Using bpftrace with Performance Co-Pilot & Grafana
summary
Combines bpftrace, Performance Co‑Pilot and Grafana to run bpftrace scripts from the browser: a PCP PMDA starts BPFtrace on demand and stops it when panels close, PMProxy exposes metrics via REST on port 44322, and a Grafana plugin renders heatmaps, flame graphs and tables. Examples: vfs_read bytes shown as a live heatmap, CPU run‑queue latency histogram in microsecond buckets, TCP connect events exported as CSV and limited to 10 table rows. Provides a container with all deps on port 3000; requires patched bpftrace for newer kernels.
AI-written summary · about
chapters (14)
- 0:00BPFtrace with PCP & Grafana
- 3:57BPFtrace fundamentals & histograms
- 6:28PCP BPFtrace architecture
- 8:12Fedora 31 installation
- 10:00BPFtrace tools & Grafana
- 12:51CSV & tables in Grafana
- 13:34Kernel stacks & flame graphs
- 14:40Quick container setup
- 15:22BPFtrace dashboards demo
- 16:46Metrics to Kafka & TSDB
- 17:29PCP archives analysis
- 18:16What is PCP
- 20:04TCP connections demo
- 21:05CPU core utilization
related talks
21mPerformance Wins with BPF: Getting Started
36mObservability and performance analysis with BPF
28mbpftrace: what is it, what's new, and where is it going
25mKernel tracing with BPF
35mbpftrace: Evolution from a command line tool to a complex observability framework
27mLow-Overhead Tracing Using eBPF for Observability into Kubernetes Apps and Services
45mPerformance lab. The power of the perf tool!
32meBPF: BPF kernel infrastructure