Observe and troubleshoot applications
Use deployment states, runtime logs, request identifiers, workflow runs, and usage records to verify real behavior.
Observability starts with identifying which environment and release you are testing. A local process, Agent preview, preview deployment, and production deployment are four different runtime states. Record the project, deployment, hostname, and request time before investigating.
Read deployment state first
Open Deployments or run sprout deployment get <deployment-id>. A build failure means SproutOS could not prepare the artifact or image. A runtime failure means the build completed but the application could not start or serve. Inspect that reason before searching application logs.
Search runtime logs
Open the project's Logs page or use:
sprout logs my-site --since 30m --limit 200
sprout logs my-site --follow
Filter by level or text when you know the failure signature. Use structured fields for operation, resource id, status, and retry count so a person can distinguish two similar messages. Avoid logging credentials, authorization headers, connection URLs, session tokens, or full sensitive payloads.
Request identifiers let you correlate a browser failure with the correct runtime line. Capture the identifier and timestamp at the boundary, then search the same project rather than scanning every service.
Inspect workflow runs
Visual workflow run detail records overall status, a human-readable error, step inputs and bounded outputs, timestamps, and measured cost. A skipped sandboxed action makes the run fail with a reason; it is never reported as success. Queue payload inspection and editing require additional permissions and every edit requires an audit reason.
For repository workers, log one stable job id from receipt through completion and record retries without including secrets or entire customer documents.
Compare with usage
Billing usage is evidence that a measured resource was consumed, not proof that the user-visible operation succeeded. Compare compute, queue residency, database, storage, and Agent dimensions with runtime logs and the final state change. A green deployment plus no matching request usually means you tested a different hostname or environment.