[linux-lvm] Pvmove while volume_ list is enabled freeze whole lvm

[linux-lvm] Pvmove while volume_ list is enabled freeze whole lvm

[Gianluca Cecchi]

[-- Attachment #1: Type: text/plain, Size: 473 bytes --]

Hello,
I have rhel 5.8 wit lvm tags enabled (volume_ list).
In volume_ list I have
@intraclusternode1
Tag, so the main vg is active with its lv.
But then I issue pvmove command that fails because it cannot activate the
temporary mirror...But now on, all commands related to lvm, also for other
vg, completely freeze..Any way to unblock?I tried removing the volume_list
directive in lvm.conf, but no way...I would like to avoid doing power off
if possible..
Thanks
Gianluca

[-- Attachment #2: Type: text/html, Size: 529 bytes --]

Re: [linux-lvm] Pvmove while volume_ list is enabled freeze whole lvm

[Zdenek Kabelac]

Dne 13.12.2012 09:23, Gianluca Cecchi napsal(a):
> Hello,
> I have rhel 5.8 wit lvm tags enabled (volume_ list).
> In volume_ list I have
> @intraclusternode1
> Tag, so the main vg is active with its lv.
> But then I issue pvmove command that fails because it cannot activate the
> temporary mirror...But now on, all commands related to lvm, also for other vg,
> completely freeze..Any way to unblock?I tried removing the volume_list
> directive in lvm.conf, but no way...I would like to avoid doing power off if
> possible..
> Thanks
> Gianluca

Usual way if something goes wrong:

Check if there is a device in 'suspend' state -
(i.e.   dmsetup info -c     and check for 's' flag in Stat column)
Resume all suspend devices -    dmsetup resume vgname-lvname &
You could easily remove those which have zero open count  - dmsetup remove
Eventually if you are brave enough, you could try 'dmsetup remove_all'.

Now you should be able to activate your LVs again.

Obviously 5.8 is somewhat older and there are some old bugs likely already 
fixed in newer releases - thus you may have far less problems with newer RHEL?
(or use support for reproducible bugs?)

Zdenek

Re: [linux-lvm] Pvmove while volume_ list is enabled freeze whole lvm

[Gianluca Cecchi]

On Thu, Dec 13, 2012 at 10:43 AM, Zdenek Kabelac  wrote:
>
>
> Usual way if something goes wrong:
>
> Check if there is a device in 'suspend' state -
> (i.e.   dmsetup info -c     and check for 's' flag in Stat column)
> Resume all suspend devices -    dmsetup resume vgname-lvname &
> You could easily remove those which have zero open count  - dmsetup remove
> Eventually if you are brave enough, you could try 'dmsetup remove_all'.
>
> Now you should be able to activate your LVs again.
>
> Obviously 5.8 is somewhat older and there are some old bugs likely already
> fixed in newer releases - thus you may have far less problems with newer
> RHEL?
> (or use support for reproducible bugs?)
>
> Zdenek
>

when I run pvmove I got:
# time pvmove /dev/mapper/mpath127 /dev/mpath/mpath23
  Not activating VG_DOCS_APPL/pvmove0 since it does not pass activation filter.
  Temporary pvmove mirror activation failed.
  Unable to deactivate open VG_DOCS_APPL-pvmove0 (253:69)
  Unable to deactivate open VG_DOCS_APPL-LV_DOCS_APPL (253:58)
  libdevmapper exiting with 1 device(s) still suspended.

This was because I'm using HA-LVM and forgot to temporarily bypass the
volume_list directive
volume_list = [ "VolGroup00", "@iclnode1" ]

Now in fact as you asked I can see
# dmsetup info -c
Name                                          Maj Min Stat Open Targ
Event  UUID
...
VG_DOCS_APPL-pvmove0                          253  69 -Isw    1    0
   0 LVM-oZIpsGRusTsze13xns0jvNeQbFuzxtRkwGQ1G6JEFZ5iM9YQRQzmFrithE6328Y5

and the I/O on the relative fs results blocked.
Apparentely the other ones are going ok.
So would it be ok to proceed this way:

1) comment out volume_list in lvm.conf
2) touch /boot/initrd*
(due to bug in ha-lv cluster monitor script)
3) dmsetup resume VG_DOCS_APPL-pvmove0

?

Thanks

Re: [linux-lvm] Pvmove while volume_ list is enabled freeze whole lvm

[Gianluca Cecchi]

On Thu, Dec 13, 2012 at 12:37 PM, Gianluca Cecchi  wrote:
> On Thu, Dec 13, 2012 at 10:43 AM, Zdenek Kabelac  wrote:
>>
>>
>> Usual way if something goes wrong:
>>
>> Check if there is a device in 'suspend' state -
>> (i.e.   dmsetup info -c     and check for 's' flag in Stat column)
>> Resume all suspend devices -    dmsetup resume vgname-lvname &
>> You could easily remove those which have zero open count  - dmsetup remove
>> Eventually if you are brave enough, you could try 'dmsetup remove_all'.
>>
>> Now you should be able to activate your LVs again.
>>
>> Obviously 5.8 is somewhat older and there are some old bugs likely already
>> fixed in newer releases - thus you may have far less problems with newer
>> RHEL?
>> (or use support for reproducible bugs?)
>>
>> Zdenek
>>
>
> when I run pvmove I got:
> # time pvmove /dev/mapper/mpath127 /dev/mpath/mpath23
>   Not activating VG_DOCS_APPL/pvmove0 since it does not pass activation filter.
>   Temporary pvmove mirror activation failed.
>   Unable to deactivate open VG_DOCS_APPL-pvmove0 (253:69)
>   Unable to deactivate open VG_DOCS_APPL-LV_DOCS_APPL (253:58)
>   libdevmapper exiting with 1 device(s) still suspended.
>
> This was because I'm using HA-LVM and forgot to temporarily bypass the
> volume_list directive
> volume_list = [ "VolGroup00", "@iclnode1" ]
>
> Now in fact as you asked I can see
> # dmsetup info -c
> Name                                          Maj Min Stat Open Targ
> Event  UUID
> ...
> VG_DOCS_APPL-pvmove0                          253  69 -Isw    1    0
>    0 LVM-oZIpsGRusTsze13xns0jvNeQbFuzxtRkwGQ1G6JEFZ5iM9YQRQzmFrithE6328Y5
>
> and the I/O on the relative fs results blocked.
> Apparentely the other ones are going ok.
> So would it be ok to proceed this way:
>
> 1) comment out volume_list in lvm.conf
> 2) touch /boot/initrd*
> (due to bug in ha-lv cluster monitor script)
> 3) dmsetup resume VG_DOCS_APPL-pvmove0
>
> ?
>
> Thanks

Zdenek,
YOU ARE THE MAN!!!!!
THANKS!

you saved me to restart three Oracle services runnign on this node

It worked with the three steps above
When you come near Milan I owe you a fresh beer ... let me know ;-)

Have a nice day