easybuilders / easybuild-easyconfigs

A collection of easyconfig files that describe which software to build using which build options with EasyBuild.
https://easybuild.io
GNU General Public License v2.0
380 stars 704 forks source link

segmentation fault when trying to load some R extensions (kernel bug) #4833

Closed boegel closed 6 years ago

boegel commented 7 years ago

I'm seeing segmentation faults pop up on our system for some R extensions, including:

Problem can be reproduced easily outside of EasyBuild as well, for example

$ cat /etc/redhat-release
CentOS Linux release 7.3.1611 (Core)
$ uname -r
3.10.0-514.21.2.el7.ug.x86_64
$ module load R/3.4.0-intel-2017a-X11-20170314
$ R -q --no-save
> library(RWeka)
Segmentation fault

The problem is apparently caused by a security fix in recent versions of the Linux kernel, see:

A workaround that seems to dance around the problem for me (from https://github.com/s-u/rJava/issues/110):

$ export _JAVA_OPTIONS="-Xss2560k -Xmx2g"
$ R -q --no-save
> library(RWeka)
Picked up _JAVA_OPTIONS: -Xss2560k -Xmx2g
> quit()

The bug is fixed is more recent Linux versions, so just logging this here for reference, it's a temporary issue linked to particular versions of the Linux kernel, so little we can do about it.

verdurin commented 7 years ago

Seeing this while building those extensions, so thought I should mention it here.

boegel commented 7 years ago

@verdurin Can you confirm that the workaround works?

I'm not sure if it's worth the trouble to keep this issue open if it works, since I don't considering this our problem to fix, giving that it's a kernel bug?

verdurin commented 7 years ago

Well, I didn't see a method of applying the workaround for individual extensions...

boegel commented 7 years ago

Hmm, even though you can now control easyconfig parameters on a per-extension basis, I'm not sure you can just put the custom $_JAVA_OPTIONS value in place for only the extensions where it's required.

Is that worth the trouble though?

boegel commented 6 years ago

Kernel update fixed this for us, so I'm closing this.