#337 storage: WebhookAuditSink needs a close()/shutdown lifecycle

[Anonymous]

Originally created by: fu351

WebhookAuditSink (src/doberman/storage/sinks.py, [#317]) starts a daemon worker thread whose loop is while True with no stop signal. Nothing crashes — daemon threads die at interpreter exit — but there is no way to stop or reconfigure a sink without killing the process, and every test that constructs an active sink leaks a live worker thread for the rest of the pytest run (roughly 15 of the 41 tests in tests/unit/test_webhook_audit_sink.py do).

What to do

1. Add a close(): set a threading.Event the worker checks each loop iteration (a small timeout on queue.get lets the worker notice it), then join with a bounded timeout. Idempotent; emit() after close is a no-op, never an error.
2. Add a pytest fixture that closes any sink a test constructs, and use it across test_webhook_audit_sink.py.
3. Two small cleanups from the same review, worth folding in:
4. test_drop_oldest_when_queue_full and test_drop_counter_increments_on_each_overflow sync on time.sleep(0.05); the same Event lets them wait deterministically instead of racing CI load.
5. If Thread.start() raises inside __init__, the failure isn't cached, so every later decision retries the config read and thread spawn. Caching the failure as an inert sink keeps resource pressure from turning into a retry storm.

The decision-path contract doesn't change: emit() stays non-blocking, sink failures stay swallowed, and close() is never called from the decision path.

[Anonymous]

Ticket changed by: fu351

• status: open --> closed