Emmanuel Dreyfus <manu@xxxxxxxxxx> wrote: > > Can you check if the 'acronyms' directory has the same gfid on all 4 servers > > (from the backend)? > Yes, it is the same on the 2 bricks (this is a Distributed-Replicate > with 2x2 bricks) I overlooked the client log. Here is it: [2011-07-09 02:39:14.798565] I [afr-common.c:902:afr_fresh_lookup_cbk] 0-gfs-replicate-1: added root inode [2011-07-09 02:39:14.798862] I [afr-common.c:801:afr_lookup_done] 0-gfs-replicate-1: background meta-data entry self-heal triggered. path: / [2011-07-09 02:39:14.800398] I [afr-self-heal-common.c:537:afr_sh_mark_sources] 0-gfs-replicate-1: split-brain possible, no source detected [2011-07-09 02:39:14.800552] E [afr-self-heal-metadata.c:518:afr_sh_metadata_fix] 0-gfs-replicate-1: Unable to self-heal permissions/ownership of '/' (possible split-brain). Please fix the file on all backend volumes [2011-07-09 02:39:14.808920] I [afr-self-heal-metadata.c:81:afr_sh_metadata_done] 0-gfs-replicate-1: split-brain detected, aborting selfheal of / [2011-07-09 02:39:14.809079] E [afr-self-heal-common.c:1554:afr_self_heal_completion_cbk] 0-gfs-replicate-1: background meta-data entry self-heal failed on / The permissions/ownerships seems okay, though: $ ls -ld /export/wd3a/ drwxr-xr-x 3 root wheel 512 Jul 8 09:53 /export/wd3a/ $ ls -ld /export/wd3a/ drwxr-xr-x 3 root wheel 512 Jul 8 09:53 /export/wd3a/ But here is a difference. According to the sources, these are used to detect split brain: silo# getextattr -x system trusted.afr.gfs-client-3 /export/wd3a/ /export/wd3a/ 00 00 00 00 00 00 00 02 00 00 00 00 silo# getextattr -x system trusted.afr.gfs-client-2 /export/wd3a/ /export/wd3a/ 00 00 00 00 00 00 00 00 00 00 00 00 hangar# getextattr -x system trusted.afr.gfs-client-3 /export/wd3a/ /export/wd3a/ 00 00 00 00 00 00 00 00 00 00 00 00 hangar# getextattr -x system trusted.afr.gfs-client-2 /export/wd3a/ /export/wd3a/ 00 00 00 00 00 00 00 02 00 00 00 01 I tried removing these attributes and the errors are gone, but the problem remain: none of the file from these two bricks are seen by readdir. -- Emmanuel Dreyfus http://hcpnet.free.fr/pubz manu@xxxxxxxxxx
