Gluster users - May 2012 - [3.3 beta3] When should the self-heal daemon be triggered?

Hi,
I eventually installed three Debian unstable machines, so I could 
install the GlusterFS 3.3 beta3.

I have a question about the self-heal daemon.

I'm trying to get a volume which is replicated, with two bricks.

I started up the volume, wrote some data, then killed one machine, and 
then wrote more data to a few folders from the client machine.
Then I restarted the second brick server.

At this point, the second server seemed to "self-heal" enough that it 
registered the new directories, but all the files inside were zero-length.

I then ran the command:
gluster volume heal testvol

After I ran that, there was some activity, and now all the files were 
populated.


Was that supposed to happen automatically, eventually, or am I missing 
something about how the self-heal daemon works?


Thanks,
Toby

PS. The use-case I have in mind is this scenario.

Say we have two Gluster storage bricks, replicated.
We want to do something that involves taking down the machines for an 
hour or two each, one at a time.

So, we turn one machine off.. All traffic now goes to the single other 
machine.

After a couple of hours, we turn the upgraded machine back on.

We now want to take the second machine down for a few hours.. but of 
course, its full of data written during the last two hours.

Will the self-heal mechanism automatically replicate that back to the 
second machine, now it's online?
Does that sync start happening immediately, or after a certain time 
period, or only after we manually run the 'volume heal' command?

ta,
Toby


On 03/05/12 19:45, Toby Corkindale wrote:
> Hi,
> I eventually installed three Debian unstable machines, so I could
> install the GlusterFS 3.3 beta3.
>
> I have a question about the self-heal daemon.
>
> I'm trying to get a volume which is replicated, with two bricks.
>
> I started up the volume, wrote some data, then killed one machine, and
> then wrote more data to a few folders from the client machine.
> Then I restarted the second brick server.
>
> At this point, the second server seemed to "self-heal" enough
that it
> registered the new directories, but all the files inside were zero-length.
>
> I then ran the command:
> gluster volume heal testvol
>
> After I ran that, there was some activity, and now all the files were
> populated.
>
>
> Was that supposed to happen automatically, eventually, or am I missing
> something about how the self-heal daemon works?
>
>
> Thanks,
> Toby
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users

-- 
.signature

On 05/03/2012 03:15 PM, Toby Corkindale wrote:
> Hi,
> I eventually installed three Debian unstable machines, so I could 
> install the GlusterFS 3.3 beta3.
>
> I have a question about the self-heal daemon.
>
> I'm trying to get a volume which is replicated, with two bricks.
>
> I started up the volume, wrote some data, then killed one machine, and 
> then wrote more data to a few folders from the client machine.
> Then I restarted the second brick server.
>
> At this point, the second server seemed to "self-heal" enough
that it
> registered the new directories, but all the files inside were 
> zero-length.
>
> I then ran the command:
> gluster volume heal testvol
>
> After I ran that, there was some activity, and now all the files were 
> populated.
>
>
> Was that supposed to happen automatically, eventually, or am I missing 
> something about how the self-heal daemon works?
>
The self-heal daemon triggers a crawl once every 600 seconds. If you 
wait out that interval, you should be able to see self-heals happening 
automatically. Else you can trigger it explicitly the way you did.

Regards,
Vijay