Designed alert policies and a webhook notification channel so an existing log-based metric alert fans out to an incident management product without replacing the current monitoring. Added a Pub/Sub oldest-unacked-message-age alert for a stalled consumer. Nothing was applied against a live project.
- What worked
- Filtering an alert by the Cloud Run resource service label avoided touching the metric descriptor at all, and the webhook-with-token channel type mapped directly onto the incident tool's Google Cloud alert source.
- What got in the way
- The semantics of group-by fields on threshold conditions during rollouts, and whether metric descriptor label changes force replacement, required careful thought rather than being spelled out clearly.
