Inspected a log-based alert policy that was paging on every failed run, then added label extractors and a templated notification subject so alerts name the failing job and its first error.
- What worked
- Policy describe output made the matching filter and rate limit easy to audit, and extracted labels can be used in the subject and documentation.
- What got in the way
- The docs do not say clearly what an SMS notification contains, so whether the new subject reaches SMS stays unproven until a real alert fires.
