docs(06): create gap-closure plan 04 — list-failed-jobs CLI and web enqueue wiring

Closes SC-3 (WORK-04) with a list-failed-jobs subcommand in cmd/worker and
SC-4 (WORK-01) by wiring a river insert-only client into cmd/web with a
/debug/enqueue-test handler that proves the web→worker enqueue path.

Co-Authored-By: Claude Sonnet 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Arthur Belleville 2026-05-15 17:12:25 +02:00
parent 7a54755618
commit 8b64b48490
No known key found for this signature in database

View file

@ -0,0 +1,281 @@
---
phase: 06-background-worker
plan: 04
type: execute
wave: 2
depends_on: [06-03]
files_modified:
- backend/cmd/worker/main.go
- backend/internal/web/router.go
- backend/internal/web/handlers.go
- backend/cmd/web/main.go
autonomous: true
requirements: [WORK-01, WORK-04]
gap_closure: true
must_haves:
truths:
- "Running `go run ./cmd/worker list-failed-jobs` (with DATABASE_URL set) prints discarded river jobs from the river_job table (or empty table output if none exist)"
- "`go run ./cmd/worker` (no subcommand) continues to start the full worker runtime as before — subcommand dispatch does not break the default path"
- "GET /debug/enqueue-test (authenticated) inserts a HeartbeatArgs job and returns 200 with plain-text confirmation; the worker picks it up within 60 seconds"
- "cmd/web builds without error after adding the river insert-only client"
- "`go build ./...` exits 0 with all changes applied"
artifacts:
- path: "backend/cmd/worker/main.go"
provides: "list-failed-jobs subcommand branching on os.Args[1]"
contains: "list-failed-jobs"
- path: "backend/internal/web/handlers.go"
provides: "EnqueueTestHandler that inserts HeartbeatArgs via river.Client"
exports: ["EnqueueTestHandler"]
- path: "backend/internal/web/router.go"
provides: "/debug/enqueue-test route wired to EnqueueTestHandler"
contains: "/debug/enqueue-test"
- path: "backend/cmd/web/main.go"
provides: "river.Client (insert-only) constructed from pgxpool and passed into router deps"
contains: "river.NewClient"
key_links:
- from: "backend/cmd/worker/main.go"
to: "river_job table"
via: "pgx query SELECT id, kind, state, attempt, max_attempts, errors, finalized_at FROM river_job WHERE state = 'discarded'"
pattern: "list-failed-jobs"
- from: "backend/cmd/web/main.go"
to: "backend/internal/web/handlers.go"
via: "river.Client passed through DebugDeps struct to EnqueueTestHandler"
pattern: "river\\.Client"
- from: "backend/internal/web/handlers.go"
to: "river_job table"
via: "riverClient.Insert(ctx, jobs.HeartbeatArgs{}, nil)"
pattern: "Insert.*HeartbeatArgs"
---
<objective>
Close two Phase 6 roadmap success-criteria gaps that Phase 6 plans 01-03 did not implement:
1. SC-3 (WORK-04): Add a `list-failed-jobs` subcommand to `cmd/worker` that queries `river_job WHERE state = 'discarded'` and prints results to stdout. This provides the "simple CLI surface" the ROADMAP success criterion requires.
2. SC-4 (WORK-01): Wire an insert-only `river.Client` into `cmd/web` and add one debug HTTP handler (`GET /debug/enqueue-test`) that enqueues a `HeartbeatArgs` job. This proves the web→worker job dispatch path end-to-end.
Purpose: Both gaps represent unproven integration contracts — the failed-job surface was only observable via logs, and the web-side enqueue path was never wired at all.
Output:
- Modified `backend/cmd/worker/main.go` with subcommand dispatch
- New `EnqueueTestHandler` in `backend/internal/web/handlers.go`
- Modified `backend/internal/web/router.go` with `/debug/enqueue-test` route
- Modified `backend/cmd/web/main.go` with river client construction and DebugDeps wiring
</objective>
<execution_context>
@/Users/arthur.belleville/Documents/perso/projects/xtablo-source/.claude/get-shit-done/workflows/execute-plan.md
@/Users/arthur.belleville/Documents/perso/projects/xtablo-source/.claude/get-shit-done/templates/summary.md
</execution_context>
<context>
@/Users/arthur.belleville/Documents/perso/projects/xtablo-source/.planning/PROJECT.md
@/Users/arthur.belleville/Documents/perso/projects/xtablo-source/.planning/ROADMAP.md
@/Users/arthur.belleville/Documents/perso/projects/xtablo-source/.planning/phases/06-background-worker/06-CONTEXT.md
@/Users/arthur.belleville/Documents/perso/projects/xtablo-source/.planning/phases/06-background-worker/06-VERIFICATION.md
<interfaces>
<!-- Key types the executor needs. Extracted from codebase. -->
From backend/internal/jobs/heartbeat.go (verified in VERIFICATION.md):
HeartbeatArgs implements river.JobArgs with Kind() string = "heartbeat"
From backend/internal/web/router.go:
func NewRouter(pinger Pinger, staticDir string, deps AuthDeps, tabloDeps TablosDeps,
taskDeps TasksDeps, fileDeps FilesDeps, csrfKey []byte, env string,
trustedOrigins ...string) http.Handler
From backend/cmd/web/main.go current signature (line 116):
router := web.NewRouter(pool, "./static", deps, tabloDeps, taskDeps, fileDeps, csrfKey, env)
From backend/go.mod (lines 42-46) — river is already an indirect dependency:
github.com/riverqueue/river v0.37.0 // indirect
github.com/riverqueue/river/riverdriver/riverpgxv5 v0.37.0 // indirect
From backend/cmd/worker/main.go — river.NewClient call for reference:
riverClient, err := river.NewClient(riverpgxv5.New(pool), &river.Config{
Logger: slog.Default(),
Workers: workers,
Queues: map[string]river.QueueConfig{river.QueueDefault: {MaxWorkers: 10}},
...
})
// Insert-only client: pass &river.Config{} with NO Workers field (nil workers = insert-only mode)
</interfaces>
</context>
<tasks>
<task type="auto">
<name>Task 1: Add list-failed-jobs subcommand to cmd/worker</name>
<files>backend/cmd/worker/main.go</files>
<read_first>
- backend/cmd/worker/main.go (full file — understand current main() structure before touching)
- backend/go.mod (confirm river and pgx imports available)
</read_first>
<action>
Modify backend/cmd/worker/main.go to dispatch on os.Args[1] before the full worker startup sequence. The subcommand check must be the FIRST thing in main() after setting up the logger — before pool creation, before river migrations, before the signal context.
Concrete dispatch logic: if len(os.Args) > 1 and os.Args[1] == "list-failed-jobs", run a separate listFailedJobs(dsn string) function and os.Exit(0). All other argument patterns (including no arguments) fall through to the existing worker startup sequence unchanged.
The listFailedJobs function must:
1. Open a pgxpool using db.NewPool(context.Background(), dsn). If dsn is empty, log "DATABASE_URL is required but unset" and os.Exit(1).
2. Execute the raw query: SELECT id, kind, state, attempt, max_attempts, errors, finalized_at FROM river_job WHERE state = 'discarded' ORDER BY finalized_at DESC LIMIT 100 — use pool.Query(ctx, query) directly (no sqlc generation needed for this one-off).
3. For each row, print one line to stdout in the format: id=<uuid> kind=<kind> attempt=<n>/<max> finalized_at=<RFC3339>. Use fmt.Printf.
4. If the query returns zero rows, print "no discarded jobs found" and exit 0.
5. Close the pool before returning.
Import requirements: the pgx rows.Scan must capture id (pgx scans uuid.UUID or [16]byte), kind (string), state (string), attempt (int), max_attempts (int), errors ([]byte or pgtype.Text — skip/ignore if scan is complex, leave as nil), finalized_at (pgtype.Timestamptz or time.Time pointer). Use pgx.CollectRows or rows.Next loop — match the pattern already used in backend/internal/db/sqlc/ for pgx v5 row scanning.
Do NOT restructure the existing main() startup sequence. The subcommand branch is a pure prefix check; if the branch is not taken, execution falls through to the existing code verbatim.
After adding the listFailedJobs function and the dispatch, run `cd backend && go build ./cmd/worker` to confirm it compiles. Fix any import or type errors before declaring done.
</action>
<verify>
<automated>cd /Users/arthur.belleville/Documents/perso/projects/xtablo-source/backend && go build ./cmd/worker && echo "build ok"</automated>
</verify>
<acceptance_criteria>
- `go build ./cmd/worker` exits 0 with the modified file
- `grep "list-failed-jobs" backend/cmd/worker/main.go` returns at least one match
- `grep "listFailedJobs" backend/cmd/worker/main.go` returns at least two matches (declaration + call site)
- `grep "os.Args" backend/cmd/worker/main.go` returns the dispatch check
- The dispatch check appears before any pool creation code — grep line numbers must show os.Args check before `db.NewPool`
- Running `go run ./cmd/worker list-failed-jobs` with DATABASE_URL unset prints "DATABASE_URL is required but unset" and exits non-zero (the early exit guard applies inside listFailedJobs too)
</acceptance_criteria>
<done>cmd/worker compiles, contains a list-failed-jobs subcommand that queries river_job WHERE state = 'discarded', and the default startup path is unchanged</done>
</task>
<task type="auto">
<name>Task 2: Wire river insert-only client into cmd/web and add /debug/enqueue-test handler</name>
<files>
backend/internal/web/handlers.go
backend/internal/web/router.go
backend/cmd/web/main.go
</files>
<read_first>
- backend/cmd/web/main.go (full file — understand existing deps wiring before adding river)
- backend/internal/web/router.go (full file — understand NewRouter signature and route groups)
- backend/internal/web/handlers.go (first 50 lines — understand handler pattern to match)
- backend/internal/jobs/heartbeat.go (confirm HeartbeatArgs type and import path)
</read_first>
<action>
Three coordinated changes. Apply them in order: handlers first, then router, then cmd/web.
STEP A — backend/internal/web/handlers.go:
Add a new DebugDeps struct and EnqueueTestHandler at the bottom of the file. DebugDeps has one field: RiverClient of type RiverInserter, where RiverInserter is a local interface defined in handlers.go:
type RiverInserter interface {
Insert(ctx context.Context, args river.JobArgs, opts *river.InsertOpts) (*river.InsertResult, error)
}
The interface approach means the handlers package does not need to import riverqueue packages directly in the non-test path — only cmd/web imports river for construction.
However, the river import IS needed for river.InsertOpts and river.InsertResult in the interface. Add import "github.com/riverqueue/river" to handlers.go imports.
Also add import "backend/internal/jobs" for jobs.HeartbeatArgs.
EnqueueTestHandler returns an http.HandlerFunc. It calls debugDeps.RiverClient.Insert(r.Context(), jobs.HeartbeatArgs{}, nil). On success, write HTTP 200 with plain text body "enqueued heartbeat job id=<result.Job.ID>". On error, write HTTP 500 with plain text body "enqueue failed: <err>". Do not use a template — plain w.Write() is correct here.
STEP B — backend/internal/web/router.go:
Add DebugDeps as a new parameter to NewRouter after fileDeps: `debugDeps DebugDeps`. Inside the router, add the route OUTSIDE the RequireAuth group (so it is accessible for integration testing without a session cookie — the handler is debug-only, not a production surface):
r.Get("/debug/enqueue-test", EnqueueTestHandler(debugDeps))
This route goes before the final healthz/demo/static block, after the RequireAuth group closing brace.
STEP C — backend/cmd/web/main.go:
Construct a river insert-only client after the pool is created but BEFORE the signal context (follow the same startup ordering as cmd/worker — river client construction is startup I/O). An insert-only client uses river.NewClient with a Config that has NO Workers field set (nil workers is the insert-only pattern per river docs):
riverInsertClient, err := river.NewClient(riverpgxv5.New(pool), &river.Config{})
if err != nil {
slog.Error("river insert client init failed", "err", err)
os.Exit(1)
}
Required imports to add to cmd/web/main.go:
"github.com/riverqueue/river"
"github.com/riverqueue/river/riverdriver/riverpgxv5"
Create the debugDeps variable:
debugDeps := web.DebugDeps{RiverClient: riverInsertClient}
Update the NewRouter call to pass debugDeps as the new final argument (before csrfKey):
router := web.NewRouter(pool, "./static", deps, tabloDeps, taskDeps, fileDeps, debugDeps, csrfKey, env)
Note: NewRouter signature after Step B is:
NewRouter(pinger, staticDir, deps, tabloDeps, taskDeps, fileDeps, debugDeps, csrfKey, env, ...trustedOrigins)
After all three changes, run `cd backend && go build ./...` to confirm the full module builds. Fix any import, type, or signature mismatches before declaring done.
The river packages are already present in go.mod as indirect dependencies (lines 42-46). Promoting them to direct imports will cause `go mod tidy` to update the indirect comment — that is expected and acceptable. Run `cd backend && go mod tidy` after the build succeeds.
</action>
<verify>
<automated>cd /Users/arthur.belleville/Documents/perso/projects/xtablo-source/backend && go build ./... && echo "full build ok"</automated>
</verify>
<acceptance_criteria>
- `go build ./...` exits 0 with all three files modified
- `grep "EnqueueTestHandler" backend/internal/web/handlers.go` returns a function declaration
- `grep "RiverInserter" backend/internal/web/handlers.go` returns the interface definition
- `grep "/debug/enqueue-test" backend/internal/web/router.go` returns the route registration
- `grep "river.NewClient" backend/cmd/web/main.go` returns the insert-only client construction
- `grep "riverpgxv5" backend/cmd/web/main.go` returns the import usage
- `grep "debugDeps" backend/cmd/web/main.go` returns at least two lines (construction + NewRouter call)
- `grep -r "river" backend/internal/web/handlers.go` returns imports and usage lines
- `go test ./internal/web/... -count=1` exits 0 — existing tests must not regress
</acceptance_criteria>
<done>cmd/web builds with river insert client; GET /debug/enqueue-test is registered; the web→worker enqueue path is provably wired</done>
</task>
</tasks>
<threat_model>
## Trust Boundaries
| Boundary | Description |
|----------|-------------|
| HTTP → EnqueueTestHandler | Unauthenticated HTTP request triggers job insertion into the database |
## STRIDE Threat Register
| Threat ID | Category | Component | Disposition | Mitigation Plan |
|-----------|----------|-----------|-------------|-----------------|
| T-06-04-01 | Elevation of Privilege | GET /debug/enqueue-test | accept | Debug-only endpoint; inserts only HeartbeatArgs which has no side effects beyond a log line. No user data accessed. Route is explicitly scoped to development use — a future hardening phase can gate it behind RequireAuth or remove it. |
| T-06-04-02 | Denial of Service | GET /debug/enqueue-test | accept | Flooding the endpoint would insert many heartbeat jobs. River's queue has MaxWorkers:10 and heartbeat Work() is a no-op log call. No meaningful amplification. Acceptable risk for a debug route. |
| T-06-04-03 | Information Disclosure | list-failed-jobs CLI output | mitigate | The CLI runs server-side only (not an HTTP surface). Output goes to stdout of an operator-controlled terminal. river_job errors column may contain job argument values — confirm HeartbeatArgs has no PII (it has no fields). OrphanCleanupArgs similarly has no user-identifying fields. Risk is low but operator should be aware output is unredacted. |
</threat_model>
<verification>
After both tasks complete, run the following to confirm both gaps are closed:
```bash
# 1. Full module builds
cd /Users/arthur.belleville/Documents/perso/projects/xtablo-source/backend && go build ./... && echo "BUILD OK"
# 2. All existing tests still pass
cd /Users/arthur.belleville/Documents/perso/projects/xtablo-source/backend && go test ./internal/jobs/... ./internal/web/... -count=1 -v 2>&1 | tail -20
# 3. SC-3 gap: list-failed-jobs subcommand exists
grep -n "list-failed-jobs" /Users/arthur.belleville/Documents/perso/projects/xtablo-source/backend/cmd/worker/main.go
# 4. SC-4 gap: river wired into cmd/web
grep -n "river" /Users/arthur.belleville/Documents/perso/projects/xtablo-source/backend/cmd/web/main.go
# 5. SC-4 gap: debug route registered
grep -n "/debug/enqueue-test" /Users/arthur.belleville/Documents/perso/projects/xtablo-source/backend/internal/web/router.go
# 6. go mod tidy does not fail
cd /Users/arthur.belleville/Documents/perso/projects/xtablo-source/backend && go mod tidy && echo "MOD TIDY OK"
```
</verification>
<success_criteria>
- SC-3 closed: `go run ./cmd/worker list-failed-jobs` is a valid invocation that queries river_job for discarded jobs and prints results (or "no discarded jobs found"). The default `go run ./cmd/worker` startup path is unaffected.
- SC-4 closed: cmd/web constructs a river insert-only client, passes it through DebugDeps, and registers GET /debug/enqueue-test which calls riverClient.Insert(ctx, HeartbeatArgs{}, nil).
- `go build ./...` exits 0 for the entire backend module.
- `go test ./internal/jobs/... ./internal/web/... -count=1` exits 0 — no regressions.
- WORK-01 and WORK-04 requirements are now fully satisfied.
</success_criteria>
<output>
After completion, create `.planning/phases/06-background-worker/06-background-worker-04-SUMMARY.md`
</output>