[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [linux-lvm] lvm fails trying to expand an LV into available space in VG.

Bryn M. Reeves wrote:
Hash: SHA1

David Timms wrote:
James Parsons wrote:
Hi - I'm the s-c-lvm guy; this will take a bit to digest. Just wanted
to let you know you were heard.
Yes, 3 minutes to scan for state seems a bit abnormal :)
Actually, its more like 5-6 minutes (I get bored waiting and start doing
other stuff). During boot up it takes about 8 seconds to activate the
lvm volumes.

Nothing to do with the s-c-lvm problem, but these times sound perfectly
normal if you have a metadata area on each of the 25 PVs in the volume

By default, pvcreate will set each PV up with a single metadata area
(MDA). This is fine for small volume groups but will kill tool
performance for volume groups with large numbers of PVs since the
runtime grows as something like N^3 in the number of metadata areas.
Hmmn, sounds like a good addition for the howto / FAQ !

I wonder if system-config-lvm has any tests for subprocesses taking to long, that might cause it to time before the process finishes ?

If the many of the existing pv assigned to the vg are full {0 bytes free}, could it cause the resizing of the lvm to fail due to no space to put more metadata ?

Note that it's only tool performance that suffers - I/O performance is
not affected.

You can prevent this by creating most of the PVs in such a VG with the
option "--metadatacopies=0" on the pvcreate command line. Create a small
number with either "--metadatacopies=1" (or just use the default - it's
1 anyway) and you should find the time to scan/activate the VG is much
Does value 1 means the original plus 1 copy, or only 1 copy total ?

Now recreate all the PVs except those that you want to contain metadata
with their original UUIDs and the --restorefile and --metadatacopies=0

# pvcreate --uuid=$UUID --restorefile=/etc/lvm/backup/<VolGroup>
- --metadatacopies=0 /dev/$DEVICE
Does this keep or trash the filesystem {ext3} within the lv that takes of most of this vg ?

Can I actually recreate the PVs without losing data ?

Next create the metadata-holding PVs:

The last time I had to do this was on a VG with 100 PVs,
Had you created 100PV's during development / testing, or was that a live system ? How long did the scan take ?

I used a script
that parsed the output of pvs to automate all this stuff. Drop me a line
off-list if you're interested & I'll see if I can pull it up.
Will do, hopefully it means finger errors wont trash the data.

Thanks, DavidT

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]