Skip to content

feat(server): add evaluation timing metrics#238

Closed
abhinav-galileo wants to merge 18 commits into
mainfrom
abhi/agent-control-eval-timing-metrics
Closed

feat(server): add evaluation timing metrics#238
abhinav-galileo wants to merge 18 commits into
mainfrom
abhi/agent-control-eval-timing-metrics

Conversation

@abhinav-galileo

@abhinav-galileo abhinav-galileo commented Jun 15, 2026

Copy link
Copy Markdown
Collaborator

Summary

  • Add timing metrics and tracing around evaluation execution and Luna evaluator stages.
  • Expose evaluation concurrency as configuration.
  • Add Luna HTTP phase/client timing and richer HTTP error metadata.

Notes

Validation

- bound connection establishment and per-statement runtime at the
  driver level for non-SQLite URLs (psycopg and asyncpg)
- default AGENT_CONTROL_DB_MAX_OVERFLOW to 10 so short bursts borrow
  transient connections instead of failing after the pool timeout
- initialize the checked-out-connections gauge so idle processes
  report 0 instead of an absent series
- Verify TLS against an optional CA bundle (ca_file arg or GALILEO_LUNA_CA_FILE) so internal API endpoints with internally-issued certificates work in internal auth mode.
- Bound connection pooling (keepalive expiry 1s, connection limits) so requests do not reuse sockets the server already closed, and retry the idempotent scorer invoke once on connection errors.
- Deprecate GALILEO_LUNA_AUTH_MODE; the auth mode is inferred from the configured credential, and setting both credentials remains an explicit error.
- Trim whitespace when resolving API base URLs from the environment.
- Update the Luna example README and cover URL precedence, TLS, retry, and deprecation with tests.
@abhinav-galileo

Copy link
Copy Markdown
Collaborator Author

Superseded by split draft PRs: #239 for runtime tuning and #240 for timing observability.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant