Why should a platform publish SLOs to internal consumers, and what should they cover?
Show the full answer Hide the answer
Why publish them
Because consuming teams are taking a dependency and must be able to reason about it.
A team building on a platform capability inherits its availability. Without a published SLO they cannot compute their own availability, cannot decide whether the dependency should be hard or soft, and cannot design a fallback proportionate to the risk.
The second reason is organisational: an unpublished SLO is assumed to be perfect, so any failure is experienced as a breach of an expectation that was never set — which is how platform teams acquire a reputation for unreliability while meeting their internal targets.
What they should cover
- Availability of each capability, separately. Provisioning, deployment, the runtime data path and the observability pipeline have different criticalities and should not share one number.
- Latency at a percentile for anything on a request path.
- Time-to-provision for self-service operations, since that is what teams plan around.
- Deployment success rate and rollback time.
- Data durability and freshness for anything that stores or processes.
- Support response time for the paths that are not self-service.
The distinction that matters most
Control plane versus data plane. A platform's control plane — provisioning, configuration, deployment — can have a materially lower availability target than the data plane, provided workloads keep running when it is unavailable.
That property must be explicit, because it is the difference between "we cannot deploy for an hour" and "everything is down". Teams need to know which of their operations depend on which.
What the platform gets from publishing
- A defensible basis for prioritisation, since work maintaining the SLO competes on stated terms.
- Fewer escalations, because teams can plan rather than escalating when something is slow.
- An error budget, which converts reliability-versus-velocity from an argument into a measurement.
The obligation
Publishing an SLO means meeting it or explaining why. A published target that is routinely missed is worse than none, because it destroys the trust the publication was meant to build — and internal consumers, unlike external ones, can and will build their own.