Bound Agent Control database connection pool

[abhinav-galileo]

Summary

• add configurable SQLAlchemy async DB pool settings for the Agent Control server
• apply pool_pre_ping, bounded pool size/overflow, pool timeout, and rollback-on-return for non-SQLite URLs
• expose a Prometheus gauge for checked-out SQLAlchemy connections
• dispose the async engine during FastAPI shutdown

Context

High-concurrency deployments can exhaust Postgres connections when connection budgets scale implicitly with server replicas. This makes Agent Control's DB connection budget explicit per server process instead of relying on SQLAlchemy defaults.

Defaults:

• AGENT_CONTROL_DB_POOL_SIZE=5
• AGENT_CONTROL_DB_MAX_OVERFLOW=0
• AGENT_CONTROL_DB_POOL_TIMEOUT_SECONDS=5.0

Verification

• uv run --package agent-control-server ruff check --config pyproject.toml server/src server/tests/test_config.py server/tests/test_db.py
• uv run --package agent-control-server mypy --config-file pyproject.toml server/src
• direct Python assertion check for Postgres/SQLite DB engine kwargs

make server-test could not complete locally because Docker is unavailable and the shared test fixture requires Postgres on localhost:5432.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!