32. #EveryoneCanContribute cafe: Continuous Profiling with Polar Signals
Continuous Profiling - Measure and understand performance over time.
- Continuous Profiling defining 4 pillars of Observability: Metrics, Traces, Logs, Profiles
- Open standard: pprof
- Polar Signals use Prometheus Service Discovery capabilities. Or push the data every X seconds.
- Query language similar to PromQL.
- Share views as a public URL, e.g. in PR/MRs for team collaboration.
- Merge queries and reports to see which code change impacted the performance
- Push metric samples to Thanos, and keep pprof profiles. How to present this over time? (to be solved)
- Idea from Andrew: OpenTelemetry?
- Automated profiling as an idea, follow Polar Signals on Twitter for updates.
- Optimize for debug symbols - try to only upload once, and link them.
- Recommendation engine: “Save time with adding this code change”.