Producing jobs
- typed, bulk, transactional, and HTTP enqueue;
- strict caller-supplied IDs, validated kind names, tags, priority, delay, pending jobs, and delete-on-completion ephemeral jobs;
- lifecycle uniqueness, throttle uniqueness, replacement, debounce, and kind exclusion;
- producer authorization, middleware, insert hooks, plugins, backpressure, outage classification, and circuit breaking;
- insert-and-await and caller-supplied connection pools.
Enqueueing guide
Scheduling and orchestration
- cron and sub-minute periodic schedules, timezones, missed-run policy, idempotent upsert, durable state, origin traceability, event audit, and periodic hooks;
- durable workflow DAG dependencies;
- named resumable steps and cursor iteration.
Periodic jobs
Workflows
Admission policy
- fleet and composable rate limits, cost weights, concurrency ceilings, and saturation strategies;
- work-conserving tenant fairness, weighted queues, queue priority ordering, and prefetch;
- poison-pill quarantine, store-supplied time, policy rejection metrics, and admission explanations;
- runtime policy writes, cross-gate rate-class validation, empty-poll backoff, and measured admission-gate overhead.
Policies
Handler execution
- typed dispatch, kind aliases, startup validation, and payload upcasting;
- retries and backoff, snooze, timeout, deadline, panic recovery, graceful shutdown, and non-consuming errors;
- progress, results, attempt logs, mid-run output, task-local typed data, client context, dependency extractors, long-task tracking, and handler-side lease control;
- isolated execution, sticky routing, and batch handlers.
Execution reliability
Handler context
Recovery and correctness
- crash accounting separate from returned failures, at-least-once delivery, lease fencing, and retry-time FIFO;
- archive/dead-letter behavior, redrive, death handlers, stuck-job callbacks, and move-suspect-to-back, including explicit abort and discard outcomes;
- idempotent effects, transactional step effects, injectable clocks, and step-resume helpers;
- orphan provenance and rate-limited-as-nonfailure semantics.
Runtime and fleet operations
- singleton duties, multi-node heartbeat, worker control, wakeups, backlog derivatives, oldest-job age, quiet-group metrics, and autoscaling signals;
- queue pause/resume, safe forced deletion, bounded memory samples, retention, eviction, per-queue history, SQL archive partitioning, and PostgreSQL index maintenance;
- rolling restart, memory guard, advisory-lock namespaces, and leader resignation;
- alternate schemas, Redis Sentinel and Cluster, migration tooling, CLI, and connection budgets.
Queue and runtime management
Integration and observability
- PostgreSQL, MySQL, and Redis stores with capability-honest optional interfaces;
- OpenTelemetry traces and metrics, envelope trace context, application subscriptions, and the embedded operations console;
- ORM transaction interop for supported native handles.
Testing and security
- in-memory stores, drain and execute-one helpers, assert-enqueued matchers, live test database management, injectable time, and a test-only uniqueness bypass;
- payload redaction by default, read-only UI/API posture, producer authorization, and client-side AES-256-GCM payload encryption.
Testing
Encryption at rest
Read boundaries as part of the feature
Not every backend implements every optional interface. Redis intentionally omits SQL transactions and archive partitioning; MySQL polls because it has no PostgreSQL-style notification channel. Encryption protects payload bytes, not policy metadata or output. Workflows currently provide immutable dependency graphs, not signals, timers, nested graph mutation, or workflow-level retry.Capability evidence
Learn how completed claims are tied to running tests and scenarios.