Cookbook

Short, concrete recipes for the things you'll wire up on day one. Code first, explanation second.

Alerting

The autoscaler emits a handful of events your app can listen to. These recipes hook them up to the three places most Laravel teams actually read:

Alert via Laravel Log — zero setup, uses your existing log stack
Alert via Slack — one webhook URL, done
Alert via Email — via Laravel Notifications
Alert on a Fuse Trip — page on-call when a queue's jobs start failing

All three transport recipes use the built-in AlertRateLimiter so you don't get 50 Slack pings for one breach.

Cluster Monitoring

Export Cluster Metrics — expose clusterMetrics() and ClusterScalingSignalUpdated to Prometheus, dashboards, or cboxdk/laravel-queue-monitor

Events the autoscaler emits

Event	When	Fires
`SlaBreached`	Oldest job age ≥ SLA target	Once on state entry
`SlaRecovered`	Oldest job age drops below SLA target	Once on state exit
`SlaBreachPredicted`	Forecaster predicts pickup time will exceed SLA	Per evaluation cycle during risk
`WorkersScaled`	Workers spawn or terminate	Per scaling action
`ScalingDecisionMade`	Any scaling decision (including HOLD)	Per evaluation cycle
`FuseTripped`	Failure rate crosses the threshold, scaling held at `workers.min`	Once on state entry
`FuseProbing`	Cooldown elapsed, a probe tests for recovery	Once per probe
`FuseRecovered`	Probe succeeded, normal scaling resumes	Once on state exit

SlaBreached/SlaRecovered and all three fuse events already fire at most once per state change, so they don't need additional rate-limiting. SlaBreachPredicted and WorkersScaled can fire rapidly during flapping — those are the events where rate-limiting matters.

The rate limiter in one line

use Cbox\LaravelQueueAutoscale\Alerting\AlertRateLimiter;

public function __construct(private AlertRateLimiter $limiter) {}

public function handle(SlaBreached $event): void
{
    if (! $this->limiter->allow("slack:{$event->connection}:{$event->queue}")) {
        return;
    }
    // send the alert
}

Default cooldown is 300 seconds (5 min), configurable via queue-autoscale.alerting.cooldown_seconds or the QUEUE_AUTOSCALE_ALERT_COOLDOWN env var.