Files
ARTEX/server/task_archive_remote_test.go
T

397 lines
12 KiB
Go

package server
import (
"bytes"
"context"
"errors"
"fmt"
"io"
"net/http"
"net/http/httptest"
"os"
"path/filepath"
"strconv"
"strings"
"sync"
"sync/atomic"
"testing"
"github.com/Autumn-27/artex/archivestore"
pgdb "github.com/Autumn-27/artex/db"
)
func TestRemoteArchiveLifecycle(t *testing.T) {
if os.Getenv("ARTEX_PG_DSN") == "" {
t.Skip("requires isolated PostgreSQL")
}
t.Setenv("ARTEX_ARCHIVE_STORAGE", "local")
m, err := NewManager(t.TempDir(), "")
if err != nil {
t.Fatal(err)
}
defer m.Close()
ctx, cancel := context.WithCancel(context.Background())
cancel()
s := New(ctx, m, t.TempDir(), t.TempDir(), t.TempDir())
s.archiveWG.Wait()
s.ctx = context.Background()
var mu sync.Mutex
objects := map[string][]byte{}
corrupt, failDelete := false, false
remote := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
mu.Lock()
defer mu.Unlock()
switch r.Method {
case "PUT":
objects[r.URL.Path], _ = io.ReadAll(r.Body)
w.Header().Set("ETag", `"test"`)
case "HEAD":
w.Header().Set("ETag", `"test"`)
w.Header().Set("Content-Length", strconv.Itoa(len(objects[r.URL.Path])))
case "GET":
data := objects[r.URL.Path]
if corrupt && len(data) > 0 {
data = bytes.Clone(data)
data[0] ^= 0xff
}
var start, end int64
if n, err := fmt.Sscanf(r.Header.Get("Range"), "bytes=%d-%d", &start, &end); err != nil || n != 2 || start < 0 || end < start || end >= int64(len(data)) || r.Header.Get("If-Match") != `"test"` {
w.WriteHeader(http.StatusRequestedRangeNotSatisfiable)
return
}
w.Header().Set("ETag", `"test"`)
w.Header().Set("Content-Length", strconv.FormatInt(end-start+1, 10))
w.Header().Set("Content-Range", fmt.Sprintf("bytes %d-%d/%d", start, end, len(data)))
w.WriteHeader(http.StatusPartialContent)
_, _ = w.Write(data[start : end+1])
case "DELETE":
if failDelete {
w.WriteHeader(403)
return
}
delete(objects, r.URL.Path)
w.WriteHeader(204)
default:
w.WriteHeader(400)
}
}))
defer remote.Close()
s.archiveStore, err = archivestore.New(archivestore.Config{Endpoint: remote.URL, Bucket: "test", Prefix: "tasks", AccessKey: "test", SecretKey: "test", PathStyle: true})
if err != nil {
t.Fatal(err)
}
task, err := m.CreateTask("S3 round trip", "preserve data", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
id, _ := strconv.ParseInt(task.ID, 10, 64)
defer func() { _, _ = m.pg.Exec(`DELETE FROM task_archives WHERE task_id=$1`, id); _ = m.pg.DeleteTask(id) }()
if err := m.pg.SetPaused(id, true); err != nil {
t.Fatal(err)
}
file := filepath.Join(m.dir, "tasks", task.ID, "workspace", "important.txt")
if err := os.MkdirAll(filepath.Dir(file), 0700); err != nil {
t.Fatal(err)
}
if err := os.WriteFile(file, []byte("keep this content"), 0600); err != nil {
t.Fatal(err)
}
pgTask, err := m.pg.GetTask(id)
if err != nil {
t.Fatal(err)
}
if _, err := m.pg.Exploration(pgTask.ExplorationID).AppendActivity(pgdb.Activity{Worker: "system", Kind: "system", Summary: "archive proof"}); err != nil {
t.Fatal(err)
}
job, err := m.pg.QueueTaskArchive(id)
if err != nil {
t.Fatal(err)
}
// A configured-but-invalid backend must fail before moving any source data.
s.archiveStoreErr = errors.New("invalid fixture configuration")
if err := s.runOneTaskArchiveJob(); !errors.Is(err, s.archiveStoreErr) {
t.Fatalf("configuration error was not returned: %v", err)
}
if got, err := os.ReadFile(file); err != nil || string(got) != "keep this content" {
t.Fatalf("bad configuration lost source: %v", err)
}
if pgTask, err := m.pg.GetTask(id); err != nil || pgTask == nil {
t.Fatal("bad configuration lost task", err)
}
s.archiveStoreErr = nil
if _, err := m.pg.QueueTaskArchive(id); err != nil {
t.Fatal(err)
}
// A corrupt readback must roll back filesystem staging and leave PG intact.
mu.Lock()
corrupt = true
mu.Unlock()
if err := s.runOneTaskArchiveJob(); err == nil {
t.Fatal("expected integrity failure")
}
if got, _ := os.ReadFile(file); string(got) != "keep this content" {
t.Fatal("upload failure lost source")
}
if task, err := m.pg.GetTask(id); err != nil || task == nil {
t.Fatal("upload failure lost task")
}
mu.Lock()
corrupt = false
mu.Unlock()
if err := s.recoverRemoteArchiveJournals(); err != nil {
t.Fatal(err)
}
if _, err := m.pg.QueueTaskArchive(id); err != nil {
t.Fatal(err)
}
if err := s.runOneTaskArchiveJob(); err != nil {
t.Fatal(err)
}
job, err = m.pg.GetTaskArchive(job.ID)
if err != nil {
t.Fatal(err)
}
if job.State != pgdb.ArchiveReady || job.Storage != "s3" || !strings.HasPrefix(job.ArchivePath, "s3://") {
t.Fatalf("unexpected archive %+v", job)
}
if _, err := os.Stat(file); !os.IsNotExist(err) {
t.Fatal("workspace not released")
}
if _, err := os.Stat(taskArchivePath(m.dir, job.ID, task.ID)); !os.IsNotExist(err) {
t.Fatal("local package retained")
}
// Simulate the crash window after commit: cleanup keeps the remote object.
journal, err := s.writeRemoteArchiveJournal(remoteArchiveJournal{ArchiveID: job.ID, URI: job.ArchivePath, LocalPath: taskArchivePath(m.dir, job.ID, task.ID), Operation: "upload"})
if err != nil {
t.Fatal(err)
}
if err := s.recoverRemoteArchiveJournal(journal); err != nil {
t.Fatal(err)
}
// A corrupted download must leave the archive and its remote package intact.
mu.Lock()
corrupt = true
mu.Unlock()
if _, err := m.pg.QueueTaskArchiveRestore(job.ID); err != nil {
t.Fatal(err)
}
if err := s.runOneTaskArchiveJob(); err == nil {
t.Fatal("expected corrupted restore to fail")
}
if _, err := os.Stat(file); !os.IsNotExist(err) {
t.Fatal("corrupted archive installed workspace files")
}
if restored, err := m.pg.IsTaskArchiveRestored(job.ID); err != nil || restored {
t.Fatalf("corrupted archive restored database: %v %v", restored, err)
}
mu.Lock()
if len(objects) == 0 {
t.Error("corrupted restore deleted the only remote package")
}
corrupt = false
mu.Unlock()
// Restoration commits files and PG first. Failed remote deletion is retryable.
mu.Lock()
failDelete = true
mu.Unlock()
if _, err := m.pg.QueueTaskArchiveRestore(job.ID); err != nil {
t.Fatal(err)
}
if err := s.runOneTaskArchiveJob(); err == nil {
t.Fatal("expected remote delete failure")
}
if got, _ := os.ReadFile(file); string(got) != "keep this content" {
t.Fatal("restored content differs")
}
if restored, err := m.pg.IsTaskArchiveRestored(job.ID); err != nil || !restored {
t.Fatalf("PG restore %v %v", restored, err)
}
mu.Lock()
failDelete = false
mu.Unlock()
if _, err := m.pg.QueueTaskArchiveRestore(job.ID); err != nil {
t.Fatal(err)
}
if err := s.runOneTaskArchiveJob(); err != nil {
t.Fatal(err)
}
if item, err := m.pg.GetTaskArchive(job.ID); err != nil || item != nil {
t.Fatal("restore did not consume stub")
}
var count int
if err := m.pg.QueryRow(`SELECT count(*) FROM activity WHERE exploration_id=$1 AND summary='archive proof'`, pgTask.ExplorationID).Scan(&count); err != nil || count != 1 {
t.Fatalf("activity restore count=%d err=%v", count, err)
}
// Re-archive, then crash/retry the deletion outbox after PG removed its stub.
job, err = m.pg.QueueTaskArchive(id)
if err != nil {
t.Fatal(err)
}
if err := s.runOneTaskArchiveJob(); err != nil {
t.Fatal(err)
}
mu.Lock()
failDelete = true
mu.Unlock()
if _, err := m.pg.QueueTaskArchiveDelete(job.ID); err != nil {
t.Fatal(err)
}
if err := s.runOneTaskArchiveJob(); err != nil {
t.Fatal(err)
}
mu.Lock()
failDelete = false
mu.Unlock()
if err := s.recoverRemoteArchiveJournals(); err != nil {
t.Fatal(err)
}
mu.Lock()
remaining := len(objects)
mu.Unlock()
if remaining != 0 {
t.Fatalf("left %d objects", remaining)
}
}
// Once PostgreSQL has committed an archive location and checksum, a subsequent
// queued operation must not make startup recovery resurrect the old workspace.
func TestRemoteArchiveStageRecoveryAfterCommit(t *testing.T) {
if os.Getenv("ARTEX_PG_DSN") == "" {
t.Skip("requires isolated PostgreSQL")
}
m, err := NewManager(t.TempDir(), "")
if err != nil {
t.Fatal(err)
}
defer m.Close()
task, err := m.CreateTask("remote stage recovery", "test", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
taskID, _ := strconv.ParseInt(task.ID, 10, 64)
defer func() {
_, _ = m.pg.Exec("DELETE FROM task_archives WHERE task_id=$1", taskID)
_ = m.pg.DeleteTask(taskID)
}()
if err := m.pg.SetPaused(taskID, true); err != nil {
t.Fatal(err)
}
job, err := m.pg.QueueTaskArchive(taskID)
if err != nil {
t.Fatal(err)
}
pgTask, err := m.pg.GetTask(taskID)
if err != nil {
t.Fatal(err)
}
for _, state := range []string{pgdb.ArchiveReady, pgdb.RestoreQueued, pgdb.DeleteQueued, pgdb.ArchiveQueued} {
t.Run(state, func(t *testing.T) {
committed := state != pgdb.ArchiveQueued
checksum := ""
if committed {
checksum = strings.Repeat("a", 64)
}
if _, err := m.pg.Exec("UPDATE task_archives SET state=$2,archive_path=$3,sha256=$4 WHERE id=$1", job.ID, state, "s3://test/tasks/package.tar.zst", checksum); err != nil {
t.Fatal(err)
}
file := filepath.Join(m.dir, "tasks", task.ID, "workspace", "remaining.txt")
if err := os.MkdirAll(filepath.Dir(file), 0700); err != nil {
t.Fatal(err)
}
if err := os.WriteFile(file, []byte("retain until commit"), 0600); err != nil {
t.Fatal(err)
}
stage, err := stageTaskArchiveFiles(m.dir, job.ID, task.ID, pgTask.ExplorationID)
if err != nil {
t.Fatal(err)
}
if err := recoverTaskArchiveStages(m.dir, m.pg); err != nil {
t.Fatal(err)
}
if _, err := os.Stat(stage.root); !os.IsNotExist(err) {
t.Fatalf("staging cleanup incomplete: %v", err)
}
if committed {
if _, err := os.Stat(file); !os.IsNotExist(err) {
t.Fatalf("resurrected committed workspace: %v", err)
}
} else if data, err := os.ReadFile(file); err != nil || string(data) != "retain until commit" {
t.Fatalf("uncommitted archive lost source: %v", err)
}
})
}
}
// Failed remote cleanup can survive switching to local storage and retrying the
// same archive ID. Its journal must never remove the newly committed local copy.
func TestRemoteArchiveJournalPreservesLocalRetry(t *testing.T) {
if os.Getenv("ARTEX_PG_DSN") == "" {
t.Skip("requires isolated PostgreSQL")
}
m, err := NewManager(t.TempDir(), "")
if err != nil {
t.Fatal(err)
}
defer m.Close()
task, err := m.CreateTask("local retry", "test", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
taskID, _ := strconv.ParseInt(task.ID, 10, 64)
defer func() {
_, _ = m.pg.Exec("DELETE FROM task_archives WHERE task_id=$1", taskID)
_ = m.pg.DeleteTask(taskID)
}()
if err := m.pg.SetPaused(taskID, true); err != nil {
t.Fatal(err)
}
job, err := m.pg.QueueTaskArchive(taskID)
if err != nil {
t.Fatal(err)
}
filename := taskArchivePath(m.dir, job.ID, task.ID)
if err := os.MkdirAll(filepath.Dir(filename), 0700); err != nil {
t.Fatal(err)
}
if err := os.WriteFile(filename, []byte("committed local archive"), 0600); err != nil {
t.Fatal(err)
}
if _, err := m.pg.Exec("UPDATE task_archives SET state=$2,archive_path=$3,sha256=$4 WHERE id=$1", job.ID, pgdb.ArchiveReady, filename, strings.Repeat("a", 64)); err != nil {
t.Fatal(err)
}
var deletes atomic.Int32
remote := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if r.Method != http.MethodDelete {
w.WriteHeader(http.StatusMethodNotAllowed)
return
}
deletes.Add(1)
w.WriteHeader(http.StatusNoContent)
}))
defer remote.Close()
store, err := archivestore.New(archivestore.Config{Endpoint: remote.URL, Bucket: "test", Prefix: "tasks", AccessKey: "test", SecretKey: "test", PathStyle: true})
if err != nil {
t.Fatal(err)
}
s := &Server{ctx: context.Background(), m: m, archiveStore: store}
journal, err := s.writeRemoteArchiveJournal(remoteArchiveJournal{
ArchiveID: job.ID, URI: store.URI(taskID, job.ID, strings.Repeat("b", 64)),
LocalPath: filename, Operation: "upload",
})
if err != nil {
t.Fatal(err)
}
if err := s.recoverRemoteArchiveJournal(journal); err != nil {
t.Fatal(err)
}
if data, err := os.ReadFile(filename); err != nil || string(data) != "committed local archive" {
t.Fatalf("old upload cleanup removed local retry: %v", err)
}
if deletes.Load() != 1 {
t.Fatal("uncommitted remote object was not removed")
}
if _, err := os.Stat(journal); !os.IsNotExist(err) {
t.Fatalf("finished journal remains: %v", err)
}
}