Description of problem: I have a volume divided into 3 bricks (one per server). Sometimes, a brick goes down. The only error I found is in the /var/log/glusterfs/brick/glusterfs-projets-brick2-projets.log log file : http://fpaste.org/118116/14054353/ I have a talk with ndevos on IRC and he asked me the result of : gdb /usr/lib/x86_64-linux-gnu/glusterfs/3.5.1/xlator/features/marker.so l *mq_loc_fill_from_name+0x86 which is : 0x14896 is in mq_loc_fill_from_name (marker-quota.c:178). 173 marker-quota.c: No such file or directory. I'm using glusterfs 3.5.1 on a Debian Wheezy server and I use NFS, Quota and ACL. # gluster volume info projets Volume Name: projets Type: Distribute Volume ID: ee585e7a-1b89-40b5-bd37-1c977920e87f Status: Started Number of Bricks: 3 Transport-type: tcp Bricks: Brick1: projet1:/glusterfs/projets-brick1/projets Brick2: projet2:/glusterfs/projets-brick2/projets Brick3: projet3:/glusterfs/projets-brick3/projets Options Reconfigured: performance.cache-size: 256MB features.quota-timeout: 0 features.quota: on
Bug 1118591 looks very much the same. This bug will handle the issue in glusterfs-3.5.x, where bug 1118591 is used to include a fix in the master branch.
(In reply to Thilam from comment #0) > The only error I found is in the > /var/log/glusterfs/brick/glusterfs-projets-brick2-projets.log log file : > http://fpaste.org/118116/14054353/ /usr/lib/x86_64-linux-gnu/glusterfs/3.5.1/xlator/features/marker.so(mq_loc_fill_from_name+0x86)[0x7f654cb48896] /usr/lib/x86_64-linux-gnu/glusterfs/3.5.1/xlator/features/marker.so(mq_readdir_cbk+0x227)[0x7f654cb49367] /usr/lib/x86_64-linux-gnu/libglusterfs.so.0(default_readdir_cbk+0xc0)[0x7f6552a22460] > I have a talk with ndevos on IRC and he asked me the result of : > gdb /usr/lib/x86_64-linux-gnu/glusterfs/3.5.1/xlator/features/marker.so > l *mq_loc_fill_from_name+0x86 > > which is : > 0x14896 is in mq_loc_fill_from_name (marker-quota.c:178). > 173 marker-quota.c: No such file or directory. $ git show v3.5.1:xlators/features/marker/src/marker-quota.c ... 176 len = strlen (oldloc->path); 177 178 if (oldloc->path [len - 1] == '/') ... Line marker-quota.c:178 caused the segfault.
(In reply to Niels de Vos from comment #2) > (In reply to Thilam from comment #0) > > The only error I found is in the > > /var/log/glusterfs/brick/glusterfs-projets-brick2-projets.log log file : > > http://fpaste.org/118116/14054353/ > > /usr/lib/x86_64-linux-gnu/glusterfs/3.5.1/xlator/features/marker. > so(mq_loc_fill_from_name+0x86)[0x7f654cb48896] > /usr/lib/x86_64-linux-gnu/glusterfs/3.5.1/xlator/features/marker. > so(mq_readdir_cbk+0x227)[0x7f654cb49367] > /usr/lib/x86_64-linux-gnu/libglusterfs.so. > 0(default_readdir_cbk+0xc0)[0x7f6552a22460] > Seems similar to bz 1118591
Backporting http://review.gluster.org/8296 should probably fix it.
The patch has been backported already for bug 1144315. *** This bug has been marked as a duplicate of bug 1144315 ***