Laravel-sorok és események

Queues, Events & Scheduler

Queued jobs, retries and idempotency

• Offload slow or external work (emails, webhooks, third-party APIs, exports, media processing) to ShouldQueue jobs. Never treat a queued job or event listener as guaranteed to run exactly once.
• Make every job idempotent: guard side effects with state checks, unique database constraints, idempotency keys or ShouldBeUnique, so retries or duplicate dispatches never double-charge, double-send or corrupt records.
• Set explicit $tries, $backoff (array of seconds), $timeout and $maxExceptions, and implement failed(\Throwable $e) to record permanent failure and clean up partial state.
• Ensure the queue connection's retry_after in config/queue.php is strictly greater than the job's $timeout; otherwise a worker can pick up a second copy while the first is still running.

Payloads, transactions and worker state

• Pass Eloquent models (via SerializesModels, which stores only the model identifier and re-queries on execution) or scalar IDs — never serialize large collections, request objects or binary blobs into the queue payload. Handle deleted or stale models cleanly when the job runs.
• When dispatching a job or event inside a database transaction (or from a model observer), implement ShouldDispatchAfterCommit or call ->afterCommit() so workers never read uncommitted rows.
• In long-lived queue workers (queue:work, Horizon, Octane), never leak static state, singleton properties, auth state or tenant context from one job or request into the next. Explicitly pass and restore tenant context per job, and note when a deploy requires php artisan queue:restart.

Events, notifications and scheduler

• Use Events and Listeners to decouple domain side effects from the primary action, and Notifications (ShouldQueue) for user-facing messages across mail, database or Slack channels.
• Keep scheduled closures/commands in routes/console.php (or app/Console/Kernel.php in Laravel 10) thin by delegating to services or jobs. Add ->withoutOverlapping() on tasks that could run past their interval and ->onOneServer() in multi-server deployments.