Continuous Profiling: The Missing Piece in Your Observability Puzzle
summary
Continuous profiling claims production-safe, low-overhead sampling (typically 1–2% CPU overhead, sometimes up to 5%) and continuous data retention so teams spot why spikes happen rather than just what. Implementation options include eBPF agentless profiling for kernel visibility or language agents that face Java safe‑point bias; open tools named are Pixie, Periscope, Parka. Practical tradeoffs: 100 ms sampling, symbolization and storage costs, dynamic sampling and downsampling strategies, and real wins like catching regressions or huge cost savings.
AI-written summary · about
chapters (15)
- 0:00Introduction and motivation
- 10:17Three pillars of observability
- 17:38Why profiling matters
- 23:38Traditional profiling challenges
- 27:11What is continuous profiling
- 31:04How continuous profiling works
- 37:03Overhead and trade-offs
- 41:42Challenges in adoption
- 46:28Cost optimization strategies
- 50:19Benefits and use cases
- 54:11Agent vs agentless profilers
- 58:05eBPF in continuous profiling
- 1:04:43Open source tools overview
- 1:07:19Live demo with Periscope
- 1:12:01Summary and closing
related talks
40mContinuous Profiling, the missing piece in your observability puzzle!
56mContinuous Profiling the missing piece in your observability puzzle
52mThe missing piece in your observability puzzle!
24mCommunity Driven Breakout Session: Continuous Profiling
51mProduction Profiling: What, Why and How
46mContinuous Profiling - the 4th Pillar of Observability
47mContinuous Profiling — what, why and how
36mContinuous Profiling in the Cloud-Native era