Scale an app automatically
Add and remove replicas of an app as its load changes, down to zero when it is idle.
Goal: an app that runs more replicas when it is busy and fewer when it is quiet, so you pay for what it uses.
Before you start
- A deployed app that can run as several replicas: it keeps no state in memory or on its own disk that other replicas need.
- Enough room on the cluster for the extra replicas, or a pool that grows with them (see Dedicated clusters).
Steps
Open the app on the project canvas and go to its Settings tab.
In the Autoscaling section, set Min Replicas and Max Replicas (at most 100).
Select Add and pick a trigger:
- CPU Utilization: scale out above a CPU percentage (70 by default).
- Memory Utilization: scale out above a memory percentage (80 by default).
- Cron Schedule: run more replicas during set hours, such as office hours.
- NATS JetStream Lag: scale workers by the number of waiting messages.
Optional: set the Cooldown (s), how long to wait before scaling in again (300 seconds by default).
Optional: turn on Scale to zero when idle. This sets Min Replicas to 0. The first request after an idle period waits while the app starts.
Apply the changes and deploy the app.
Without a trigger, the app does not scale: the section shows No triggers configured.
Verify
Put load on the app. Its number of replicas rises, up to Max Replicas, and falls back to Min Replicas after the cooldown once the load is gone. Follow CPU and memory in the Metrics on the Overview tab.
Troubleshooting
- The app never scales out: check that a trigger is configured and that its threshold is reached.
- New replicas stay pending: the cluster has no room. Add servers, or let the pool grow with pool autoscaling.
- Replicas go up and down all the time: raise the Cooldown (s).
In IaC
The autoscaling block takes min, max, triggers, scale_to_zero and
cooldown_seconds. See the example under Scaling.
Next steps
- Read logs and metrics
- Dedicated clusters: pools that add servers when apps need them.