Red Hat Bugzilla – Bug 877988
creating hard link fails with 'File Exist' message even though the hard link file doesn't exist.
Last modified: 2013-11-25 08:49:07 EST
Description of problem: ======================== In a replicate volume (1x2) , creating hard link fails with 'File Exist' message even though the hard link file doesn't exist. Version-Release number of selected component (if applicable): ============================================================== [11/19/12 - 02:45:35 root@king ~]# gluster --version glusterfs 3.3.0.5rhs built on Nov 8 2012 22:30:35 How reproducible: ================= Often Steps to Reproduce: ==================== 1.Create a replicate volume (1x2) . 2. set self-heal-daemon volume option to off 3. start the volume 4. create a fuse mount 5. create a directory "testdir" 6. create a file under "testdir" : dd if=/dev/urandom of=testdir/a bs=1k count=1024 7. bring down brick1 8. create hard link to the file "a" : "ln testdir/a testdir/b" 9. bring back brick1 and immediately bring down brick2 10. create hard link to the file "a" : "ln testdir/a testdir/b" Actual results: ================== [11/19/12 - 05:51:18 root@flea rep]# ln testdir/a testdir/b ln: creating hard link `testdir/b': File exists [11/19/12 - 05:53:58 root@flea rep]# ls testdir [11/19/12 - 05:54:01 root@flea rep]# ls testdir/ a Expected results: =================== Should allow to create the hard link "testdir/b"to the file "testdir/a"
Seems to be a meta-data-cache translator issue. If we disable stat-prefetch which in turn disables meta-data-cache. ln succeeds. Remove the commented out state-prefetch volume set to see the issue. I automated the script for ease of re-creating the issue: HOSTNAME=`hostname` mkdir /mnt/r2 mkdir /gfs gluster --mode=script volume create r2 replica 2 $HOSTNAME:/gfs/r2_0 $HOSTNAME:/gfs/r2_1 gluster --mode=script volume start r2 gluster --mode=script volume set r2 client-log-level DEBUG gluster --mode=script volume set r2 brick-log-level DEBUG #gluster --mode=script volume set r2 stat-prefetch off mount -t glusterfs $HOSTNAME:/r2 /mnt/r2 cd /mnt/r2 mkdir /mnt/r2/testdir dd if=/dev/urandom of=/mnt/r2/testdir/a bs=1k count=1024 kill -9 $(cat /var/lib/glusterd/vols/r2/run/$HOSTNAME-gfs-r2_0.pid) ln /mnt/r2/testdir/a /mnt/r2/testdir/b gluster --mode=script volume start r2 force && kill -9 $(cat /var/lib/glusterd/vols/r2/run/$HOSTNAME-gfs-r2_1.pid) sleep 1 ln /mnt/r2/testdir/a /mnt/r2/testdir/b Here is the run with stat-prefetch disabled: [root@pranithk-laptop ~]# glusterd && bash -x /home/pranithk/workspace/rhs-glusterfs/877988.sh ++ hostname + HOSTNAME=pranithk-laptop + mkdir /mnt/r2 mkdir: cannot create directory `/mnt/r2': File exists + mkdir /gfs mkdir: cannot create directory `/gfs': File exists + gluster --mode=script volume create r2 replica 2 pranithk-laptop:/gfs/r2_0 pranithk-laptop:/gfs/r2_1 Creation of volume r2 has been successful. Please start the volume to access data. + gluster --mode=script volume start r2 Starting volume r2 has been successful + gluster --mode=script volume set r2 client-log-level DEBUG Set volume successful + gluster --mode=script volume set r2 brick-log-level DEBUG Set volume successful + gluster --mode=script volume set r2 stat-prefetch off Set volume successful + mount -t glusterfs pranithk-laptop:/r2 /mnt/r2 + cd /mnt/r2 + mkdir /mnt/r2/testdir + dd if=/dev/urandom of=/mnt/r2/testdir/a bs=1k count=1024 1024+0 records in 1024+0 records out 1048576 bytes (1.0 MB) copied, 0.13735 s, 7.6 MB/s ++ cat /var/lib/glusterd/vols/r2/run/pranithk-laptop-gfs-r2_0.pid + kill -9 31244 + ln /mnt/r2/testdir/a /mnt/r2/testdir/b + gluster --mode=script volume start r2 force Starting volume r2 has been successful ++ cat /var/lib/glusterd/vols/r2/run/pranithk-laptop-gfs-r2_1.pid + kill -9 31250 + sleep 1 + ln /mnt/r2/testdir/a /mnt/r2/testdir/b [root@pranithk-laptop ~]# ls -l /mnt/r2/testdir/ total 2046 -rw-r--r-- 2 root root 1047552 Nov 20 12:26 a -rw-r--r-- 2 root root 1047552 Nov 20 12:26 b Here is the run with stat-prefetch enabled: [root@pranithk-laptop ~]# glusterd && bash -x /home/pranithk/workspace/rhs-glusterfs/877988.sh ++ hostname + HOSTNAME=pranithk-laptop + mkdir /mnt/r2 mkdir: cannot create directory `/mnt/r2': File exists + mkdir /gfs mkdir: cannot create directory `/gfs': File exists + gluster --mode=script volume create r2 replica 2 pranithk-laptop:/gfs/r2_0 pranithk-laptop:/gfs/r2_1 Creation of volume r2 has been successful. Please start the volume to access data. + gluster --mode=script volume start r2 Starting volume r2 has been successful + gluster --mode=script volume set r2 client-log-level DEBUG Set volume successful + gluster --mode=script volume set r2 brick-log-level DEBUG Set volume successful + mount -t glusterfs pranithk-laptop:/r2 /mnt/r2 + cd /mnt/r2 + mkdir /mnt/r2/testdir + dd if=/dev/urandom of=/mnt/r2/testdir/a bs=1k count=1024 1024+0 records in 1024+0 records out 1048576 bytes (1.0 MB) copied, 0.136642 s, 7.7 MB/s ++ cat /var/lib/glusterd/vols/r2/run/pranithk-laptop-gfs-r2_0.pid + kill -9 31531 + ln /mnt/r2/testdir/a /mnt/r2/testdir/b + gluster --mode=script volume start r2 force Starting volume r2 has been successful ++ cat /var/lib/glusterd/vols/r2/run/pranithk-laptop-gfs-r2_1.pid + kill -9 31538 + sleep 1 + ln /mnt/r2/testdir/a /mnt/r2/testdir/b ln: failed to create hard link `/mnt/r2/testdir/b': File exists [root@pranithk-laptop ~]# ls -l /mnt/r2/testdir total 1023 -rw-r--r-- 1 root root 1047552 Nov 20 12:28 a [root@pranithk-laptop ~]#
Pranith, Thanks for the quick turn-around time on this. Avati, do you think this should be marked as KnownIssue and close it as WONTFIX? (as workaround exists)... Let me know what you think. I don't see any particular dev effort for this particular bug.
updated the doc text. Avati, need your input on this.
-------- Cause: with 'md-cache' translator in the graph, quick entry operations may have few issues when replicate bricks are not connected properly (ie, one brick down, while other is up). Consequence: We would hit EEXIST error for a link() (hardlink) call. Workaround (if any): 'gluster volume set <VOL> stat-prefetch off' Result: The issues are not seen. -------- Please re-open if the work around is not working :-)
This is documented as a known issue for the Big Bend Update 1 Release Notes. Here is the link: http://documentation-devel.engineering.redhat.com/docs/en-US/Red_Hat_Storage/2.1/html/2.1_Update_1_Release_Notes/chap-Documentation-2.1_Update_1_Release_Notes-Known_Issues.html