Last Minute Notes
  • Github
  • Docs
  • Learn
    @sayanstwt
Lessons
0/37 done
  • Timeouts, Retries & Breakers
  • Multi-Region Design
  • Cells & Shuffle Sharding
  • SLIs, SLOs & Error Budgets
  • Observability
  • Deployment Strategies
  • Backups & Disaster Recovery
Learn/System Design/Reliability & Operations

SLIs, SLOs & Error Budgets

6 min Intermediate 7 steps

Turn reliability into a number you can spend.

Dashboard
green
Users
?
Target
not defined
0m15m30m45m60mCPUCPU 40%: fineUser requests
normal slow / failed marker
In this lessonStep 1 of 7
  1. The problem
  2. SLI
  3. SLO and budget
  4. Budget policy
  5. Burn rate
  6. Two windows
  7. Recap
Step 1 / 7problem

Green dashboards, unhappy users

CPU sat at a calm 40% for the whole hour. Meanwhile users got 5xx errors for almost 20 minutes.

Without a measured target, nobody can say how reliable is reliable enough, or whether this hour was acceptable.

Key idea

Reliability needs a number measured where users feel it, plus a target for that number.

←→ step·R replay·Space pause

You'll learn

  • Define an SLI from the user’s point of view
  • Turn an SLO into an error budget in minutes
  • Use the budget policy to decide between features and reliability
  • Alert on burn rate with a long and a short window
Read the full notesReliability and Operations, 1