No `perf.data` generated within Docker container

[mateusfigmelo]

I run the my rust app with cargo flamegraph in my local environment.
At that time, I could see the perf.data as temp file in my working dir.
And when the CTRL+C, I could also see the flamegraph.svg
Question 1: How to see the flamegraph.svg in real time without CTRL+C

And then I needed to run the cargo flamegraph with docker compose.
So I wrote the docker compose including cargo flamegraph.
After running docker, all processes were running correctly, but I couldn't see the perf.data in working dir of docker container.
Instead of generating perf.data, I could see too long logs of docker containter (docker logs -f xxx) including known unicode letters.
Maybe it seems like the contents of perf.data.
I am not sure why perf.data is not generated within docker container.
But I tried to run the command cargo flamegraph within docker container.
At that time, perf.data was generated within working dir of docker container.
Question 2:
Why the results of run the command as entrypoint of docker compose and as direct run within docker container are different?

[djc]

From a quick look at the source code, I don't see any reason for a difference in behavior between running natively in a Linux (terminal) vs running in a Docker container. However, flamegraph is a fairly thin wrapper over perf (at least on Linux) and the inferno crate, so it's possible that perf behaves differently in your environment.

Beyond, I probably won't be able to troubleshoot your issue much more on a volunteer basis. If you want to discuss commercial support, feel free to send me an email.

[mateusfigmelo]

Thanks for your response.
Anyways, I took out the part to check the performance from docker compose for now.
In the future, I will need to check the performance within docker container.
If you find relevant datas with it, plz lmk.
🙏

[RealJoL]

I have encountered this problem before (including binary/Unicode on std output). Have you verified that the container has the required capabilities (at least PERFMON or even privileged) to use perf in your compose deployment?

In the past, manually starting the flame graph via docker exec or wrapping the binary call into a .sh file also worked for me, but your mileage may vary.