Persistent SRE Antipatterns: Pitfalls On the...
summary
Alerts target user-experience alerts not individual boxes: wake people only when a significant fraction of a cluster fails or velocity spikes, not when host 473 runs out of memory. Build automated remediation so alerts fire only on human-actionable failures and use ML grouping and velocity checks to cut noise. Run on-call rotations big enough (about 8 people on rotation, primary+secondary) so SREs spend ≤50% time reactive. Prefer immutable images and canary rollbacks tied to business metrics; ditch fixed war rooms.
AI-written summary · about
chapters (12)
- 0:00Start
- 3:52User Experience Alerts
- 5:10What Makes a Good Alert
- 26:56Stages in the Development of an Automated Autonomous System
- 28:14Delaying Resolution
- 33:09Relationship between Sres and the Product Teams
- 33:48The Embedded Sra Model
- 34:55Consulting Model
- 38:41Risk versus Reward
- 40:37Never Get Involved in a Land War in Asia
- 44:23Impact of the Outage
- 48:31The Issues You Have with Other Teams within Your Own Organization
related talks
47mPersistent SRE Antipatterns: Pitfalls on the Road to Creating a Successful
24mHow to Destroy an SRE
33mWhat are the common Anti Patterns of SRE?
18mDeploying Humans at the Edge of SRE
25mUnleashing the potential of SRE for modern operations
39mVirtuous Cycles: Enabling SRE via automated feedback loops
18mWe need to talk about SRE burnout
52mPerformance & Stability Anti patterns