Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
Hadoop >> mail # user >> Java->native .so->seg fault->core dump file?

Copy link to this message
Java->native .so->seg fault->core dump file?
My Java mappers use JNI to call native .so files compiled from C++.  In some cases, the task status ends with exit 139, which generally indicates a seg-fault.  I would like to see the core-dump, but I can't seem to get it to work.

I have this line in my driver setup (Yes, it's v.19):
conf.setBoolean("keep.failed.task.files", true);

and I have this in my .bashrc (which I believe should be propagated to the slave nodes):
ulimit -c unlimited

and in my native code I call rlimit() and write the results, where I see:
RLIMIT_CORE:  18446744073709551615     18446744073709551615

which indicates the "unlimited" setting, but I can't find any core dump files in the node's hadoop directories after the job runs.

Any ideas what I'm doing wrong?

Keith Wiley               [EMAIL PROTECTED]               www.keithwiley.com

"It's a fine line between meticulous and obsessive-compulsive and a slippery
rope between obsessive-compulsive and debilitatingly slow."
  -- Keith Wiley