fix(redis-worker): stop fair queue leaking concurrency slots - #4540
fix(redis-worker): stop fair queue leaking concurrency slots#4540matt-aitken wants to merge 10 commits into
Conversation
A slot was only released when the message's in-flight record could still be read, and the release ran after that record had already been deleted. Any failure in between left the slot held with nothing able to reclaim it. Once a tenant leaked its whole limit, every queue it owned stalled for good.
🦋 Changeset detectedLatest commit: 72947fe The changes in this PR will be included in the next version bump. This PR includes changesets to release 27 packages
Not sure what this means? Click here to learn what changesets are. Click here if you're a maintainer who wants to add another changeset to this PR |
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
WalkthroughFair queue consumers now release concurrency slots before completing, requeueing, retrying, or moving messages to the DLQ. Fallback queue metadata supports cleanup when in-flight records are missing or invalid. Timed-out reclamation releases slots per message and removes dangling in-flight entries. Regression tests cover completion, failure, reclaim, and timeout cleanup paths. Batch queues now use configurable visibility timeouts and heartbeat long-running callbacks. A patch changeset documents the fair queue fix. 🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
…lease path Concurrency groups can key on queue metadata, not just the tenant. The fallback descriptor dropped metadata and skipped the descriptor cache, so those groups released against the wrong Redis set and leaked exactly as before. Prefer the cached descriptor in both branches, and move the retry path's release ahead of the re-queue so a redelivery cannot have its fresh reservation deleted by the previous holder.
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
@trigger.dev/build
trigger.dev
@trigger.dev/core
@trigger.dev/python
@trigger.dev/react-hooks
@trigger.dev/redis-worker
@trigger.dev/rsc
@trigger.dev/schema-to-json
@trigger.dev/sdk
commit: |
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
… queue Four paths could strand a slot with nothing left to reclaim it. Reclaim released the slot only after requeuing, and only for messages that survived the requeue, so a message whose requeue threw kept its slot forever. It now releases before the message becomes claimable, which also stops a redelivery having its fresh reservation deleted by the previous holder. The dead-letter branch completed the message before releasing, so anything throwing in between stranded the slot. Release now happens first. failMessage returned early when the stored message was missing or unparseable without releasing, completing or requeuing, leaving the slot held. The requeue script returned without removing the in-flight entry when the payload was gone, so the member was rescanned on every reclaim tick forever. Enough of them fill the scan window and starve a whole shard's reclaim. reclaimTimedOut takes an optional pre-requeue hook; no existing caller changes.
There was a problem hiding this comment.
🧹 Nitpick comments (2)
packages/redis-worker/src/fair-queue/index.ts (2)
1302-1311: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low valueConsider extracting the repeated descriptor construction.
Four sites now build the same fallback
QueueDescriptor:completeMessage(Lines 1249-1253),releaseMessage(Lines 1302-1306),#releaseOrphanedConcurrency(Lines 1346-1350), andfailMessage(Lines 1389-1393). The precedence rule is identical: cache first, then stored-message tenant and metadata, thenextractTenantId(queueId). A single private helper keeps that rule in one place and prevents the sites from drifting apart.The release-before-requeue ordering itself is correct.
♻️ Suggested helper
`#resolveDescriptor`( queueId: string, storedMessage?: { tenantId: string; metadata?: Record<string, unknown> } | null ): QueueDescriptor { return ( this.queueDescriptorCache.get(queueId) ?? { id: queueId, tenantId: storedMessage?.tenantId ?? this.keys.extractTenantId(queueId), metadata: storedMessage?.metadata ?? {}, } ); }
1547-1574: 🧹 Nitpick | 🔵 TrivialConsider a metric for the swallowed release failure.
The catch block logs and continues, which is the right choice here. One failure must not stop the shard scan. However, when the release fails, the message is still requeued and the concurrency slot stays held. That is a silent slot leak that only appears in logs.
A counter incremented in the catch block would make the condition alertable. Keep the attributes low cardinality, for example a bounded outcome label only, and do not attach
messageId,queueId, ortenantId.As per coding guidelines, OTEL metric attributes must use only enums, booleans, bounded error codes, or bounded shard IDs, and must not use IDs such as
runIdorprojectId.Source: Coding guidelines
ℹ️ Review info
⚙️ Run configuration
Configuration used: Repository UI
Review profile: CHILL
Plan: Pro Plus
Run ID: b6d6bee1-d525-4333-866f-fadfa29629b4
📒 Files selected for processing (4)
packages/redis-worker/src/fair-queue/index.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.tspackages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/visibility.ts
📜 Review details
⏰ Context from checks skipped due to timeout. (21)
- GitHub Check: webapp / 🧪 Unit Tests: Webapp (10, 12)
- GitHub Check: webapp / 🧪 Unit Tests: Webapp (8, 12)
- GitHub Check: webapp / 🧪 Unit Tests: Webapp (12, 12)
- GitHub Check: webapp / 🧪 Unit Tests: Webapp (6, 12)
- GitHub Check: webapp / 🧪 Unit Tests: Webapp (5, 12)
- GitHub Check: webapp / 🧪 Unit Tests: Webapp (9, 12)
- GitHub Check: webapp / 🧪 Unit Tests: Webapp (1, 12)
- GitHub Check: webapp / 🧪 Unit Tests: Webapp (11, 12)
- GitHub Check: webapp / 🧪 Unit Tests: Webapp (4, 12)
- GitHub Check: webapp / 🧪 Unit Tests: Webapp (3, 12)
- GitHub Check: webapp / 🧪 Unit Tests: Webapp (7, 12)
- GitHub Check: webapp / 🧪 Unit Tests: Webapp (2, 12)
- GitHub Check: packages / 🧪 Unit Tests: Packages (2, 3)
- GitHub Check: packages / 🧪 Unit Tests: Packages (3, 3)
- GitHub Check: packages / 🧪 Unit Tests: Packages (1, 3)
- GitHub Check: typecheck / typecheck
- GitHub Check: internal / 🧪 Unit Tests: Internal
- GitHub Check: runops-guard / runops-guard
- GitHub Check: e2e-webapp / 🧪 E2E Tests: Webapp
- GitHub Check: code-quality / code-quality
- GitHub Check: Analyze (javascript-typescript)
🧰 Additional context used
📓 Path-based instructions (8)
**/*.{ts,tsx}
📄 CodeRabbit inference engine (.github/copilot-instructions.md)
**/*.{ts,tsx}: Use types over interfaces for TypeScript
Avoid using enums; prefer string unions or const objects instead
**/*.{ts,tsx}: Prefer static imports over dynamicimport(); use dynamic imports only for unresolvable circular dependencies, genuine performance code splitting, or conditional runtime loading.
Import Trigger.dev tasks from@trigger.dev/sdk; never use@trigger.dev/sdk/v3or deprecatedclient.defineJob.
Add agentcrumbs while writing code using approved namespaces; mark lines with//@Crumbsor blocks with `// `#region` `@crumbs, and strip them before merging.
Files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.tspackages/redis-worker/src/fair-queue/visibility.tspackages/redis-worker/src/fair-queue/index.ts
**/*.{ts,tsx,js,jsx}
📄 CodeRabbit inference engine (.github/copilot-instructions.md)
Use function declarations instead of default exports
Files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.tspackages/redis-worker/src/fair-queue/visibility.tspackages/redis-worker/src/fair-queue/index.ts
**/*.{test,spec}.{ts,tsx}
📄 CodeRabbit inference engine (.github/copilot-instructions.md)
Use vitest for all tests in the Trigger.dev repository
**/*.{test,spec}.{ts,tsx}: Use Vitest exclusively and never mock dependencies; use Testcontainers for integration dependencies.
Place test files next to the source files they test.
Files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.ts
**/*.ts
📄 CodeRabbit inference engine (.cursor/rules/otel-metrics.mdc)
**/*.ts: When creating or editing OTEL metrics (counters, histograms, gauges), ensure metric attributes have low cardinality by using only enums, booleans, bounded error codes, or bounded shard IDs
Do not use high-cardinality attributes in OTEL metrics such as UUIDs/IDs (envId, userId, runId, projectId, organizationId), unbounded integers (itemCount, batchSize, retryCount), timestamps (createdAt, startTime), or free-form strings (errorMessage, taskName, queueName)
When exporting OTEL metrics via OTLP to Prometheus, be aware that the exporter automatically adds unit suffixes to metric names (e.g., 'my_duration_ms' becomes 'my_duration_ms_milliseconds', 'my_counter' becomes 'my_counter_total'). Account for these transformations when writing Grafana dashboards or Prometheus queries
Files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.tspackages/redis-worker/src/fair-queue/visibility.tspackages/redis-worker/src/fair-queue/index.ts
packages/redis-worker/src/fair-queue/**/*.ts
📄 CodeRabbit inference engine (packages/redis-worker/CLAUDE.md)
Keep the fair dequeueing algorithm within
src/fair-queue/.
Files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.tspackages/redis-worker/src/fair-queue/visibility.tspackages/redis-worker/src/fair-queue/index.ts
packages/redis-worker/**/*.{ts,tsx}
📄 CodeRabbit inference engine (packages/redis-worker/CLAUDE.md)
Use
@trigger.dev/redis-workerfor all background jobs in the webapp.
Files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.tspackages/redis-worker/src/fair-queue/visibility.tspackages/redis-worker/src/fair-queue/index.ts
packages/redis-worker/**/*.{test,spec}.{ts,tsx}
📄 CodeRabbit inference engine (packages/redis-worker/CLAUDE.md)
Test Redis worker behavior with ioredis and testcontainers for Redis.
Files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.ts
packages/**/*.{ts,tsx}
📄 CodeRabbit inference engine (AGENTS.md)
For public packages, use
buildfor verification.
Files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.tspackages/redis-worker/src/fair-queue/visibility.tspackages/redis-worker/src/fair-queue/index.ts
🧠 Learnings (12)
📚 Learning: 2026-03-22T13:26:12.060Z
Learnt from: ericallam
Repo: triggerdotdev/trigger.dev PR: 3244
File: apps/webapp/app/components/code/TextEditor.tsx:81-86
Timestamp: 2026-03-22T13:26:12.060Z
Learning: In the triggerdotdev/trigger.dev codebase, do not flag `navigator.clipboard.writeText(...)` calls for `missing-await`/`unhandled-promise` issues. These clipboard writes are intentionally invoked without `await` and without `catch` handlers across the project; keep that behavior consistent when reviewing TypeScript/TSX files (e.g., usages like in `apps/webapp/app/components/code/TextEditor.tsx`).
Applied to files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.tspackages/redis-worker/src/fair-queue/visibility.tspackages/redis-worker/src/fair-queue/index.ts
📚 Learning: 2026-03-22T19:24:14.403Z
Learnt from: matt-aitken
Repo: triggerdotdev/trigger.dev PR: 3187
File: apps/webapp/app/v3/services/alerts/deliverErrorGroupAlert.server.ts:200-204
Timestamp: 2026-03-22T19:24:14.403Z
Learning: In the triggerdotdev/trigger.dev codebase, webhook URLs are not expected to contain embedded credentials/secrets (e.g., fields like `ProjectAlertWebhookProperties` should only hold credential-free webhook endpoints). During code review, if you see logging or inclusion of raw webhook URLs in error messages, do not automatically treat it as a credential-leak/secrets-in-logs issue by default—first verify the URL does not contain embedded credentials (for example, no username/password in the URL, no obvious secret/token query params or fragments). If the URL is credential-free per this project’s conventions, allow the logging.
Applied to files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.tspackages/redis-worker/src/fair-queue/visibility.tspackages/redis-worker/src/fair-queue/index.ts
📚 Learning: 2026-05-18T08:21:27.694Z
Learnt from: d-cs
Repo: triggerdotdev/trigger.dev PR: 3632
File: apps/webapp/sentry.server.ts:4-21
Timestamp: 2026-05-18T08:21:27.694Z
Learning: When handling Prisma error P1001 ("Can't reach database server") in TypeScript, don’t assume a single error shape. Prisma can surface P1001 via two different error classes/fields: `PrismaClientKnownRequestError` exposes it as `err.code === "P1001"` (common during mid-query connection drops), while `PrismaClientInitializationError` exposes it as `err.errorCode === "P1001"` (common on client startup failure). Therefore, predicates should use `err.code === "P1001" || err.errorCode === "P1001"`. Do not flag `err.code === "P1001"` as “unreachable/never matches,” as it is expected in production.
Applied to files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.tspackages/redis-worker/src/fair-queue/visibility.tspackages/redis-worker/src/fair-queue/index.ts
📚 Learning: 2026-05-18T08:21:27.694Z
Learnt from: d-cs
Repo: triggerdotdev/trigger.dev PR: 3632
File: apps/webapp/sentry.server.ts:4-21
Timestamp: 2026-05-18T08:21:27.694Z
Learning: When handling Prisma errors for P1001 ("Can't reach database server"), do not assume it only appears under a single property name. Prisma may surface P1001 via either `PrismaClientKnownRequestError` (`err.code === "P1001"`, e.g., mid-query connection drops) or `PrismaClientInitializationError` (`err.errorCode === "P1001"`, e.g., client startup connection failure). To reliably detect the condition, check `err.code === "P1001" || err.errorCode === "P1001"`, and avoid review rules that would incorrectly flag `err.code === "P1001"` as unreachable/never-matching.
Applied to files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.tspackages/redis-worker/src/fair-queue/visibility.tspackages/redis-worker/src/fair-queue/index.ts
📚 Learning: 2026-06-13T19:53:13.759Z
Learnt from: ericallam
Repo: triggerdotdev/trigger.dev PR: 3937
File: packages/trigger-sdk/skills/realtime-and-frontend/SKILL.md:258-260
Timestamp: 2026-06-13T19:53:13.759Z
Learning: When reviewing code that uses `trigger.dev/react-hooks`’s `useRealtimeRun`, preserve the call signature where the first argument is the full realtime handle object (not `handle.id`). This is intentional to maintain type-safety and is consistent with the official docs; do not suggest changing the first argument from the handle object to `handle.id`.
Applied to files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.tspackages/redis-worker/src/fair-queue/visibility.tspackages/redis-worker/src/fair-queue/index.ts
📚 Learning: 2026-06-17T17:13:49.929Z
Learnt from: matt-aitken
Repo: triggerdotdev/trigger.dev PR: 3948
File: apps/webapp/app/routes/_app.orgs.$organizationSlug.projects.$projectParam.env.$envParam.bulk-actions.$bulkActionParam/route.tsx:48-62
Timestamp: 2026-06-17T17:13:49.929Z
Learning: In triggerdotdev/trigger.dev, within `dashboardLoader`/`dashboardAction` (or similar context resolver code) whenever you resolve an organization ID from an organization slug for RBAC/enterprise authorization scope, always read from the primary Prisma client (`prisma`), not `$replica`. Using `$replica` can hit replica-lag and cause the RBAC lookup/authorization to run without the correct org scope (bypassing intended role enforcement). Implement the slug→org lookup with `prisma.organization.findFirst(...)` (or equivalent primary-client query) and add an inline comment documenting why the primary client is required (replica lag could lead to unscoped RBAC checks).
Applied to files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.tspackages/redis-worker/src/fair-queue/visibility.tspackages/redis-worker/src/fair-queue/index.ts
📚 Learning: 2026-06-23T13:04:21.413Z
Learnt from: carderne
Repo: triggerdotdev/trigger.dev PR: 4023
File: apps/webapp/app/services/upsertBranch.server.ts:14-18
Timestamp: 2026-06-23T13:04:21.413Z
Learning: In TypeScript, it’s valid to `import { type X }` and then use `typeof X` in a type-only position, e.g. `type Alias = z.infer<typeof X>`. The `type` modifier suppresses the runtime import, but the type checker still has the full exported type so `z.infer<typeof X>` can resolve correctly. In code reviews, don’t flag this as a TypeScript compile error as long as `typeof X` is used in a type context (e.g., with `z.infer`, `type` aliases, generics), not as a runtime value.
Applied to files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.tspackages/redis-worker/src/fair-queue/visibility.tspackages/redis-worker/src/fair-queue/index.ts
📚 Learning: 2026-05-18T14:40:02.173Z
Learnt from: ericallam
Repo: triggerdotdev/trigger.dev PR: 3658
File: packages/core/src/v3/realtimeStreams/manager.test.ts:1-147
Timestamp: 2026-05-18T14:40:02.173Z
Learning: In this repo’s trigger.dev codebase, the “never mock — use testcontainers” guideline should only be applied to integration tests that talk to real external services (e.g., Redis, Postgres, S2). For unit tests that validate in-memory logic (e.g., deduplication/cache behavior in StandardRealtimeStreamsManager and similar module-boundary call counting), it is allowed to use Vitest mocks like `vi.fn()` and to stub/mock `ApiClient` objects to count calls or simulate in-process collaborators. Do not flag `vi.fn()`-based mocks as policy violations in these unit-test scenarios; reserve the rule for true external-service integration tests.
Applied to files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.ts
📚 Learning: 2026-05-18T14:40:02.173Z
Learnt from: ericallam
Repo: triggerdotdev/trigger.dev PR: 3658
File: packages/core/src/v3/realtimeStreams/manager.test.ts:1-147
Timestamp: 2026-05-18T14:40:02.173Z
Learning: In the triggerdotdev/trigger.dev repo, the policy “Never mock anything — use testcontainers instead” should only be enforced for integration tests that interact with real external services (e.g., Redis, Postgres) via actual infrastructure. For unit tests that exercise pure in-memory logic (e.g., cache semantics) it is OK to stub collaborators such as `ApiClient` using Vitest (`vi.fn()`) to assert call counts or control behavior. Do not flag `vi.fn()`-based `ApiClient` stubs in unit tests as violations of the testcontainers policy.
Applied to files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.ts
📚 Learning: 2026-06-04T18:16:35.386Z
Learnt from: nicktrn
Repo: triggerdotdev/trigger.dev PR: 3836
File: apps/supervisor/src/backpressure/backpressureMonitor.ts:3-5
Timestamp: 2026-06-04T18:16:35.386Z
Learning: When reviewing TypeScript in this repo, apply the rule “prefer type aliases over interfaces” only to data/object shapes and union/intersection type modeling. If an interface is being used as a behavioral contract for collaborators to implement (e.g., method-shape interfaces that define required behavior, such as `BackpressureLogger` / `BackpressureSignalSource` in `apps/supervisor/src/backpressure/backpressureMonitor.ts`), keep it as an `interface` and do not flag it as a type-alias-vs-interface violation.
Applied to files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.tspackages/redis-worker/src/fair-queue/visibility.tspackages/redis-worker/src/fair-queue/index.ts
📚 Learning: 2026-06-09T17:58:04.699Z
Learnt from: 0ski
Repo: triggerdotdev/trigger.dev PR: 3879
File: apps/webapp/app/models/vercelIntegration.server.ts:619-630
Timestamp: 2026-06-09T17:58:04.699Z
Learning: In this codebase, outbound raw `fetch` calls should typically rely on Node/undici’s default request timeout (about ~300s) rather than adding a per-call `AbortController` + `setTimeout` wrapper inside individual functions (e.g. in files like `apps/webapp/app/models/vercelIntegration.server.ts`). During code review, do not flag the absence of a per-call timeout on a single `fetch` as an issue; if per-call timeouts are needed, they should be implemented via a codebase-wide convention (e.g., a shared fetch wrapper or documented pattern) rather than ad-hoc per-function changes.
Applied to files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.tspackages/redis-worker/src/fair-queue/visibility.tspackages/redis-worker/src/fair-queue/index.ts
📚 Learning: 2026-06-16T09:19:47.637Z
Learnt from: d-cs
Repo: triggerdotdev/trigger.dev PR: 3960
File: apps/webapp/test/prismaInfrastructureErrorCapture.test.ts:0-0
Timestamp: 2026-06-16T09:19:47.637Z
Learning: In this repo’s Vitest setup, `vitest.config.ts` uses `globals: true`, so identifiers like `vi`, `describe`, `it`, and `expect` are available as globals in Vitest test files. During code review, do not flag missing `vi`/`describe`/`it`/`expect` imports as a runtime error or correctness issue when they’re used in `*.test.ts/tsx` or `*.spec.ts/tsx` files. Explicit imports are still preferred for consistency, but they’re not required for runtime behavior.
Applied to files:
packages/redis-worker/src/fair-queue/tests/visibility.test.tspackages/redis-worker/src/fair-queue/tests/fairQueue.test.ts
🔇 Additional comments (13)
packages/redis-worker/src/fair-queue/visibility.ts (4)
401-402: LGTM!
443-451: LGTM!
719-724: LGTM!
795-799: LGTM!packages/redis-worker/src/fair-queue/index.ts (6)
28-28: LGTM!
1249-1262: LGTM!
1335-1354: LGTM!
1371-1387: LGTM!
1433-1473: LGTM!
1576-1594: LGTM!packages/redis-worker/src/fair-queue/tests/fairQueue.test.ts (2)
1520-1590: LGTM!
1592-1659: LGTM!packages/redis-worker/src/fair-queue/tests/visibility.test.ts (1)
916-979: LGTM!
…ite fails The message was removed from in-flight before the dead-letter entry was written, and the write went through a pipeline whose per-command errors were never inspected. A failed write therefore lost the message silently: gone from in-flight, absent from the dead-letter queue, invisible to the reclaim loop. Write the entry first and only complete the message once it lands. On failure the message stays in-flight, so the reclaim loop picks it up and it is retried rather than dropped. A persistently failing write now loops visibly instead of discarding work.
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
…s stranded or run twice Releasing a slot could silently do nothing. The release pipeline never inspected its per-command errors, and ioredis resolves a pipeline whose commands failed, so a failed SREM reported success and the caller went on to destroy the in-flight record. Release now throws on a failed command. Reclaim then acted on that false success: it released per message, swallowed any error, and requeued regardless, which is exactly how a slot ends up held with the message gone. It now frees the whole timed-out batch in one pipeline before any of them is requeued, and a failure aborts the requeue so the messages stay in-flight for the next tick. That also restores the single round trip per shard. Requeuing removed the message from in-flight before writing it back to the queue. Redis does not roll back a script that fails partway, so a failed queue write lost the message outright. The order is now write-then-remove. Batch items were never heartbeated, so any item slower than the visibility timeout was reclaimed and handed to a second consumer while the first was still running it, executing the item twice. Items are now heartbeated for as long as their callback runs, and the visibility timeout is configurable so this is testable.
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
… stop one bad key stalling a shard Each beat extended the deadline by the same amount as the gap between beats, so timer drift and round-trip latency left the message briefly past its deadline on every cycle, and a reclaim scan landing in that window still handed a running item to a second consumer. Beats now extend by a full visibility timeout while the tick stays at a third of it. Reclaim released the whole timed-out batch in one pipeline and aborted every requeue if any command failed, so a single unusable concurrency key stalled reclaim for every other tenant in that shard. A failed batch now falls back to releasing message by message and only the messages whose slot could not be freed are held back.
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
… is lost The heartbeat already learns when another consumer has taken the item over: the in-flight member is gone, so the extend returns false. That signal was discarded, leaving the original consumer free to finish and complete over the new owner's in-flight record. It now stops and discards its result instead, so the consumer that actually owns the item is the one that finishes it. Also treat a discarded release pipeline as a failure rather than a success, since a null result means none of the commands ran.
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
The in-flight member is keyed only by message and queue id, so once another consumer re-claims a reclaimed item the member exists again and an extend from the previous consumer succeeds. The signal therefore only catches the window where the item is back on the queue and unclaimed, which is narrower than the comment and the changeset claimed.
The heartbeat fixes a separate pre-existing bug (items slower than the visibility timeout are redelivered and executed twice) and shares no files with the concurrency slot fixes, so it ships on its own.
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
The test waited for the message to be queued and absent from in-flight at the same instant, a state that only lasts about one dispatch interval before the message is re-claimed. It now waits for the stuck in-flight entry's deadline to move forward instead, which is monotonic once reclaim has run.
| Fair queue consumers no longer leak the concurrency slots that gate a tenant's throughput. Slots were held by messages that had already finished, were never reclaimed, and once enough of them accumulated every queue belonging to that tenant stopped being served. Slots are now freed on the paths that previously skipped them, freed before the record needed to recover them is discarded, and released before a reclaimed message goes back on the queue. A failed release is now surfaced instead of being silently treated as success. | ||
|
|
||
| Concurrency groups keyed on queue metadata rather than the tenant can still resolve to the wrong group when a consumer completes a message it did not enqueue, so this does not yet cover that case. |
There was a problem hiding this comment.
🟡 Release note text explains internal implementation instead of user impact
The release note added for this change (.changeset/fair-queue-concurrency-slot-leak.md:5-7) describes internal mechanics and a maintainer-facing caveat instead of one plain sentence about what changed for the user, which is what the repository requires because this text ships verbatim in user-visible release notes.
Impact: Users reading the release notes get implementation detail and an internal caveat rather than a clear statement of the behaviour change.
Rule in AGENTS.md on changeset wording
AGENTS.md ("Changesets and Server Changes") states: "Write the description for users, not maintainers. Both changesets and .server-changes/ notes ship verbatim in user-visible release notes. Lead with what changed for the user - one plain sentence describing behavior, not implementation, and never naming internal tools or infra."
The note instead says slots are "freed before the record needed to recover them is discarded, and released before a reclaimed message goes back on the queue. A failed release is now surfaced instead of being silently treated as success.", and the second paragraph documents a remaining internal limitation about "concurrency groups keyed on queue metadata".
| Fair queue consumers no longer leak the concurrency slots that gate a tenant's throughput. Slots were held by messages that had already finished, were never reclaimed, and once enough of them accumulated every queue belonging to that tenant stopped being served. Slots are now freed on the paths that previously skipped them, freed before the record needed to recover them is discarded, and released before a reclaimed message goes back on the queue. A failed release is now surfaced instead of being silently treated as success. | |
| Concurrency groups keyed on queue metadata rather than the tenant can still resolve to the wrong group when a consumer completes a message it did not enqueue, so this does not yet cover that case. | |
| Fixes a bug where a tenant's queued work could stop running permanently after messages finished processing. |
Was this helpful? React with 👍 or 👎 to provide feedback.
2 participants
Summary
A tenant could reach a state where none of its queued work ever ran again. The fair queue's per-tenant concurrency limiter was holding slots for messages that had already finished, and nothing reclaimed them. Once the leaked slots reached the tenant's limit, every queue belonging to that tenant stalled permanently, with items sitting in Redis untouched.
Root cause
completeMessageandreleaseMessageboth read the message's in-flight record to build a queue descriptor, then skipped the concurrency release entirely when that read came back empty:The release does not need that record. The descriptor built on the line above already falls back to
extractTenantId(queueId), andtenantIdis the only fieldrelease()reads.Ordering made it worse: the release ran after
visibilityManager.complete()had deleted the in-flight record, so the reclaim loop, the one thing that can recover a slot, had already lost its handle on the message. Any failure between those two awaits stranded the slot with nothing able to free it.The fix is to always release the slot, and to release it before the in-flight record is removed, so an interruption leaves a reclaimable message rather than an orphaned slot.
The regression test deletes the in-flight record before completing a message on a tenant with a limit of 1. Without the fix the next message never gets a slot and the test times out, which is exactly the observed failure.