I''m curious about if there are any potential problems with using LVM
metadevices as ZFS zpool targets. I have a couple of situations where using a
device directly by ZFS causes errors on the console about "Bus and lots of
"stalled" I/O. But as soon as I wrap that device inside an LVM
metadevice and then use it in the ZFS zpool things work perfectly fine and
smoothly (no stalls).
Situation 1 is when trying to use Intel X25-E or X25-M SSD disks in a Sun X4240
server with the LSI SAS controller - I never could get things to run without
errors no matter what. (Tried multiple LSI controllers and multiple SSD disks).
Jul 8 09:43:31 merope scsi: [ID 365881 kern.info] /pci at 0,0/pci10de,375 at
f/pci1000,3150 at 0 (mpt0):
Jul 8 09:43:31 merope Log info 31126000 received for target 15.
Jul 8 09:43:31 merope scsi_status=0, ioc_status=804b, scsi_state=c
Jul 8 09:43:31 merope scsi: [ID 365881 kern.info] /pci at 0,0/pci10de,375 at
f/pci1000,3150 at 0 (mpt0):
Jul 8 09:43:31 merope Log info 31126000 received for target 15.
Jul 8 09:43:31 merope scsi_status=0, ioc_status=804b, scsi_state=c
Jul 8 09:43:31 merope scsi: [ID 107833 kern.warning] WARNING: /pci at
0,0/pci10de,375 at f/pci1000,3150 at 0/sd at f,0 (sd32):
Jul 8 09:43:31 merope Error for Command: write Error Level:
Retryable
Jul 8 09:43:31 merope scsi: [ID 107833 kern.notice] Requested Block: 64256
Error Block: 64256
Jul 8 09:43:31 merope scsi: [ID 107833 kern.notice] Vendor: ATA
Serial Number: CVEM8493
00BM
Jul 8 09:43:31 merope scsi: [ID 107833 kern.notice] Sense Key: Unit
Attention
Jul 8 09:43:31 merope scsi: [ID 107833 kern.notice] ASC: 0x29 (power on,
reset, or bus reset occurred), ASCQ: 0x0, FRU
: 0x0
Situation 2 is when I installed an X25-E in a X4500 Thumper. Here I
didn''t see any errors on the console of the server, but performance
would at regular intervals "drop" to zero (felt the same as in the LSI
case above, just without the console errors).
(In situation 1 above things would work perfectly fine when I was using an
Adaptec controller instead).
Anyway, when I put the 4GB partition of that SSD disk that I was using for
testing inside a simple LVM metadevice all errors vanished and performance
increased many many times. And no "hickups".
But I wonder... Is there anything in a setup like this that might be dangerous -
something that might come back and bite me in the future? LVM(disksuite) is
really mature technology and something I''ve been using without problems
on many servers for many years so I think it can be trusted but anyway...?
(I use the partition of that SSD-in-a-LVM-metadevice as a SLOG device for the
ZFS zpools on those servers and performance is now really *really* good).
--
This message posted from opensolaris.org
