Memory alert fires but autoscaler does not restart container on Scalingo

Automatic Restarts and Memory Alerts

No, Scalingo’s autoscaler cannot be configured to automatically restart containers based on the platform's memory-usage alerts. The memory alert (triggered when RAM exceeds 80% for more than five minutes) is a notification mechanism only; it does not trigger an operational action like a container restart.

Health Checks vs. Memory Alerts

While memory alerts are passive, Scalingo does support automatic restarts via health checks. If you define a custom health check in your scalingo.json file, the platform will monitor the specified endpoint. If that endpoint fails to respond or returns an error, Scalingo will restart the container to attempt recovery.

To implement a restart mechanism for memory issues, you must build a custom health check endpoint within your application that monitors its own memory usage and returns a non-200 status code when a critical threshold is reached.

Implementation Steps

1. Create an Endpoint: Develop a route (e.g., /health) in your application that checks current memory consumption.
2. Define Logic: Program the endpoint to return an HTTP 500 or 503 error if memory usage exceeds your desired limit.
3. Configure scalingo.json: Add the health check configuration to your scalingo.json file to point to this endpoint.
4. Deploy: Push the updated configuration to the platform.

Risks and Constraints

• State Loss: Automatic restarts will wipe all in-memory data. For stateful applications, ensure all critical data is persisted in an external database or object storage.
• Restart Loops: If a memory leak occurs immediately upon startup, the container may enter a crash loop. Scalingo may throttle restart frequency to prevent platform instability.
• Billing: Restarts do not impact billing, as the allocated dyno remains the same.

Verification

To verify your current configuration and monitor behavior, use the following commands:

# Check if a health check is currently configured
scalingo -a your-app-name config:get

# Monitor logs for restart events following a memory spike
scalingo -a your-app-name logs

# Check current container status
scalingo -a your-app-name ps

Diagnostic Detail Needed: Does your application currently expose an HTTP endpoint that can be used for health monitoring, or would you need to implement one specifically for this purpose?