Version 1.2.0
Automatic scaling
Automatically adjust worker and rolling-service replicas within configured limits using resource and traffic metrics.
Enable automatic scaling
The monitor's scaling controller adjusts replicas for worker and rolling-service applications. Set explicit lower and upper bounds. Scaling uses the current artifact and does not rebuild or create an application version.
deploy:
strategy: rolling
replicas: 2
autoscaling:
enabled: true
min_replicas: 1
max_replicas: 5
interval: 15s
metrics:
cpu:
scale_up: 70
scale_down: 30
scale_up:
evaluation_windows: 3
cooldown: 60s
scale_down:
evaluation_windows: 5
cooldown: 180s
stabilization_window: 300sWorkers also accept the autoscaling block with their existing worker deployment configuration. If max_replicas is omitted, it defaults to deploy.replicas, so enabling the controller alone does not expand your declared fleet.
Decision policy and metrics
Scaling adds one replica when any enabled metric crosses its upper threshold and removes one only when every enabled metric is below its lower threshold. Scale-up wins conflicts. Evaluation windows, cooldowns, and scale-down stabilization prevent rapid reversals.
CPU is measured per logical core. Memory thresholds use each replica's configured memory limit. Request rate and latency apply to applications enrolled in the proxy. Missing metrics block scale-down, and stale samples do not trigger decisions.
Pause, resume, and inspect
phelix autoscale status api
phelix autoscale pause api
phelix autoscale resume api
phelix autoscale evaluate api
phelix autoscale history apiPause survives monitor restarts. Enable and disable change configuration; evaluate requests an immediate evaluation. Decision history includes the reason and execution outcome. Per-replica metrics remain available when automatic scaling is disabled.
Remote management and recovery
Remote controls use the connected monitor and require scale authorization. Automatic scaling shares the application deployment lock with rebuild, rollback, and manual scaling. It refuses conflicting rollouts, health-checks added rolling replicas, drains removed replicas, and recovers interrupted scaling on monitor restart.