR-Exported") ?? "0") || 0;
const matched = Number(r.headers.get("X-ARTEX-HAR-Matched") ?? "0") || 0;
const truncated = r.headers.get("X-ARTEX-HAR-Truncated") === "true";
const truncatedReason = r.headers.get("X-ARTEX-HAR-Truncated-Reason") ?? undefined;
const objectUrl = URL.createObjectURL(blob);
const a = document.createElement("a");
a.href = objectUrl;
a.download = "artex-traffic.har";
document.body.appendChild(a);
a.click();
a.remove();
URL.revokeObjectURL(objectUrl);
return { exported, matched, truncated, truncatedReason };
},
trafficHosts: () => get<{ hosts: TrafficHost[] }>(`/traffic/hosts`),
trafficDeleteHost: (host: string) => del<{ deleted: number }>(`/traffic?host=${encodeURIComponent(host)}`),
+33
View File
@@ -903,6 +903,36 @@ export interface TrafficExchange {
status: number;
content_type: string;
resp_len: number;
rewritten?: boolean;
}
export interface TrafficRewriteStats {
large_responses: number;
capacity_refusals: number;
tombstone_drops: number;
conversions: number;
sweeps: number;
}
export interface TrafficHARImportError {
index: number;
url?: string;
reason: string;
}
export interface TrafficHARImportResult {
imported: number;
skipped: number;
failed: number;
truncated?: boolean;
errors?: TrafficHARImportError[];
}
export interface TrafficHARExportMeta {
exported: number;
matched: number;
truncated: boolean;
truncatedReason?: string;
}
export interface TrafficResp {
@@ -913,6 +943,7 @@ export interface TrafficResp {
page?: number;
size?: number;
exchanges?: TrafficExchange[];
rewrite?: TrafficRewriteStats;
}
// Full raw request/response of one exchange (lazy-loaded on row select).
@@ -955,12 +986,14 @@ export interface TrafficReplayHeader {
}
// POST /api/traffic/replay. Host is not sent: the server takes it from url.
// apply_rules defaults to false (bypasses capture rewrite rules).
export interface TrafficReplayRequest {
method: string;
url: string;
headers: TrafficReplayHeader[];
body: string;
body_base64: boolean;
apply_rules?: boolean;
}
export interface TrafficReplayResult {