[Gluster-devel] Specific bug question
Stephan von Krawczynski
skraw at ithnet.com
Thu Aug 23 18:53:08 UTC 2012
I have a question for people with indepth glusterfs code knowledge.
Lets assume I have a 2 brick distributed setup with one client and a mounted path /gluster.
brick1 exports /brick1, brick2 exports /brick2
I do the following in bash:
chown dummyuser:dummygroup /tmp/testfile
mv /tmp/testfile /gluster/dir1/
mv /gluster/dir1/testfile /gluster/dir2/
this works most of the time. But sometimes I see this on one brick (not always the same brick):
[2012-08-23 20:30:01] E [posix.c:1515:posix_chown] posix: lchown on /dir1/testfile failed: No such file or directory
[2012-08-23 20:30:01] E [posix.c:1461:posix_chmod] posix: chmod on /dir1/testfile failed: No such file or directory
[2012-08-23 20:30:01] E [posix.c:1636:posix_utimens] posix: utimes on /brick1/dir1/testfile failed: No such file or dir
Of course the last line can also be
[2012-08-23 20:30:01] E [posix.c:1636:posix_utimens] posix: utimes on /brick1/dir2/testfile failed: No such file or dir
depending on the brick.
If you look at brick1/2 afterwards you see that on the brick with these errors testfile is own by root:root, whereas client shows correct owner dummyuser:dummygroup, just like the brick without the error does. The problem is not solved by selfheal.
Now my question is: is this a solved bug? Can someone point me to a patch? I am not using the latest glusterfs and don't want to. I am only looking for information to backport the bugfix.
The client uses replicate-iocache-writebehind-iothreads btw.
The bug obviously shows that mv across glusterfs does not work atomic and the fs is not even serializing fs actions concerning the same file.
More information about the Gluster-devel