Troubleshooting
Diagnose the common gaps between a valid request and a working workload.
No compute is available
Check the current catalog and selected tier/geography. Confirm verified email, a positive usable balance, compatible model/memory, configured pricing, enabled capacity, and a connected runtime. A supported profile can remain unavailable.
Service stays queued or starting
Read service events and logs. Image download, container initialization, application startup, or capacity can delay readiness. Do not issue duplicate creates with new keys after a timeout. Inspect the existing resource first.
Endpoint returns 503
The service may be waking or the runtime may be unavailable. Honor Retry-After, then check state, logs, configured port, and the application's listening interface. An HTTP server bound only to 127.0.0.1 inside a container may not be reachable through the proxy.
A quiet service will not sleep
Inspect activity records. A stale task remains protective because network silence does not prove task completion. Confirm the task ended before releasing it. Open streams and WebSockets also count as activity.
Billing continues after pause
A pause request can still be releasing compute. Check for confirmed release. Provisioned storage remains billable during sleep and must be deleted separately if it is no longer needed. Deleting storage removes data.
Volume expansion fails
The volume cannot shrink. Confirm current revision, size limits, placement, and that attached compute is paused. Wait for storage readiness before resuming the service. Inspect errors before submitting another update.
A configuration update conflicts
Read the latest service or template revision. Review whether another client changed it. Apply your intended edit to the current version; do not blindly overwrite the full resource.
Command result is uncertain
Check logs and filesystem state before replaying a command. A timeout does not prove the command did not execute. Avoid retrying commands with side effects automatically.
Collect useful details
Keep the resource ID, operation time, lifecycle state, error code, and a redacted log excerpt. Do not share API keys, environment values, browser launch URLs, or private datasets in diagnostic material.