Runbook: Redis Unavailable

Project-Agent operations docs/ops/runbooks/redis-unavailable.md

Severity: SEV-2 by default; escalate to SEV-1 if queueing, rate limiting, or LLM failover becomes unsafe.
Page: on-call engineer when /health/components shows Redis unhealthy for more than 2 minutes.
Triage:
Confirm impact on rate limiting, circuit breaker state, and retry queues.
Check whether webhooks and approval escalations are backing up.
Mitigation:
Disable webhook fanout if retries are compounding load.
Pause approval auto-escalation if state churn is unsafe.
Prefer restoring Redis before rolling out any new deployment wave.