Alerting

Page when users cannot complete a core action or a backlog will not drain by itself.

Alert	Likely cause	First move
ServiceDown	Process crashed or scrape failed	Container logs, then restart that service
KongHigh5xxRate / HighErrorRate5xx	Bad deploy or dependency	Trace Explorer for a sample id, then roll back the image
OutboxLagHigh	Relay or Redis	Outbox runbook
OutboxFailedBacklog	Poison payload	Fix handler, then replay dead letters
InboxBacklogHigh	Consumer stuck	Pending entries for that group
PushDeadLetterRateHigh	APNs or FCM rejecting	Push runbook
HikariPoolExhausted	Slow queries or a leak	Database runbook
ScheduledJobFailing	Cron exception	Service logs for the job name
RedisMemoryHigh / RedisEvictions	Cache or stream maxlen	Check SHARED_EVENTS_STREAM_MAXLEN and presence keys
JvmHeapPressure	Leak or undersized heap	Heap dump on a non-prod clone first if you can; restart if the error budget is burning

Warning-level host alerts (disk warning, CPU high, I/O wait) go to a ticket, not a night page, unless they accompany ServiceDown.

Silence an alert only with an expiry and an incident note.