Site Reliability on Rails
summary
Argues that the first fix is visibility: deploy APMs, record event streams and minute snapshots, and you can time-travel to incidents. Recommends concrete measures: snapshot PG stat statements every 10 minutes, treat availability as request‑based (target 99.9% of requests succeed) and turn that into an error budget. Shows how to measure jobs (wait, service, response), use Apdex and an agony score to prioritize, add alert tiers and dashboards, and reports reducing an initial ~150% error‑budget overrun down to within budget.
AI-written summary · about
chapters (22)
- 0:00Intro
- 0:18Site Reliability on Rails Anthony Crumley
- 3:20The Cloud of Unknowing
- 5:47Service Levels
- 6:02Service Level Indicator (SLI)
- 6:18Service Level Objective (SLO)
- 6:40Service Level Agreement (SLA)
- 7:26Availability
- 7:39Time based Periodic pings
- 8:21Request based
- 9:00Time Travel
- 9:32Event Streams Requests, jobs, exceptions, events
- 10:03Snapshots Load, cpu, memory, 10, queue size
- 14:51Jobs are snowflakes Snapshot: Queue sizes
- 15:53Maximum Be aware of outside causes. Early indicator of performance problems.
- 16:40Percentiles Reliable indicator for SLOS.
- 18:09Time Consuming Source of baseline load
- 23:59Slow Queries APM, slow query logs, database stats
- 25:19Errors Don't ignore. Handle responsibly.
- 25:51Error Budget Guilt and conflict freefish error management.
- 29:24Filters
- 31:04Communication
related talks
31mOperating Rails: What About After You Deploy?
01h 04mDevOps Without Measurement is a Fail
30mImproving Reliability w/ Error Budgets & SRE
30mReliability Engineering for Humans
26mTechnology and Cultural Transformation using Site Reliability Engineering
37mMonitoring the Health of Your App
37mManaging incidents in microservices
31mNo-Nonsense Observability Improvement