ai-enabled-hr-talent-automation

On-Call Runbook

Title: On-Call Runbook Version: 1.0 Owner: [TENANT_CONFIGURATION_REQUIRED — DevOps/SRE] Status: Draft Last reviewed: 2026-09-07 Next review: [TENANT_CONFIGURATION_REQUIRED] Reviewers: DevOps, Architecture

Purpose and scope

Operational (non-security) incident response for on-call engineers, complementing incident-response-runbook.md (security-specific).

Alert triage

Alert First checks Likely cause Runbook link
API error-rate spike Recent deploy? DB health? Upstream provider status? Bad deploy, DB connection exhaustion, dependency outage Rollback procedure, DB connection-pool tuning
Outbox/DLQ depth growing Which adapter? Circuit breaker open? Downstream integration outage Check integration-adapter-catalog.md, manual replay after root cause fixed
Agent Plane latency spike Model provider status? Token budget exhaustion? Provider throttling, runaway prompt size Check model-routing-and-cost-controls.md fallback routing
RAG retrieval failures Vector store health? Vector store outage (non-critical, no-answer fallback should engage) Verify graceful degradation is active; escalate to vector store provider if prolonged
SLA-breach escalation flood Recent workflow-config change? Misconfigured SLA thresholds Check config/defaults/workflow.default.yaml recent changes

Escalation path

On-call engineer → Service owner → Architecture on-call → Engineering leadership, per [TENANT_CONFIGURATION_REQUIRED] paging policy. Security-flavored incidents (suspected breach, injection success) are immediately escalated per incident-response-runbook.md instead of this runbook.

Standard operating procedures

Change control

Version Date Author Change
1.0 2026-09-07 Documentation package generation Initial creation