Logs and Monitoring¶
Logs¶
Aegaeon writes INFO logs to stdout and to AEGAEON_LOG_FILE, which defaults to output.log. You can also set an absolute log path before startup:
mkdir -p logs
export AEGAEON_LOG_FILE="$PWD/logs/server.log"
The path is forwarded to Ray actors, so its parent directory must exist on every node. Ray actor stdout and stderr are also available in the Ray session log directory.
Runtime interfaces¶
| Endpoint | Contents |
|---|---|
/health |
API liveness |
/v1/models |
The service's deployed model list |
/v1/aegaeon/runtime |
Nodes, engines, replicas, routing, caches, and Decode load |
/v1/aegaeon/events |
Engine-state snapshots and model-switch events over SSE |
/v1/aegaeon/metrics |
Per-model request and generated-token rates, counts, and configured limits |
curl -sS http://127.0.0.1:8000/v1/aegaeon/runtime
curl -N http://127.0.0.1:8000/v1/aegaeon/events
Use replica state ready in the runtime snapshot to confirm model availability. Request rates and generated-token rates are per-model averages over a sliding 60-second window. Set AEGAEON_MAX_REQUESTS_PER_SECOND and AEGAEON_MAX_TOKENS_PER_SECOND before API startup to limit those rates; exceeding a limit returns HTTP 429.