[Gluster-users] Self healing does not see files to heal
Pranith Kumar Karampuri
pkarampu at redhat.com
Sat Aug 20 11:22:42 UTC 2016
We do need to consider this as bug and fix full self-heal to handle the
case where it has to look at both the bricks to see if there are any files
missing in the bricks. We won't be letting this happen on the mounts though
because it will slow down performance. Be very careful about deleting files
directly from the brick though. It is always recommended you take back up
of the good file before attempting heal.
On Wed, Aug 17, 2016 at 4:28 PM, Дмитрий Глушенок <glush at jet.msk.su> wrote:
> You are right, stat triggers self-heal. Thank you!
>
> --
> Dmitry Glushenok
> Jet Infosystems
>
> 17 авг. 2016 г., в 13:38, Ravishankar N <ravishankar at redhat.com>
> написал(а):
>
> On 08/17/2016 03:48 PM, Дмитрий Глушенок wrote:
>
> Unfortunately not:
>
> Remount FS, then access test file from second client:
>
> [root at srv02 ~]# umount /mnt
> [root at srv02 ~]# mount -t glusterfs srv01:/test01 /mnt
> [root at srv02 ~]# ls -l /mnt/passwd
> -rw-r--r--. 1 root root 1505 авг 16 19:59 /mnt/passwd
> [root at srv02 ~]# ls -l /R1/test01/
> итого 4
> -rw-r--r--. 2 root root 1505 авг 16 19:59 passwd
> [root at srv02 ~]#
>
> Then remount FS and check if accessing the file from second node triggered
> self-heal on first node:
>
> [root at srv01 ~]# umount /mnt
> [root at srv01 ~]# mount -t glusterfs srv01:/test01 /mnt
> [root at srv01 ~]# ls -l /mnt
>
>
> Can you try `stat /mnt/passwd` from this node after remounting? You need
> to explicitly lookup the file. `ls -l /mnt` is only triggering readdir on
> the parent directory.
> If that doesn't work, is this mount connected to both clients? i.e. if you
> create a new file from here, is it getting replicated to both bricks?
>
> -Ravi
>
> итого 0
> [root at srv01 ~]# ls -l /R1/test01/
> итого 0
> [root at srv01 ~]#
>
> Nothing appeared.
>
> [root at srv01 ~]# gluster volume info test01
>
> Volume Name: test01
> Type: Replicate
> Volume ID: 2c227085-0b06-4804-805c-ea9c1bb11d8b
> Status: Started
> Number of Bricks: 1 x 2 = 2
> Transport-type: tcp
> Bricks:
> Brick1: srv01:/R1/test01
> Brick2: srv02:/R1/test01
> Options Reconfigured:
> features.scrub-freq: hourly
> features.scrub: Active
> features.bitrot: on
> transport.address-family: inet
> performance.readdir-ahead: on
> nfs.disable: on
> [root at srv01 ~]#
>
> [root at srv01 ~]# gluster volume get test01 all | grep heal
> cluster.background-self-heal-count 8
>
> cluster.metadata-self-heal on
>
> cluster.data-self-heal on
>
> cluster.entry-self-heal on
>
> cluster.self-heal-daemon on
>
> cluster.heal-timeout 600
>
> cluster.self-heal-window-size 1
>
> cluster.data-self-heal-algorithm (null)
>
> cluster.self-heal-readdir-size 1KB
>
> cluster.heal-wait-queue-length 128
>
> features.lock-heal off
>
> features.lock-heal off
>
> storage.health-check-interval 30
>
> features.ctr_lookupheal_link_timeout 300
>
> features.ctr_lookupheal_inode_timeout 300
>
> cluster.disperse-self-heal-daemon enable
>
> disperse.background-heals 8
>
> disperse.heal-wait-qlength 128
>
> cluster.heal-timeout 600
>
> cluster.granular-entry-heal no
>
> [root at srv01 ~]#
>
> --
> Dmitry Glushenok
> Jet Infosystems
>
> 17 авг. 2016 г., в 11:30, Ravishankar N <ravishankar at redhat.com>
> написал(а):
>
> On 08/17/2016 01:48 PM, Дмитрий Глушенок wrote:
>
> Hello Ravi,
>
> Thank you for reply. Found bug number (for those who will google the
> email) https://bugzilla.redhat.com/show_bug.cgi?id=1112158
>
> Accessing the removed file from mount-point is not always working because
> we have to find a special client which DHT will point to the brick with
> removed file. Otherwise the file will be accessed from good brick and
> self-healing will not happen (just verified). Or by accessing you meant
> something like touch?
>
>
> Sorry should have been more explicit. I meant triggering a lookup on that
> file with `stat filename`. I don't think you need a special client. DHT
> sends the lookup to AFR which in turn sends to all its children. When one
> of them returns ENOENT (because you removed it from the brick), AFR will
> automatically trigger heal. I'm guessing it is not always working in your
> case due to caching at various levels and the lookup not coming till AFR.
> If you do it from a fresh mount ,it should always work.
> -Ravi
>
> Dmitry Glushenok
> Jet Infosystems
>
> 17 авг. 2016 г., в 4:24, Ravishankar N <ravishankar at redhat.com>
> написал(а):
>
> On 08/16/2016 10:44 PM, Дмитрий Глушенок wrote:
>
> Hello,
>
> While testing healing after bitrot error it was found that self healing
> cannot heal files which were manually deleted from brick. Gluster 3.8.1:
>
> - Create volume, mount it locally and copy test file to it
> [root at srv01 ~]# gluster volume create test01 replica 2 srv01:/R1/test01
> srv02:/R1/test01
> volume create: test01: success: please start the volume to access data
> [root at srv01 ~]# gluster volume start test01
> volume start: test01: success
> [root at srv01 ~]# mount -t glusterfs srv01:/test01 /mnt
> [root at srv01 ~]# cp /etc/passwd /mnt
> [root at srv01 ~]# ls -l /mnt
> итого 2
> -rw-r--r--. 1 root root 1505 авг 16 19:59 passwd
>
> - Then remove test file from first brick like we have to do in case of
> bitrot error in the file
>
>
> You also need to remove all hard-links to the corrupted file from the
> brick, including the one in the .glusterfs folder.
> There is a bug in heal-full that prevents it from crawling all bricks of
> the replica. The right way to heal the corrupted files as of now is to
> access them from the mount-point like you did after removing the
> hard-links. The list of files that are corrupted can be obtained with the
> scrub status command.
>
> Hope this helps,
> Ravi
>
> [root at srv01 ~]# rm /R1/test01/passwd
> [root at srv01 ~]# ls -l /mnt
> итого 0
> [root at srv01 ~]#
>
> - Issue full self heal
> [root at srv01 ~]# gluster volume heal test01 full
> Launching heal operation to perform full self heal on volume test01 has
> been successful
> Use heal info commands to check status
> [root at srv01 ~]# tail -2 /var/log/glusterfs/glustershd.log
> [2016-08-16 16:59:56.483767] I [MSGID: 108026]
> [afr-self-heald.c:611:afr_shd_full_healer] 0-test01-replicate-0: starting
> full sweep on subvol test01-client-0
> [2016-08-16 16:59:56.486560] I [MSGID: 108026]
> [afr-self-heald.c:621:afr_shd_full_healer] 0-test01-replicate-0: finished
> full sweep on subvol test01-client-0
>
> - Now we still see no files in mount point (it becomes empty right after
> removing file from the brick)
> [root at srv01 ~]# ls -l /mnt
> итого 0
> [root at srv01 ~]#
>
> - Then try to access file by using full name (lookup-optimize and
> readdir-optimize are turned off by default). Now glusterfs shows the file!
> [root at srv01 ~]# ls -l /mnt/passwd
> -rw-r--r--. 1 root root 1505 авг 16 19:59 /mnt/passwd
>
> - And it reappeared in the brick
> [root at srv01 ~]# ls -l /R1/test01/
> итого 4
> -rw-r--r--. 2 root root 1505 авг 16 19:59 passwd
> [root at srv01 ~]#
>
> Is it a bug or we can tell self heal to scan all files on all bricks in
> the volume?
>
> --
> Dmitry Glushenok
> Jet Infosystems
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-users
>
>
>
>
>
>
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://www.gluster.org/mailman/listinfo/gluster-users
>
--
Pranith
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.gluster.org/pipermail/gluster-users/attachments/20160820/d7377f9c/attachment.html>
More information about the Gluster-users
mailing list