Frederic Branczyk
Frederic Branczyk
Random thought: It would be incredibly useful if we could hook into the OOM-Killer and take a heap profile just before the process is killed.
If the JVM is started with the above-linked perf-map-agent then soon there will be support. The important one is https://github.com/parca-dev/parca-agent/pull/101, but we still lack support for translating container PID to...
Starting with the [`v0.3.0`](https://github.com/parca-dev/parca-agent/releases/tag/v0.3.0) release if the perf-map agent is used, then JVM can be profiled with parca-agent today!
Looks like there is also some precedence of profilers at the intersection of eBPF and java, that some inspiration could come from: https://github.com/weixingsun/jBProF
I think I'd like to see us building full profiles with the stack traces that call the allocation causing functions. I think it would be neat to build two separate...
It does not seem to me that the libbpfgo one handles address ranges correctly. I distinctly remember that ranges were important and lookups do not work with exact addresses. Eg....
We can propose it, but I suspect there’s a bit of a trade off at play. We traded off memory for spending more cpu. It depends a lot on how...
That would be super awesome I didn't even know about the JITDUMP format! If the erlang stacks always require unwinding then unfortunately that's not something we can do today (@kakkoyun...
That is amazing!!!
Could we maybe even have the Parca Agent do a few tests on startup to identify whether the features it needs are available and error out with that exact feature...