Previously we've measured the memory consumption for every received block, but ideally we'd instrument the binary similarly to how we do it with perf to make sure the peaks and trends are also captured.
We can likely do that with valgrind as long as they play nice with perf. This would be for an instrumented run, we can't rely on the speed characteristics of an instrumented run, but it shouldn't affect the flame graphs significantly.
This should help us understand any memory changes of any PR (e.g. bitcoin#30039 or bitcoin#31645)
Previously we've measured the memory consumption for every received block, but ideally we'd instrument the binary similarly to how we do it with
perfto make sure the peaks and trends are also captured.We can likely do that with
valgrindas long as they play nice withperf. This would be for an instrumented run, we can't rely on the speed characteristics of an instrumented run, but it shouldn't affect the flame graphs significantly.This should help us understand any memory changes of any PR (e.g. bitcoin#30039 or bitcoin#31645)