Architecture

Pipeline stages

1. Connect to Spanner and warm the session pool.
2. Read all enabled pipelines with their last modified timestamps.
3. Compute expected delay windows by pipeline type and PIMS.
4. Log warnings when pipelines exceed the delay + grace period.

System context

Spanner (Pipelines, Organizations) -> Watcher Job -> Logs/Sentry

Dependencies

• Upstream: Cloud Spanner.
• Downstream: Logging/Sentry only.

Failure handling

• No retries beyond Cloud Run Job retry policy.
• Idempotent by design; read-only checks.