Skip to content

E20: Operations

Goal: The platform is deployable, observable, recoverable and supportable.

Progress (Stage 9): aap integrations status|run, connector_runs, Integrations page, local integration stack and the integration bootstrap runbook IMPLEMENTED.

F01: Infrastructure

E20-F01-S01: Terraform for platform environments · M · Phase 1

  • AC: Modules: network (VNet integration, private DNS, NAT/firewall rules), Container Apps env (workload profiles, internal), PG Flexible (zone-redundant HA, geo-redundant backup on creation, Entra auth), Storage (WORM containers, private endpoint, shared key disabled), Key Vault, ACR Premium, Log Analytics/App Insights. dev and prod envs. Checkov clean or with recorded exceptions. azurerm ~> 4.x.
  • Depends on: The platform's own VNet (modules/platform, IMPLEMENTED as code).

E20-F01-S02: Platform CI/CD with provenance verification · M · Phase 1

  • AC: PR: lint, type-check, unit, SAST, SCA, secrets, IaC, container scan, SBOM, dependency review, coverage. Main: build → attest → deploy to dev (migration as a Container Apps job) → smoke → manual approval → prod. CD verifies the image attestation before deploy (ADR-030). Post-deploy verification (health, synthetic webhook).

F02: Observability

E20-F02-S01: OpenTelemetry, SLOs and alerts · M · Phase 4

  • AC: Traces webhook → job → evaluation. SLO dashboards (ADR-027). Alerts with linked runbooks. A redaction test on logs.

E20-F02-S02: Connector health in product · M · Phase 4–8

  • AC: Per connector: last success, error rate, rate-limit headroom. A stale connector turns dependent evaluations to error with the last known state and shows a banner.

F03: Resilience

E20-F03-S01: Backup, restore and DR drill · M · Phase 9

  • AC: PITR restore to a scratch server + chain verification succeeds within RTO. Regional restore tabletop. The drill record is stored as platform evidence.

E20-F03-S02: Runbooks · M · Phase 9

  • AC: Runbooks: platform incident, CVE emergency (including the DT fallback query), evidence integrity mismatch, key rotation, connector outage, dead-letter replay, partition maintenance, onboarding support.