[Gluster-users] symbolic link problems

[Keith Freedman]

I'm testing now with qa60, so hopefully that includes the patches you 
specified.
I'm testing the symbolic link issue as well as a couple other problems.

here are my findings  (my config is at the bottom of the email, but 
it's a basic 2 server AFR config with each other as clients and no 
other clients).

1) symlinks
I created a symlink to a non-existing file.  this went fine and 
appeared on both underlying filesystems.
I removed the link from one of the underlying filesystems and tried 
to do an ls of it
here's the results:
]# ls -al /home/testlink
/bin/ls: /home/testlink: No such file or directory
/bin/ls: cannot read symbolic link /home/testlink: No such file or directory
lrwxrwxrwx 1 root root 10 2008-11-17 21:08 /home/testlink

gluster logs:
2008-11-17 21:08:41 E [posix.c:1074:posix_chmod] home1: lstat on 
/gluster/home/testlink failed: No such file or directory
2008-11-17 21:08:41 E 
[afr-self-heal-metadata.c:296:afr_sh_metadata_sync_cbk] home: setting 
attributes failed for /testlink on posix-locks-home1 (No such file or 
directory)
2008-11-17 21:08:41 W [posix.c:1240:posix_utimens] home1: utimes on 
/gluster/home/testlink: No such file or directory
2008-11-17 21:08:41 E 
[afr-self-heal-metadata.c:296:afr_sh_metadata_sync_cbk] home: setting 
attributes failed for /testlink on posix-locks-home1 (No such file or 
directory)
2008-11-17 21:08:41 W 
[afr-self-heal-metadata.c:97:afr_sh_metadata_done] home: completed 
self heal of /testlink
2008-11-17 21:08:41 E [posix.c:2182:posix_getxattr] home1: listxattr 
failed on /gluster/home/testlink: No such file or directory

I can then remove the link from gluster and re-created it with 
gluster, but again, it creates properly if it's not on any node, but 
once it's on one, and it's not on the other it doesn't properly auto-heal.
the problem here is when you bring a new empty filesystem online it 
should auto-heal and ultimately sync all the data to the new node, 
but it can't in this scenario.


2) one of the AFR servers goes offline the permissions seem to be 
wrong on all the files until the other AFR server comes back online
--this problem existed in qa 56 and 57 but seems resolved in 60

I'll keep an eye on things and report if I find anything else.
Things seem to be shaping up quite nicely.

Keith


At 10:29 AM 11/17/2008, Anand Avati wrote:
>There is a bug fix in self-healing symlinks in patch-628. Can you 
>try the revision off tla? I tried to recreate a similar scenario as 
>you mentioned and things worked fine for me. The log message there 
>too has been fixed (it was not right to delete files in that case 
>from the backend). Thank you for reporting!
>
>avati
>
>2008/11/17 Keith Freedman 
><<mailto:freedman at freeformit.com>freedman at freeformit.com>
>so I got a strange error in gluster logs telling me that I need to
>delete a directory (tree) from all but the preferred node.
>
>gluster 1.4.0qa57
>
>so I deleted the thing from the undelying filesystem on one of my AFR'ed nodes
>
>I did the find command and it mostly rebuilt and autohealed.
>however, it got hung up anytime it ran into a symlink to a file that
>either doesn't exist OR
>does exist but the destination file exists outside of the underlying
>filesystem.
>
>Either seems to be pretty undesirable.
>
>I presume this is a bug.
>
>Keith
>