On-Call Experience at Uber
summary
Operates 4,000 microservices and ingests about 600 million aggregated events/sec, claims that integrated tooling turns noisy alerts into action. On-call dashboard links alerts to M3 graphs and a simulated black box trip runner, incident manager creates prefilled incidents, and automated rollback returns a bad deploy to a known good state. Shows concrete numbers, the tools wired together, the mitigation flow that actually clears alerts, and postmortems that feed fixes back into code and runbooks.
AI-written summary · about
chapters (13)
- 0:00Uber and on-call teams
- 2:29What on-call means at Uber
- 3:35Simulating an outage demo
- 5:44First alert and dashboard use
- 7:54Diagnosing issues tools
- 11:48Alerts versus incidents
- 13:55Incident Manager setup
- 16:25Root cause and mitigation
- 18:56Resolution and incident close
- 20:42Shift reports metrics
- 22:11Post-mortem writing
- 24:23On-call tools integrations
- 26:50On-call lessons learned
related talks
23mDevOps at Uber
30mOn call does not have to suck
24mOn-Call the Easy Way With Agents
26mBuild a data-driven on-call workflow for your team with atomic habits
31mHow to Build a Healthy On-call Culture
22mRaise a Production System:Making on-call human-scale w/ Observability & Trust
48mDevOps Transformation: Introducing Incident Management and Maximizing Monitoring Value
19mYou Can't Stop Fires with an Ambulance