fix(db): dispose engine connection pool at process exit

[JanSobus]

What this PR does

Registers engine.dispose() as an atexit handler in reana_db/database.py so that all pooled database connections are cleanly closed — with a proper TCP FIN — before any process using reana-db exits.

Root cause

The module-level engine in database.py holds a SQLAlchemy connection pool. When a short-lived process exits (e.g. a CronJob container), this pool is garbage-collected without calling engine.dispose(). Python's GC does not send PostgreSQL Terminate messages, so the server sees an abrupt disconnect and logs:

LOG: unexpected EOF on client connection with an open transaction

For Flask-based commands this is compounded by teardown_appcontext calling only session.remove() (which returns connections to the pool) without disposing the pool itself.

Affected CronJob containers

All four CronJob containers that use reana-db are fixed by this single change, since they all share the same module-level engine:

CronJob	Command
reana-system-status	flask reana-admin status-report
reana-retention-rules-apply	flask reana-admin retention-rules-apply
reana-resource-quota-update	reana-db quota resource-usage-update
reana-interactive-session-cleanup	flask reana-admin interactive-session-cleanup

Note: reana-interactive-session-cleanup was not mentioned in the original issue but is affected by the same root cause.

The fix

atexit.register(engine.dispose)

atexit is Python's standard library — no new package dependency. The handler fires once when the process is about to exit, regardless of how the process was started (Flask CLI, plain Click CLI, or long-running server). For long-running services, this is equally correct: engine.dispose() on graceful shutdown cleanly closes all idle pooled connections.

Tests

A unit test (tests/test_database.py) verifies that engine.dispose is registered as an atexit handler by reloading the module with a mocked atexit.register and asserting it was called.

Verifying the fix end-to-end

The unit test confirms the registration, but the most direct proof is observing PostgreSQL's connection logs before and after. To reproduce on any deployment with access to PostgreSQL logs:

1. Enable connection logging:

ALTER SYSTEM SET log_connections = on;
ALTER SYSTEM SET log_disconnections = on;
SELECT pg_reload_conf();

2. Run a CronJob command without the fix:

reana-db quota resource-usage-update

3. Check PostgreSQL logs — you'll see:

LOG: unexpected EOF on client connection with an open transaction

4. Apply this fix and run the same command again — you'll see instead:

LOG: disconnection: session time: 0:00:00.012 user=postgres database=postgres ...

A clean disconnection entry confirms that engine.dispose() sent the proper Terminate message before the process exited.

Closes reanahub/reana#943