[Gluster-users] Fwd: Problems with self heal on replicated DHT

[Raghavendra G]

Hi Roland,

Instead of deleting files from back end, can you simulate the failures by
just killing one of the glusterfs servers and then deleting the files from
mount point? Once you are simulating failures like this, you can try to
self-heal after restarting the killed server. Glusterfs does not support
modification of files directly from back end directories. All the operations
should be done from mount point.

regards,
On Wed, Jan 6, 2010 at 12:29 AM, Roland Rabben <roland at jotta.no> wrote:

> Hi
> I have now tested Glusterfs 3.0 on a different system with the same
> self-heal problems I experienced in 2.0.6.
>
> To summarize:
> I use cluster/replicate and cluster/distribute
> Brick 1 on server 1 gets replicated to brick 1 on server 2.
> Brick 2 on server 1 gets replicated to brick 2 on server 2. etc...
>
> If I delete a file called "movie.avi" from a brick behind Glusterfs back on
> server 2, self-heal works fine with the "ls -lR" command.
>
> If I delete a file called "movie.avi" from a brick behind Glusterfs back on
> server 1, self-heal with "ls -lR" doesn't work.
>
> If I delete a file called "movie.avi" from a brick behind Glusterfs back on
> server 1, self-heal with "ls -lR movie.avi" works.
>
> So for self-heal to work properly you need to specify the missing file. I
> wonder how many do that? I think most people run "ls -lR" thinking
> everything will be fine.
>
> Hope this is useful.
>
> Regards
> Roland
>
>
>
> 2009/12/15 Raghavendra G <raghavendra at gluster.com>
>
> ---------- Forwarded message ----------
>> From: Raghavendra G <raghavendra.hg at gmail.com>
>> Date: 2009/12/15
>> Subject: Re: [Gluster-users] Problems with self heal on replicated DHT
>> To: Roland Rabben <roland at jotta.no>
>> Cc: gluster-users at gluster.org
>>
>>
>> Hi Roland,
>>
>> Can you try 3.0.0 and check whether the problem with self-heal persists?
>>
>> regards,
>> On Tue, Dec 15, 2009 at 3:14 PM, Roland Rabben <roland at jotta.no> wrote:
>>
>> > Hi
>> > I am using Gluster 2.06 on my servers and clients. My setup is a
>> replicated
>> > with cluster/replicate and then distribued with DHT.
>> >
>> >  My bricks are paired. Brick1 on server1 is replicated to Brick2 on
>> server
>> > 2
>> > and so on.
>> >
>> > My problem is that it seems that self heal only works on the secondary
>> > bricks (Brick2 in this example).
>> >
>> > For a test I can rm a file from Brick2 behind GlusterFS's back, run a
>> self
>> > heal on my dfs with ls -lR, and the file will be recovered on Brick2 as
>> > expected.
>> >
>> > However if I remove a file from Brick1 and run the same self heal, it
>> > doesn't work. Not only that, but the file is now gone from my dfs. (It
>> is
>> > still stored on Brick2)
>> >
>> > I have tried using option "lookup-unhashed on" and option
>> "lookup-unhashed
>> > yes". (There seem to be some confusion to what is the correct syntax in
>> the
>> > docs. The example<
>> >
>> http://gluster.com/community/documentation/index.php/Translators/cluster/distribute
>> > >uses
>> > "yes" and the text below says default is "on").
>> >
>> > Is this the expected behavior?
>> >
>> > Please advise on how to fix this. I want a self heal to fix both my
>> primary
>> > and secondary bricks.
>> >
>> > Regards
>> >
>> > Roland Rabben
>> >
>> > _______________________________________________
>> > Gluster-users mailing list
>> > Gluster-users at gluster.org
>> > http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
>> >
>> >
>>
>>
>> --
>> Raghavendra G
>>
>>
>>
>>
>> --
>> Raghavendra G
>>
>> _______________________________________________
>> Gluster-users mailing list
>> Gluster-users at gluster.org
>> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
>>
>>
>
>
> --
> Roland Rabben
> Founder & CEO Jotta AS
> Cell: +47 90 85 85 39
> Phone: +47 21 04 29 00
> Email: roland at jotta.no
>



-- 
Raghavendra G