Gluster users - Oct 2012 - lost data, help!

Hi,all:



       I have a problem now?Can you help me?

       When I use replace-brick command to use some new machines to replace some
old machines in my cluster.
   But it dosen't susscess! It only create some link files in the new
machines,it dosen't copy the real data to the

   new machines.So we have an idea?we copy the data in the bricks of  all the
old machines to the volume of

   my cluster from the mount point of the volume.(There is only one volume in my
cluster.) But we ingnor the

   link files in the bricks of the old methines.When we copy all the files in
the bricks of the old methines to the
   volume , the link files overwrite  all the real data files which are linked
by the link files. So we  lost much data.

  And when we read the files which are lost data, gluster will choose the
file's real file in dead loop,So client

  time out and there are too many socket links.

        How can we find our data? We try to use some disk tools to recovery the
lost data? We analyse all
   copyed link files to find out which data wo lose.

        Can someone give me some suggestions,thank you!


-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://supercolony.gluster.org/pipermail/gluster-users/attachments/20121010/8d1fc4bd/attachment.html>

I don't think glusterfs can help you out, and I guess that the
"cp" command
will do a trucate before copy data. While you were trying to copy files
from bricks to mountpoint, it was obvious that you were trying to copy a
file to itself, the cp command would truncate the file first, then
"cp"
found there was no data to copy, that was why data became to lose. Now this
problem turn out to be a problem of low layer filesystem, and I'm not sure
if a truncate syscall will reset the data of the block which will be
allocaed.. Files that placed at new bricks must have contained valid data,
so keep trying to use recover tools.

2012/10/10 renqiang <renqiang at 360buy.com>
>  Hi,all:
>
>
>
>        I have a problem now?Can you help me?
>
>        When I use replace-brick command to use some new machines to
> replace some old machines in my cluster.
>    But it dosen't susscess! It only create some link files in the new
> machines,it dosen't copy the real data to the
>
>    new machines.So we have an idea?we copy the data in the bricks of  all
> the old machines to the volume of
>
>    my cluster from the mount point of the volume.(There is only one
> volume in my cluster.) But we ingnor the
>
>    link files in the bricks of the old methines.When we copy all the
> files in the bricks of the old methines to the
>    volume , the link files overwrite  all the real data files which are
> linked by the link files. So we  lost much data.
>
>   And when we read the files which are lost data, gluster will choose the
> file's real file in dead loop,So client
>
>   time out and there are too many socket links.
>
>         How can we find our data? We try to use some disk tools to
> recovery the lost data? We analyse all
>    copyed link files to find out which data wo lose.
>
>         Can someone give me some suggestions,thank you!
>
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://supercolony.gluster.org/mailman/listinfo/gluster-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://supercolony.gluster.org/pipermail/gluster-users/attachments/20121011/caf92a09/attachment.html>