<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<meta content="text/html;charset=ISO-8859-1"
http-equiv="Content-Type">
</head>
<body bgcolor="#ffffff" text="#000000">
Lars Ellenberg wrote:
<blockquote cite="mid:20070926101001.GC9808@barkeeper1.linbit"
type="cite">
<pre wrap="">On Wed, Sep 26, 2007 at 05:56:42PM +0800, Petr Cervenka wrote:
</pre>
<blockquote type="cite">
<pre wrap="">Lars Ellenberg wrote:
</pre>
<blockquote type="cite">
<pre wrap="">well.
I guess you used internal meta data.
if the RAID got it right, then all the old data is still there,
lying at "the end" of the previous device, so somewhere at 220G.
drbd only looks for its meta data at the end of the device,
so at offset ~400G currently.
there is nothing but nonsense there.
you have to create new drbd meta-data then,
as if you'd set it up for the first time.
</pre>
</blockquote>
<pre wrap="">Aha, well this is just test before we go live, so no damage there, so is
it better to store metadata in another disk , can this be just file? How
big does it have to be? As we expect to expend the disk more often. How
come with LVM this works? Does LVM extend differently ?
</pre>
</blockquote>
<pre wrap=""><!---->
resizing with internal meta data is fragile,
and works only when you can do the resize while drbd stays online.
while it is running, it sort of "knows" where its meta data lies,
and can the relocate it when told so.
if it crashes/is unconfigured/unloaded when the device is already
resized, but before the meta data got relocated, sorry, out of luck.
resizing works much better with external meta disk, since the data there
does not have to be relocated.
</pre>
</blockquote>
yeah i turned off drbd, remove scsci device (hotplug) add it again
with the right size, then start drbd try :<br>
root@darla1:/proc/scsi # drbdsetup /dev/drbd0 resize <br>
Failure: (138) Device does not have a disk-config<br>
<br>
:( so i try put size option in drbd.conf..<br>
same result.. so resize does not work at all? do i have to create new
meteadata, sync it with other side, and do the same on node 1? <br>
<br>
<blockquote cite="mid:20070926101001.GC9808@barkeeper1.linbit"
type="cite">
<pre wrap="">it cannot be a file.
you could fake a file to be a disk with losetup, but don't do that.
it introduces a potential for deadlock, because block io
re-entering file-io path.
</pre>
</blockquote>
ok we made another small partition and moved all metadata there, btw
dump and restored from file did not work .. gave us error :<br>
v08 Magic number not found<br>
SIGSEGV!<br>
This feels like a bug.<br>
debug hint: last memset: md_disk_08_to_cpu:560: @0x61d048 104<br>
Sorry.<br>
<br>
validateing the dump file was fine.. <br>
<br>
<blockquote cite="mid:20070926101001.GC9808@barkeeper1.linbit"
type="cite">
<pre wrap="">it has to be a "partition", "logical volume" or something like that.
man drbd.conf should tell you the rest.
</pre>
</blockquote>
<br>
<br>
thanks
</body>
</html>