Java off heap allocation profiling (with mprotect event) issue

[RoySunnySean007]

Dear experts,

According to https://youtu.be/c755fFv1Rnk, I use profiler.sh -d <duration> -e mprotect -f <absolute path of html file> <java pid> to profile java off-heap object allocation, but interestingly, there are ~50% samples just show mprotect.
My JDK is JDK11, async version is 2.9, and I run java application in container, and set kernel.perf_event_paranoid=1 and kernel.kptr_restrict=0 in the host machine.
Could u please kindly let me know if there is any wrong? Thanks in advance!

sh-4.2# ./profiler.sh -v
Async-profiler 2.9 built on Nov 26 2022
Copyright 2016-2021 Andrei Pangin
sh-4.2# ./profiler.sh -d 60 -e mprotect -f /tmp/sample.html 1 
Profiling for 60 seconds
sh-4.2# java -version
openjdk version "11.0.10" 2021-01-19 LTS
OpenJDK Runtime Environment Corretto-11.0.10.9.1 (build 11.0.10+9-LTS)
OpenJDK 64-Bit Server VM Corretto-11.0.10.9.1 (build 11.0.10+9-LTS, mixed mode)
sh-4.2# 

[apangin]

The default stack walking algorithm cannot get a stack trace of mprotect, because the system library is apparently compiled with omit-frame-pointer optimization.

Try --cstack dwarf option - this should help to recover correct stack traces for mprotect.

[RoySunnySean007]

It works, turned out to be GC related. Thanks @apangin !

[RoySunnySean007]

Dear @apangin , sorry to interrupt you again...

My previous assumption is GC related, am I right? If possible, could u please kindly let me know how can I correlate these native async profiler samples to specific functions? Thanks in advance!

There are two parts: deoptimization & _GI_Clone

[apangin]

I'm not sure I understand. Neither of these stack traces come from GC.

This SO answer explains how to find cause of deoptimization. In any case, I don't think deoptimization relates to off-heap memory leak. From your profile, netty buffers are more likely suspect.

[RoySunnySean007]

Thanks @apangin for ur clear explanation with SO answer!