Re: dmcache RAID1 bug?

[Leonardo Santos <heiligerstein@xxxxxxxxx>]

I tested updating my xbox to same production kernel (3.17.2) and the problem appears now.

The old kernel (ok) is 3.13.0-37.

About updating lvm, I think it should not be, because this time I did not use them and did creating only through the dmsetup command.

I will continue testing with other kernels.

Can anyone help me?

On Mon, Nov 17, 2014 at 4:14 PM, Leonardo Santos <heiligerstein@xxxxxxxxx> wrote:

On Mon, Nov 17, 2014 at 12:46 PM, Mike Snitzer <snitzer@xxxxxxxxxx> wrote:

On Mon, Nov 17 2014 at  9:24am -0500,
Leonardo Santos <heiligerstein@xxxxxxxxx> wrote:

> I'm working in tests for my msc thesis envolving dm_cache module. I'm
> having problems with the following scenario:
>
> Slowstorage: RAID1 (mdadm) with 4 HDDs;
> - /dev/md/raid1_4ssds

You mean /dev/md/raid1_4hdds ?

A symlink to raid1 created using 4 ssds.

mdadm --create --verbose /dev/md/raid0_4ssds --raid-devices=4 --force --level=1 /dev/sdd /dev/sdc /dev/sdb /dev/sda

 

> -- /dev/sda
> -- /dev/sdb
> -- /dev/sdc
> -- /dev/sdd
>
> Faststorage: RAID1 (mdadm) with 4 SSDs
> - /dev/md/raid1_4hdds

And /dev/md/raid1_4ssds here?

mdadm --create --verbose /dev/md/raid0_4hdds --raid-devices=4 --force --level=1  /dev/sdh /dev/sdg /dev/sdf /dev/sde

 

> -- /dev/sde
> -- /dev/sdf
> -- /dev/sdg
> -- /dev/sdh
>
>
> Cachestorage: dm_cache envolving both devices
> - /dev/mapper/dmcache
> -- /dev/md/raid1_4ssds
> -- /dev/md/raid1_4hdds
>

Commands:

vgcreate cache /dev/md/raid0_4ssds -y

lvcreate cache -n metadata -l 5%FREE

lvcreate cache -n block -l 100%FREE

dd if=/dev/zero of=/dev/cache/metadata bs=1M count=100

# 16773120 = blockdev --getsize /dev/md/raid0_4hdds

dmsetup create dmcache --table '0 16773120 cache /dev/cache/metadata /dev/cache/block /dev/md/raid0_4hdds 512 1 writeback default 0'

 

> Other informations (strange facts):
> In RAID0 it's working very well.
> In a virtual machine (vbox) it's working as well (in this case disks have
> 2GB only for big tests automatization, in production are 160GB SSDs and 2TB
> HDDs);
>
> Have you seen about this error? Syslog is bellow.

Never seen this before... have you taken care to zero the dm-cache
metadata's superblock (first 512B of the cache metadata device) each
time you've reconfigured the cache device?

Have you tried using the latest lvm2 to create the dm-cache device?

I've used LVM2 from distribution (debian) and a latest kernel version.

LVM2 latest today is 2.02.112 (I'll try with latest LVM2 version and post results).

# lvm version

  LVM version:     2.02.95(2) (2012-03-06)

  Library version: 1.02.74 (2012-03-06)

  Driver version:  4.27.0

# uname -a

Linux steinhager 3.17.2 #2 SMP Fri Nov 7 12:49:16 BRST 2014 x86_64 GNU/Linux

--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel