Process Logs
Crow surfaces the process logs of each agent and autoscaler — the zerolog output the producer would normally write to stdout — directly in the Crow web UI. This lets operators debug a misbehaving producer without SSHing into the host.
Where to find logs
Section titled “Where to find logs”- As an administrator, open Admin → Agents (or Autoscalers).
- Click on a row to open the detail page.
- Select the Logs tab.
The viewer streams live entries via SSE and replays the most recent batch from the database so you have context immediately.
What is retained
Section titled “What is retained”Process logs are stored in the proc_log_entries table, bounded per source.
By default, the 5,000 most recent lines per producer are retained; older lines are trimmed by a periodic janitor.
Stdout is not replaced — docker logs crow-agent (or your platform’s equivalent) continues to work exactly as before.
During short network outages
Section titled “During short network outages”If an agent temporarily loses gRPC connectivity to the server, lines are kept in a small on-agent ring buffer (default 500 lines) and back-filled on reconnect with their original timestamps. Longer outages may drop lines beyond the buffer size.
Disabling
Section titled “Disabling”Operators who do not want process-log persistence can disable the feature entirely:
When disabled, ingest endpoints return Unimplemented / 404, the SSE stream returns immediately, and the janitor does not run.
Producers fall back to stdout-only.
Tuning
Section titled “Tuning”See the server env vars and agent env vars pages for retention, batch, and rate-limit controls.