SLO Aggregation

[ian-bartholomew]

Problem to solve

Currently, multiple SLOs can encompass a single user journey, without a single SLO that measures the user experience.

Proposal

Add the ability to aggregate SLOs, and roll them up into a single SLO

This is similar to Keptn's Quality Gates, and proposed by Andres Grabner. More info here: https://www.youtube.com/watch?v=bMnMkOKVzdg

Further details

Key features:

• Performance Signature
• Synthetic SLI from multiple SLOs
• Key SLOs
  • All will fail if this fails
• Weighted
  • Total weight is the aggregation/sum of all weights
• Performance testings
• Regression Detection

[mmazur]

The simplest usable (to me) syntax for this would be:

spec:
  description: Aggregate SLO
  budgetingMethod: ?
  indicator:
    objectiveMetric:
      source: http://localhost:9090
      queryType: promql
      query: component:latency:slo_ok_5m{component="prod-comp-1"}
  objectives:
  - target: 0.95
  timeWindows:
  - count: 28
    unit: Day
    isRolling: true

With the evaluation engine being smart enough to compute this by creating the following query for prom and just storing the result:

avg_over_time(component:latency:slo_ok_5m{component="prod-comp-1"}[28d])

[mmazur]

Hmm, I think I misunderstood the goal of this issue. I'll move over to my own issue :)

[r3code]

@ian-bartholomew In sloth.dev definition I do it by calculating a raw error_query_ratio it as sum of all good events and all totals from SLIs.