root_cause_category: infrastructure
equivalent_root_cause_categories:
  - failover_event
  - multi_az_failover_health_check
  - failover
required_keywords:
  - failover
  - Multi-AZ
  - health check failure
  - primary evidence source
  - workload resumed normally
  - failover initiated
  - failover in progress
  - failover completed
  - instance available
required_evidence_sources:
  - aws_rds_events
optimal_trajectory:
  - query_grafana_logs
  - query_grafana_metrics
  - query_grafana_alert_rules
  - describe_rds_instance
  - describe_rds_events
max_investigation_loops: 4
model_response: |
  ROOT_CAUSE: Based on the RDS event timeline (primary evidence source), a Multi-AZ automatic failover occurred on payments-prod, triggered by a health check failure on the primary host. This was an infrastructure event rather than resource exhaustion, and the system recovered and workload resumed normally after failover completion.
  ROOT_CAUSE_CATEGORY: infrastructure

  VALIDATED_CLAIMS:
  - DatabaseConnections dropped to 0 at 08:05Z and recovered fully by 08:07Z. [evidence: aws_cloudwatch_metrics]
  - CPUUtilization and WriteIOPS both dropped to near-zero during the same window, confirming the instance was unavailable during failover rather than suffering a connection pool issue. [evidence: aws_cloudwatch_metrics]
  - RDS events show the full failover sequence: failover initiated (08:04:18Z) -> failover in progress (08:04:21Z) -> failover completed (08:04:58Z) -> instance available (08:05:04Z). [evidence: aws_rds_events]
  - Performance Insights shows no anomalous queries before or after the failover window, and the system recovered and workload resumed normally afterward. [evidence: aws_performance_insights]

  NON_VALIDATED_CLAIMS:
  - Application connection retry logic should be reviewed to minimize reconnection time after future failovers.

  CAUSAL_CHAIN:
  - AWS RDS detected a health check failure on the primary instance in us-east-1a.
  - Multi-AZ automatic failover was initiated, promoting the standby in us-east-1b to primary.
  - The DNS endpoint was updated to point clients to the promoted standby host.
  - A brief outage occurred during the failover window while existing database connections dropped and clients reconnected.
  - The instance became available again, and the system recovered and workload resumed normally.