logql-cache-refresh-peak-lag

Instruction

For the earlier user-service cache incident, I want the warning-log view rather than Prometheus. From the cache refresh warning logs over roughly the last 12 hours, tell me how high `lag_seconds` got, about how many `stale_keys` were attached to that worst lag record, and whether those warnings look like one recurring background job or a mixed bag of unrelated user-service warnings.