hello.
we use glusterfs 3.2.0
2 glusterfs-server with SLES 11.1
and several clients which access the gfs-volumes.
configuration:
info
----
type=2
count=2
status=1
sub_count=2
version=1
transport-type=0
volume-id=05168b54-6a5c-4aa3-91ee-63d16976c6cd
brick-0=10.0.1.xxx:-glusterstorage-macm03
brick-1=10.0.1.xxy:-glusterstorage-macm03
macm03-fuse.vol
---------------
volume macm03-client-0
type protocol/client
option remote-host 10.0.1.xxx
option remote-subvolume /glusterstorage/macm03
option transport-type tcp
end-volume
volume macm03-client-1
type protocol/client
option remote-host 10.0.1.xxy
option remote-subvolume /glusterstorage/macm03
option transport-type tcp
end-volume
volume macm03-replicate-0
type cluster/replicate
subvolumes macm03-client-0 macm03-client-1
end-volume
volume macm03-write-behind
type performance/write-behind
subvolumes macm03-replicate-0
end-volume
volume macm03-read-ahead
type performance/read-ahead
subvolumes macm03-write-behind
end-volume
volume macm03-io-cache
type performance/io-cache
subvolumes macm03-read-ahead
end-volume
volume macm03-quick-read
type performance/quick-read
subvolumes macm03-io-cache
end-volume
volume macm03-stat-prefetch
type performance/stat-prefetch
subvolumes macm03-quick-read
end-volume
volume macm03
type debug/io-stats
subvolumes macm03-stat-prefetch
end-volume
macm03.10.0.1.xxx.glusterstorage-macm03.vol
-------------------------------------------
volume macm03-posix
type storage/posix
option directory /glusterstorage/macm03
end-volume
volume macm03-access-control
type features/access-control
subvolumes macm03-posix
end-volume
volume macm03-locks
type features/locks
subvolumes macm03-access-control
end-volume
volume macm03-io-threads
type performance/io-threads
subvolumes macm03-locks
end-volume
volume /glusterstorage/macm03
type debug/io-stats
subvolumes macm03-io-threads
end-volume
volume macm03-server
type protocol/server
option transport-type tcp
option auth.addr./glusterstorage/macm03.allow *
subvolumes /glusterstorage/macm03
end-volume
macm03.10.0.1.xxy.glusterstorage-macm03.vol
-------------------------------------------
volume macm03-posix
type storage/posix
option directory /glusterstorage/macm03
end-volume
volume macm03-access-control
type features/access-control
subvolumes macm03-posix
end-volume
volume macm03-locks
type features/locks
subvolumes macm03-access-control
end-volume
volume macm03-io-threads
type performance/io-threads
subvolumes macm03-locks
end-volume
volume /glusterstorage/macm03
type debug/io-stats
subvolumes macm03-io-threads
end-volume
volume macm03-server
type protocol/server
option transport-type tcp
option auth.addr./glusterstorage/macm03.allow *
subvolumes /glusterstorage/macm03
end-volume
client
------
the client has mounted the volume via fstab like this:
server:/macm03 /srv/www/GFS glusterfs defaults,_netdev 0 0
now we registered strange behavior and i have some questions:
1) files with size 0
we find many files with size 0. in server-log we only find this. what does this mean?
(most all files in one directory has size 0).
[2011-04-28 23:52:00.630869] I [server-resolve.c:580:server_resolve] 0-macm03-server: pure path
resolution for /xxx/preview/4aa76fa541413.jpg (LOOKUP)
[2011-04-28 23:52:00.637384] I [server-resolve.c:580:server_resolve] 0-macm03-server: pure path
resolution for /xxx/preview/4aa76fa541413.jpg (UNLINK)
[2011-04-28 23:52:00.693183] I [server-resolve.c:580:server_resolve] 0-macm03-server: pure path
resolution for /xxx/preview/4aa76fa541413.jpg (LOOKUP)
[2011-04-28 23:52:00.711092] I [server-resolve.c:580:server_resolve] 0-macm03-server: pure path
resolution for /xxx/preview/4aa76fa541413.jpg (MKNOD)
[2011-04-28 23:52:00.746289] I [server-resolve.c:580:server_resolve] 0-macm03-server: pure path
resolution for /xxx/preview/4aa76fa541413.jpg (SETATTR)
[2011-04-28 23:52:16.373532] I [server-resolve.c:580:server_resolve] 0-macm03-server: pure path
resolution for /xxx/preview/4aa76fa541413.jpg (LOOKUP)
2) then the client is selfhealing meta-data all the time.... (because the file has has size 0 on one
of the servers???)
but we triggered selfhealing severalt time like this:
http://europe.gluster.org/community/documentation/index.php/Gluster_3.1:_Triggering_Self-Heal_on_Replicate
[2011-04-29 07:55:27.188743] I [afr-common.c:581:afr_lookup_collect_xattr] 0-macm03-replicate-0:
data self-heal is pending for /videos12/29640/preview/4aadf4b757de6.jpg.
[2011-04-29 07:55:27.188829] I [afr-common.c:735:afr_lookup_done] 0-macm03-replicate-0: background
meta-data data self-heal triggered. path: /videos12/29640/preview/4aadf4b757de6.jpg
[2011-04-29 07:55:27.194446] W [dict.c:437:dict_ref]
(-->/opt/glusterfs/3.2.0/lib64/glusterfs/3.2.0/xlator/protocol/client.so(client3_1_fstat_cbk+0x2bb)
[0x2aaaaafe833b]
(-->/opt/glusterfs/3.2.0/lib64/glusterfs/3.2.0/xlator/cluster/replicate.so(afr_sh_data_fstat_cbk+0x17d)
[0x2aaaab11c9ad]
(-->/opt/glusterfs/3.2.0/lib64/glusterfs/3.2.0/xlator/cluster/replicate.so(afr_sh_data_fix+0x1fc)
[0x2aaaab11c64c]))) 0-dict: dict is NULL
3)
on some of the clients we then can not access the whole directory:
# dir xxx/preview/
/bin/ls: reading directory xxx/preview/: File descriptor in bad state
total 0
in logs we find this:
[2011-04-29 08:36:17.224301] W [afr-common.c:634:afr_lookup_self_heal_check] 0-macm03-replicate-0:
/videos12/30181: gfid different on subvolume
[2011-04-29 08:36:17.241330] I [afr-common.c:680:afr_lookup_done] 0-macm03-replicate-0: entries are
missing in lookup of /xxx/preview.
[2011-04-29 08:36:17.241373] I [afr-common.c:735:afr_lookup_done] 0-macm03-replicate-0: background
meta-data data entry self-heal triggered. path: /xxx/preview
[2011-04-29 08:36:17.243160] I [afr-self-heal-metadata.c:595:afr_sh_metadata_lookup_cbk]
0-macm03-replicate-0: path /videos12/30181/preview on subvolume macm03-client-0 => -1 (No such file
or directory)
[2011-04-29 08:36:17.302228] I [afr-dir-read.c:120:afr_examine_dir_readdir_cbk]
0-macm03-replicate-0: /videos12/30181/preview: failed to do opendir on macm03-client-0
[2011-04-29 08:36:17.303836] I [afr-dir-read.c:174:afr_examine_dir_readdir_cbk]
0-macm03-replicate-0: entry self-heal triggered. path: /xxx/preview, reason: checksums of directory
differ, forced merge option set
4)
sometimes when we umount glusterfs-volume on client and remount it again, we can access the dirctory
which was in bad state before -> and then also selfhealing works at it should
but sometimes also a remount does not work.
any help would be appreciated.
thank you very very much!
thx
christopher
