Ready Steady Crash
summary
Load testing a new internal biz ops API using Artillery against a staging replica reveals a reproducible crash: 25 RPS average, latency jumping from 67 ms to >30,000 ms, >90% 500/503s, and a short 30s crash phase peaking at 130 VUs. Profiling shows a GraphQL query fetching 600 systems drives CPU to 100%. Metrics flow to Graphite/Grafana; attempts at pool tuning, cipher profiling, indexing and upgrading Neo4j hardware fail to fix it; a generated 0x flame graph shows no app hotspot, and the failure inexplicably disappears overnight.
AI-written summary · about
chapters (13)
- 0:00Start
- 4:171. The Challenge 2. Getting started 3. Identify performance issue 4. Troubleshooting 5. Outcomes
- 7:48Choose a Testing framework
- 9:40Targeted towards continuous integration.
- 10:41Decide on the type of requests.
- 12:49Before the release of a push notification
- 15:36Crash Arrival rate of 100 virtual users/second for 30
- 18:45What should the first script look like?
- 27:39Application issue?
- 31:25Cypher tuning?
- 33:08Configuration issue with db?
- 36:33Able to measure, assess and improve the overall performance of our application
- 37:27To fully understand the performance and reliability of what you're building, you should load test your application
related talks
07mSuccess Breeds Panic: shares the story of his first security incident
39mEvery way we DDoSed our Cortex
29mThe osquery Agent. Performance Tuning @Scale
48mMove fast Build things… Safely!
29mDamn GraphQL - Defending and Attacking APIs
38mPerformance Engineering At MasterCard
35mMove fast, build things… Safely!
20mPremiere