Dependency SLO ownership
observability/slo/dependency-ownership.yml is the starter's machine-readable
ownership catalog. Each enabled dependency should have one accountable owner,
an escalation route, an availability target, a latency target, an alert, and a
documented failure mode.
The default catalog treats PostgreSQL as required for user-service, while
Redis, email, and messaging are optional. Optional integrations should not
silently become release blockers: use their fail-open, queue, or outbox behavior
and alert at the dependency owner rather than paging the API owner for every
transient provider error.
At each quarterly SLO review:
- Replace example team names with real owners and on-call routes.
- Confirm each enabled dependency has a probe or telemetry signal matching its alert name.
- Compare dependency error/latency trends with the API's SLO burn and capacity results.
- Review timeout, retry, queue, and fail-open behavior for overload risks.
- Record changes to targets, ownership, escalation, or dependency criticality in the release/reliability review issue.
Capacity reviews should include the dependency ownership catalog, fixture volume, and result artifact. This makes it explicit whether a failure is caused by API compute, PostgreSQL data/index volume, cache behavior, or an optional integration.