bregman-arie When to use
Use when burn alerts fire or SLO dashboards show accelerated error budget consumption.
Preconditions
- The SLO definition and signals are known (good/bad events or latency threshold).
Procedure
- Confirm the SLO signal and the burn window.
- Identify whether burn is driven by errors, latency, or missing telemetry.
- Break down by endpoint, region, customer segment, and dependency.
- Correlate with deploys, config changes, and incidents.
- Choose the fastest mitigation that reduces burn (rollback, disable, scale, failover).
Decision points
- Telemetry missing: fix instrumentation/pipeline to avoid false burn.
- Dependency-induced burn: mitigate upstream (timeouts, circuit breaker, failover).
- Deploy-correlated burn: rollback or feature flag off.
Verification
- Burn rate returns to normal and stays stable.
- Alert conditions clear.
Rollback / undo
- Revert mitigations that cause new errors or correctness regressions.
Escalation
- Escalate to dependency owners if they are the dominant contributor.
- Escalate to service owners for code fixes.
Examples
Start by decomposing the SLO into error vs latency vs availability contributions.