Andrew: Thanks so much for running the tests. Let me take a look at the debug output and then let you know if more debugging needs to be turned on. In addition, I need to stand up a VM with your version of the kernel and Emmanuel's changes to see if the code path has changed.
I'll get back in touch with you when I have something to report. Thanks again, Becky On Thu, May 17, 2012 at 4:01 PM, Andrew Savchenko <birc...@gmail.com> wrote: > Becky, > > I found that pvfs2fuse client works without this problems, I was able > to run make -j10 successfully, so problem is definitely somewhere in > the kernel module. > > For tests I changed my OFS configuration to be close to yours (see > attached new.config.log), but please note there is one significant > difference: I use a patch for 3.2 kernels by Emmanuel Florac: > > http://www.beowulf-underground.org/pipermail/pvfs2-developers/attachments/20120321/548b7c8e/patch.k3.1-3.obj > otherwise I wouldn't be able to use kernel module at all. > > Problem indeed doesn't depend on server configuration: I tried with > 15 data+meta and with 3 data+meta serves. Also this is not a compiler > issue, I tried a build with disabled CFLAGS and LDFLAGS, problem was > not affected. > > On Thu, 17 May 2012 12:20:50 -0400 Becky Ligon wrote: > > So, if you remount the filesystem, you don't get "?" when you "ls"? > > Yes. > > > If this is the case, then I suspect that OrangeFS isn't handling an error > > condition correctly, resulting in incorrect inode information. When you > > run your tests, if you will turn on OrangeFS kernel debugging on the > > client, I may see where the problem is happening. > > > > To turn on the appropriate kernel debug flags while the system is > running: > > > > echo "file" > /proc/sys/pvfs2/kernel-debug > > > > The kernel debug messages will appear in your /var/log/messages file (or > > wherever your system messages go on your system). > > > > I would only need to see the messages from a test that failed. > > I enabled file debug and ran two tests, results are attached. Aside > from kernel log I added make.log (to see failed files) and strace for > processes accessing failed files. > > From test1 it is clearly seen that kernel.log has no messages about > failed file itself and ends just before corruption. Each kernel.log > ends with debug mask change message to be sure buffers were synced. > > Maybe I should enable another debug level? > > Best regards, > Andrew Savchenko > -- Becky Ligon OrangeFS Support and Development Omnibond Systems Anderson, South Carolina
_______________________________________________ Pvfs2-developers mailing list Pvfs2-developers@beowulf-underground.org http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
