Resolving performance issues in practice
summary
Methodology prescribes measuring with RED (requests errors durations) and drilling into components with USE (utilization saturation errors), using tracers/profilers, Java Mission Control/Flight Recorder or bpf utils as needed. A real incident shows 99th‑percentile latency jumping from ~100ms to 3s, the DB reporting 4,000 queued requests and 90% utilization; root cause is a BI job moved to 3pm, stopping the job and reverting its schedule restores metrics to baseline and meets acceptance criteria.
AI-written summary · about
chapters (12)
- 0:00Introduction and speaker intro
- 1:41Why performance matters
- 5:15Defining system performance
- 8:05Challenges in performance engineering
- 12:23Common anti-patterns
- 17:26Methodology for resolving issues
- 19:34Step 1: Confirming the problem
- 23:06Step 2: Setting stop points
- 25:12Step 3: Locating root causes
- 31:13Real-life example walkthrough
- 44:18Recap and mindset
- 46:27Q&A session
related talks
01h 01mThe Performance Investigator's Field Guide
30mSeeing RED
48mFantastic Performance and Where to Find It
49mResponding in a Timely Manner
38mTechniques and Tools for a Coherent Discussion about Performance in Complex...
46mTaming the 9s
32mSoftware Performance: A Shape Not a Number
35mThe RED Method: How To Instrument Your Services [B]