fix fd leak

[yanglong1010]

hi,

Our project uses the java-profiler code, and the Java process has a Too many open files error. After using strace -k -e trace=openat -f java-pid, I feel that I may have found the reason. After I modified it slightly, I have verified that the problem can be solved. I don't know if I can contribute a little code to java-profiler.

[WARN] perf_event_open for TID 30704 failed: Too many open files

ulimit -n
256

lsof -p 30499|grep '/proc/30499/task'|wc -l
202

lsof -p 30499|grep '/proc/30499/task'|head
java 30499 root 23r DIR 0,3 0 61771716 /proc/30499/task
java 30499 root 24r DIR 0,3 0 61771716 /proc/30499/task
java 30499 root 51r DIR 0,3 0 61771716 /proc/30499/task
java 30499 root 52r DIR 0,3 0 61771716 /proc/30499/task
java 30499 root 53r DIR 0,3 0 61771716 /proc/30499/task
java 30499 root 54r DIR 0,3 0 61771716 /proc/30499/task
java 30499 root 55r DIR 0,3 0 61771716 /proc/30499/task
java 30499 root 56r DIR 0,3 0 61771716 /proc/30499/task
java 30499 root 57r DIR 0,3 0 61771716 /proc/30499/task
java 30499 root 58r DIR 0,3 0 61771716 /proc/30499/task

[richardstartin]

Hi @yanglong1010 it's interesting that you're running the wallclock profiler without the tracer-guided thread filter - we were considering removing the fallback mode. Are you running the profiler without dd-java-agent, or are you setting -Ddd.profiling.ddprof.wall.context.filter=false?

[yanglong1010]

Yes, we did not use dd-java-agent, but a java agent written by ourselves.

I found that you recorded the span id in the wall clock sample event. This is a very good idea. I ported this part of the code into our project to achieve a similar continuous profiling effect. When running in continuous mode, we set the thread filter and everything is fine. The other mode is a one-time short-term opening without setting the thread filter, so it will enter the fallback mode.

[richardstartin]

@yanglong1010 Thank you so much for this patch. This is a public mirror of a repository we do new work in internally, so I can't merge your patch here, but I will apply your commit manually and sync it here later.

Further patches are very welcome, and if you ever want to discuss features or give feedback, please email me at richard [dot] startin [at] datadoghq.com

[richardstartin]

@yanglong1010 your commit was merged here 6f95dc4