As any respectable software performance person knows, perf is your best (only?) friend. For example, perf report -g has shined a light into the deepest, darkest corners of debugging territory. Since you asked, it can happily run in a container, too (albeit requiring elevated privileges, but we’re debugging here…).
Typically console formatted output is fine for grokking perf reports, but having recently become addicted to go’s pprof visualization (dot format), handy flame graphs, and on the morbid occassion, VTune, I started looking around for a way to more clearly understand a particular perf recording.
Googling turned up an interesting QT-based tool called hotspot by a company called KDAB. Screenshots indicate it might be worth kicking the tires.
After some bouncing around figuring out Fedora equivalent package names, I was able to quickly build and run hotspot. I ran a quick perf record to see if it was going to work at all:
$ sudo perf record --call-graph dwarf sleep 10 $ ./bin/hotspot ./perf.data
Folks at KDAB even included a built-in flame graph:
Interface is clean, bug-free and useful. Trying to load a large perf.data file was a bit ugly and RAM-intensive; I would likely stick to command-line parsing for those. Or, as we do in pbench, reduce the collection frequency to 100Hz and take bite-sized samples over the life of the test.