Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

Review feedback addressed — 2f717c6d1

Thanks for the thorough pass. Validity-checked each finding; here's the disposition.

Fixed

• 🔴 nohup no-liveness (run-worker.sh) — detached launch now verifies kill -0 after sleep 1, prints the log tail + returns non-zero on a dead fork. Applied to the shared detach path (benefits every worker).
• 🟠 wrong-registry observability (consumer.py) — run() logs the registered application task names at startup, so a non-empty-but-mismatched registry is diagnosable from one line (… 7 application task(s) registered: …, send_webhook_notification).
• 🟠 _env casting + typing — TypeVar-typed (was cast: type -> object, erasing types at the typed __init__); a bad value now raises Invalid WORKER_PG_QUEUE_CONSUMER_<X>=… instead of a context-free ValueError.
• 🟡 test gaps — added the guard's positive / require_tasks=False bypass / built-in-filter arms.
• 🟢 pgrep || true — warns on rc > 1 (operational/regex error) instead of collapsing it to "not running".
• 🟢 comment accuracy ×3 — build_celery_app configures but doesn't import tasks; notification is the worker that owns the leaf task send_webhook_notification; generalised the "every notification dropped" example.
• SonarCloud S1192 — extracted the repeated "pg_queue_consumer" literal into a readonly constant.

Deferred (with rationale, replied inline)

• 🟡 opt-in-hides-crash — start-time crashes now caught by the liveness check; steady-state (died-later) detection needs the future health endpoint (already a rollout item).
• 🟡 queue-skew — producer/consumer queues are verified-aligned for the one migrated task; cross-queue detection belongs with the second leaf migration (don't want untested cross-queue logic now).
• 🟡 poll-loop escalation past a failure threshold — pre-existing 9c (UN-3539 [FEAT] PG Queue Phase 9c — consumer poll loop (claim → run → ack) #2045) behaviour, and it changes crash semantics (exit-on-sustained-fault vs log-and-continue) — worth its own discussion/PR rather than bundling here.

Re-verified live after the changes: ./run-worker.sh -r -d pg-queue-consumer → liveness passes, --status shows RUNNING, startup log lists the 7 notification tasks. Consumer test suite green.

sequenceDiagram
    participant Shell as run-worker.sh
    participant UV as uv run python
    participant Main as pg_queue_consumer.__main__
    participant Worker as worker.py (bootstrap)
    participant Consumer as PgQueueConsumer.run()
    participant PG as pg_queue_message

    Shell->>Shell: "export WORKER_TYPE=pg_queue_consumer"
    Shell->>Shell: "export WORKER_PG_QUEUE_CONSUMER_QUEUE=notifications"
    Shell->>Shell: "export WORKER_PG_QUEUE_CONSUMER_WORKER_TYPE=notification"
    Shell->>UV: uv run python -m pg_queue_consumer
    UV->>Main: "__name__ == "__main__" → _bootstrap_and_run()"
    Main->>Main: "os.environ["WORKER_TYPE"] = "notification""
    Main->>Worker: import worker (registers notification tasks)
    Worker-->>Main: tasks registered in current_app
    Main->>Consumer: main() → PgQueueConsumer(...).run()
    Consumer->>Consumer: "require_tasks guard: _registered_task_count() > 0?"
    alt registry empty (wrong bootstrap)
        Consumer-->>Shell: RuntimeError (fail loud)
    else tasks present
        Consumer->>Consumer: install signal handlers
        loop poll loop
            Consumer->>PG: SKIP LOCKED claim batch
            PG-->>Consumer: messages (or empty)
            Consumer->>Consumer: task.apply() in-process
            Consumer->>PG: DELETE msg_id (ack)
        end
    end

_{Reviews (2): Last reviewed commit: "UN-3541 [FIX] Address Greptile review fe..." | Re-trigger Greptile}

Greptile feedback addressed — 5bec0f8b4

Thanks @greptile-apps (confidence 4/5). All three P2 findings handled:

• Module-level side-effects in __main__.py — moved the WORKER_TYPE mutation + import worker into a guarded _bootstrap_and_run() called only under if __name__ == "__main__":. Accidental imports are now inert (verified: import leaves WORKER_TYPE untouched and doesn't bootstrap worker; python -m path unchanged).
• usage() help text (outside-diff comment) — added pg-queue-consumer to the worker-type list plus a note documenting the WORKER_PG_QUEUE_CONSUMER_WORKER_TYPE / _QUEUE overrides.
• Liveness check scope — kept general but clarified in the comment that it's a best-effort fast-fail for immediate (sub-second) crashes (import/config faults, which affect any worker), explicitly not a connectivity check. Left the sleep at 1s (bumping it trades latency for only partial slow-failure coverage, which needs the real health endpoint); all is unaffected since those subshells overlap the 1s with the inter-launch sleep. Rationale on the thread.

Re-verified live after the changes: python -m pg_queue_consumer boots the 7 notification tasks, --status → RUNNING, --help lists the worker.

All three issues are well-addressed. The __main__.py fix is clean — _bootstrap_and_run() correctly encapsulates both the env mutation and the import side-effect, and the guard ensures tests/IDEs importing the module stay inert. The docstring and inline comment both do a good job explaining the "why" for the os.environ overwrite (vs setdefault).

No further concerns on these changes. The PR looks good from a review standpoint on these items.

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud