Java Profiling of long running MapReduce container

[#1]

Hi dear community!

i tried to track MR container (mapper) through Java Mission Control (JMC) and through visualvm.

In both cases I got some error:

if i would profile spark - no problem.
Goal of this action to understand which Java classes consume much more CPU resource.
could anybody tell how to fix this profiling or which tool could show which Java Classes use much more resources in MR container.

also add example with visualvm with NodeManager (it doesn't works as well as MR container, and Spark worker profile - it's what i want to get)

what I want to get:

thanks in advance!

[#2 · Accepted solution]

What user are the MR and Spark containers run as? The Yarn processes all run as user "yarn", and I wonder if a user-name mismatch is causing issues.

Also, what user are you running the profiling tools?

[#3]

thanks a lot! it works!