Skip to content

SLO Burn Response Runbook

Use this runbook when a workload service-level objective (SLO) emits slo.error_budget_burn. It verifies the objective and source data, confirms the short- and long-window burn, correlates likely context, and routes a governed incident response without treating missing data as healthy.

Thresholds, objective values, metric queries, and notification routes are configuration supplied by the downstream fork. This template does not define one universal burn policy.

Start with the detected issue ID, SLO and service IDs, evaluated windows, source timestamp, and configured route. Assign an owner for verification and record the next decision deadline before beginning deeper investigation.

Required inputWhat to verify
SLO versionObjective, indicator, target, evaluation period, and owner
Metric sourceQuery or projection, aggregation, dimensions, and freshness
Burn windowsShort and long window boundaries, thresholds, and sample counts
Error budgetConsumed and remaining budget for the active period
ScopeService, region, operation, dependency, and explicit exclusions
ContextDeployments, maintenance, capacity events, and open incidents
  1. Confirm the detected issue references the currently active SLO version.
  2. Verify the service-level indicator (SLI), which is the measured signal behind the objective, uses the intended scope and dimensions.
  3. Check source health, ingestion delay, sampling, and missing-data behavior.
  4. Recompute or independently inspect both configured windows from the same source.
  5. Confirm threshold comparisons and remaining error budget without rounding away a breach.
  6. Compare the detected issue timestamp with deployments, maintenance, capacity, and incidents.

If the detected issue is invalid, record why and route the labeled case to alert tuning. Do not simply close it as noise.

  1. Record the baseline. Capture the window values, error budget, source freshness, affected dimensions, and current user impact.
  2. Check for an incident. Update an existing correlated incident or open a new one using a stable correlation key.
  3. Set severity. Apply configured severity policy to measured user impact, burn, duration, and scope. The burn alert alone does not bypass policy.
  4. Assign and notify. Select the configured owner, send durable notification, and verify delivery or fallback outcome.
  5. Investigate context. Start a bounded investigation across recent changes, capacity, dependencies, and related detected issues.
  6. Prepare mitigation. For any proposed change, record evidence, intended effect, scope, what-if result, stop conditions, and rollback.
  7. Route the proposal. Send the typed action through risk and approval policy.
  8. Observe recovery. Continue both burn windows through the configured recovery period before declaring the SLO stable.
Detected issue stateResponse
Both windows breach and impact is confirmedTriage or update the incident immediately
Short window breaches but long window does notMonitor to the next deadline and inspect acute context
Long window breaches without a short-window spikeInvestigate sustained degradation and budget trend
Burn is valid but no user impact is yet visibleKeep the detected issue active and investigate before budget exhaustion
Source or SLI scope is invalidRecord an invalid detected issue and begin measured alert tuning
Existing incident already covers the scopeAdd evidence to that incident rather than opening a duplicate

Stop when samples are stale, the SLI is mis-scoped, missing data was treated as zero, window boundaries differ from configuration, or rollback and impact bounds are unavailable. Stop state transitions when the incident has changed concurrently, refresh it, and repeat the decision.

Recovery requires more than one healthy sample. Verify that:

  • Windows: short and long burn values remain below their configured recovery conditions.
  • Budget: remaining error budget and projection are recorded after mitigation.
  • Impact: affected operations and dependencies pass their health checks.
  • Change: any mitigation has a known active version and rollback reference.
  • Incident: the state moves to monitoring with a defined review deadline.

If a mitigation worsens the burn or another guard condition, follow incident mitigation and rollback.

Record SLO version, window values, source timestamp, incident ID, proposal ID, decision, and terminal outcome. Also record SLI dimensions, source-health checks, error-budget values, correlation context, notification outcome, and recovery window.

Complete the response when the detected issue is classified as valid or invalid, the incident and owner are known, every proposal has a terminal decision, and the SLO has either passed its recovery window or remains open with a next decision deadline. Preserve invalid detected issues as labeled tuning scenarios.

To continue withRead
Establish incident scope and ownershipIncident triage
Improve an invalid or noisy detectorAlert tuning
Execute a verified response safelyIncident mitigation and rollback