Service Monitoring Manual—2018 Edition
summary
Modern monitoring treats observability as first-class: separate agents emit host, service, mesh and aggregate data; counters and latency histograms go to a time-series DB (Facebook’s ODS in-memory time-series), structured logs feed a query layer (Facebook’s Scoob real-time queries), and service frameworks expose uniform primitives so teams share p99/reqs/sec language. Alarms page only actionable symptoms; examples include a memory leak visible in RSS charts, a deadlock producing a sawtooth busy-thread pattern, and a tenfold latency spike when a worker moved across regions.
AI-written summary · about
chapters (28)
- 0:00Intro
- 0:50About Us
- 1:10What to Expect
- 3:50Technology Monitoring
- 4:35Know Your Audience
- 5:19Run Code in Production
- 5:58Service Failure
- 6:17Debugging
- 7:31Knowledge
- 7:48Data
- 9:13Agent Model
- 10:11Collection Pipeline
- 10:56Timeseries DB
- 11:20ODS
- 12:07Logs
- 12:41Structure
- 13:15Scoob
- 14:01Facebook
- 14:19Service Frameworks
- 15:11Alarms
- 15:52Practices
- 17:22War Stories
- 17:44ODS System
- 18:07Metrics
- 19:19Data Validation
- 19:53Phantom Glitch
- 20:46Wrong Assumptions
- 20:57Soundbites
related talks
09mMonitoring? Not OPS problem
35mMonitoring Microservices
30mBest practices for application observability at scale
33mEffective observability in microservice architectures
47mMonitoring at Scale: Intuitive Dashboard Design
52mHolistic Monitoring
44mActionable insight with derived second order monitoring data
41mMetrics Applied