Skip to content

Adaptive capacity

Scale workers and rolling services automatically

Let capacity respond to measured demand within limits you choose. Keep the controls and the reason for every replica change.

Use the metrics that fit your workload

Set replica bounds and CPU or memory thresholds for workers and rolling services. Add request-rate and latency thresholds when the service uses the proxy. Missing or stale data remains explicit.

Use the metrics that fit your workload: Set replica bounds and CPU or memory thresholds for workers and rolling services

This fits workers and rolling services whose useful throughput improves with additional instances. Choose thresholds that represent that demand and a maximum the host can sustain. If a database or external provider is the bottleneck, more replicas can add pressure without improving the outcome, so begin with an understanding of where the workload is constrained.

Use the metrics that fit your workload: This fits workers and rolling services whose useful throughput improves with additional instances

Add capacity and drain it carefully

Evaluation windows, cooldowns, and stabilization prevent rapid reversals. Scaling uses the active artifact, shares the deployment lock, health-checks new rolling replicas, and drains replicas before removal. Each adjustment changes capacity within the declared limits while keeping the current application artifact and its normal operating protections.

Add capacity and drain it carefully: Evaluation windows, cooldowns, and stabilization prevent rapid reversals

A deliberate adjustment should preserve the behavior customers and background tasks depend on. Leave time for new instances to become ready and for removed instances to finish their work. Cooldowns and stabilization make the controller easier to observe during changing demand, while the shared deployment lock keeps capacity changes from racing a release.

Add capacity and drain it carefully: A deliberate adjustment should preserve the behavior customers and background tasks depend on

Retain manual and remote control

Pause and resume the controller remotely under scale authorization, inspect current resources, and review decision history. Pause state survives restarts, and interrupted scaling is reconciled when the monitor returns. Decision history preserves the reason and outcome of capacity changes so an operator can review what happened before resuming automatic control.

Retain manual and remote control: Pause and resume the controller remotely under scale authorization, inspect current resources, and review decision history

Keep a clear manual procedure for pausing changes during an incident and resuming after the cause is understood. Review the recorded decision reasons, current capacity, and resource measurements together. Scaling adjusts replicas on the current host rather than provisioning servers, so the limits and recovery plan must still reflect the machine supporting the workload.

Retain manual and remote control: Keep a clear manual procedure for pausing changes during an incident and resuming after the cause is understood

One binary between git push and production.

Install the CLI, verify the host, and follow the quick start to build your first application.

Read the docs