Xen pv_ops dom0 2.6.32.13 issues

Xen pv_ops dom0 2.6.32.13 issues

[greno]

[-- Attachment #1: Type: text/html, Size: 2022 bytes --]

[-- Attachment #2: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Re: Xen pv_ops dom0 2.6.32.13 issues

[Jeremy Fitzhardinge]

On 06/09/2010 02:02 PM, greno@verizon.net wrote:
>
> Ok, I've been running this 2.6.32.13 pv_ops dom0 kernel for several
> weeks and it has twice killed my domU's.  I get numerous CPU soft
> lockup bug errors and at times it will freeze which means a power
> cycle boot.

The lockups are in dom0 or domU?  Do the backtraces indicate a common
subsystem, or are they all over the place?

> This has resulted in things like:
> EXT-fs error (device dm-0): ext4_lookup: deleted inode reference
> EXT-fs error (device dm-0): ext4_lookup: deleted inode reference
> in the domU boots which has killed two of them.

What's your storage path from guest device to media?  Are they using
barriers?

    J

Re: Re: Xen pv_ops dom0 2.6.32.13 issues

[greno]

[-- Attachment #1: Type: text/html, Size: 1008 bytes --]

[-- Attachment #2: blkbackd.log --]
[-- Type: application/octet-stream, Size: 67 bytes --]

xenstore_scan: /local/domain/0/backend/blkbackd
quit on signal: 15

[-- Attachment #3: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Re: Xen pv_ops dom0 2.6.32.13 issues

[Jeremy Fitzhardinge]

On 06/09/2010 04:27 PM, greno@verizon.net wrote:
> blkbackd

Using phy: in your config file?  That really isn't recommended because
it has poor integrity; the writes are buffered in dom0 so writes can be
reordered or lost on crash, and the guest filesystem can't maintain any
of its own integrity guarantees.

tap:aio: is more resilient, since the writes go directly to the device
without buffering.

That doesn't directly relate to your lockup issues, but it should
prevent filesystem corruption when they happen.

    J



>
>
>
> Jun 9, 2010 07:13:23 PM, jeremy@goop.org wrote:
>
>     On 06/09/2010 04:05 PM, greno@verizon.net wrote:
>     > Jeremy,
>     > The soft lockups seemed to be occurring in different systems. And I
>     > could never make sense out of what was triggering them. I have not
>     > mounted any file systems with "nobarriers" in guests. The guests are
>     > all a single /dev/xvda. The underlying physical hardware is LVM over
>     > RAID-1 arrays. I'm attaching dmesg, kern.log, and messages in case
>     > these might be useful.
>
>     Using what storage backend? blkback? blktap2?
>
>     J
>
>     _______________________________________________
>     Xen-devel mailing list
>     Xen-devel@lists.xensource.com
>     http://lists.xensource.com/xen-devel
>

Which disk backend to use in domU?

[Neobiker]

Hello Jeremy,


Jeremy Fitzhardinge wrote:
> Using phy: in your config file?  That really isn't recommended because it
> has poor integrity; the writes are buffered in dom0 so writes can be
> reordered or lost on crash, and the guest filesystem can't maintain any
> of its own integrity guarantees.
> 
> tap:aio: is more resilient, since the writes go directly to the device
> without buffering.

Do you mean that using tap:aio with a disk.image is prefered against using
phy: with LVM-device?

Best Regards
Jens Friedrich aka Neobiker (www.neobiker.de)
-- 
View this message in context: http://old.nabble.com/Xen-pv_ops-dom0-2.6.32.13-issues-tp28835895p28857720.html
Sent from the Xen - Dev mailing list archive at Nabble.com.

Re: Which disk backend to use in domU?

[Valtteri Kiviniemi]

Hi,

I am also using phy: with LVM-partitions, and I also would like to know 
if there is a better or more preferred way.

- Valtteri Kiviniemi

Neobiker kirjoitti:
> Hello Jeremy,
> 
> 
> Jeremy Fitzhardinge wrote:
>> Using phy: in your config file?  That really isn't recommended because it
>> has poor integrity; the writes are buffered in dom0 so writes can be
>> reordered or lost on crash, and the guest filesystem can't maintain any
>> of its own integrity guarantees.
>>
>> tap:aio: is more resilient, since the writes go directly to the device
>> without buffering.
> 
> Do you mean that using tap:aio with a disk.image is prefered against using
> phy: with LVM-device?
> 
> Best Regards
> Jens Friedrich aka Neobiker (www.neobiker.de)

Re: Which disk backend to use in domU?

[Valtteri Kiviniemi]

Hi,

Ah, misunderstanding sorry, you were talking about disk images :)

Valtteri Kiviniemi kirjoitti:
> Hi,
> 
> I am also using phy: with LVM-partitions, and I also would like to know 
> if there is a better or more preferred way.
> 
> - Valtteri Kiviniemi
> 
> Neobiker kirjoitti:
>> Hello Jeremy,
>>
>>
>> Jeremy Fitzhardinge wrote:
>>> Using phy: in your config file?  That really isn't recommended 
>>> because it
>>> has poor integrity; the writes are buffered in dom0 so writes can be
>>> reordered or lost on crash, and the guest filesystem can't maintain any
>>> of its own integrity guarantees.
>>>
>>> tap:aio: is more resilient, since the writes go directly to the device
>>> without buffering.
>>
>> Do you mean that using tap:aio with a disk.image is prefered against 
>> using
>> phy: with LVM-device?
>>
>> Best Regards
>> Jens Friedrich aka Neobiker (www.neobiker.de)
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel

Re: Which disk backend to use in domU?

[Neobiker]

Hi

Valtteri Kiviniemi-2 wrote:
> 
> Hi, Ah, misunderstanding sorry, you were talking about disk images :)
> 
I'm talking about this config:
disk        = [
                  'phy:/dev/vm/vm01,xvda1,w',
                  'phy:/dev/vm/vm01-swap,xvda2,w',
                  'phy:/dev/daten/devel_debian_amd64,xvda3,w',
              ]
BR neobiker
-- 
View this message in context: http://old.nabble.com/Xen-pv_ops-dom0-2.6.32.13-issues-tp28835895p28858517.html
Sent from the Xen - Dev mailing list archive at Nabble.com.

Re: Which disk backend to use in domU?

[Jeremy Fitzhardinge]

On 06/11/2010 07:11 PM, Neobiker wrote:
> Hi
>
> Valtteri Kiviniemi-2 wrote:
>   
>> Hi, Ah, misunderstanding sorry, you were talking about disk images :)
>>
>>     
> I'm talking about this config:
> disk        = [
>                   'phy:/dev/vm/vm01,xvda1,w',
>                   'phy:/dev/vm/vm01-swap,xvda2,w',
>                   'phy:/dev/daten/devel_debian_amd64,xvda3,w',
>               ]
>   

file: is definitely unsafe; its IO gets buffered in the dom0 pagecache,
which means the guests writes aren't really writes.  I believe phy: has
similar problems, whereas tap:aio: implemented direct IO.  But someone
more storagey can confirm.

    J

Re: Which disk backend to use in domU?

[Daniel Stodden]

On Mon, 2010-06-14 at 06:49 -0400, Jeremy Fitzhardinge wrote:
> On 06/11/2010 07:11 PM, Neobiker wrote:
> > Hi
> >
> > Valtteri Kiviniemi-2 wrote:
> >   
> >> Hi, Ah, misunderstanding sorry, you were talking about disk images :)
> >>
> >>     
> > I'm talking about this config:
> > disk        = [
> >                   'phy:/dev/vm/vm01,xvda1,w',
> >                   'phy:/dev/vm/vm01-swap,xvda2,w',
> >                   'phy:/dev/daten/devel_debian_amd64,xvda3,w',
> >               ]
> >   
> 
> file: is definitely unsafe; its IO gets buffered in the dom0 pagecache,
> which means the guests writes aren't really writes.  I believe phy: has
> similar problems, whereas tap:aio: implemented direct IO.  But someone
> more storagey can confirm.

Unless there's a difference in type names between XCP and .org, 'phy'
means a bare LUN plugged into blkback?

Those run underneath the entire block cache subsystems, which ironically
has caching issues of it's own.

But your writes are safe.

Daniel

Re: Which disk backend to use in domU?

[Pasi Kärkkäinen]

On Mon, Jun 14, 2010 at 11:49:45AM +0100, Jeremy Fitzhardinge wrote:
> On 06/11/2010 07:11 PM, Neobiker wrote:
> > Hi
> >
> > Valtteri Kiviniemi-2 wrote:
> >   
> >> Hi, Ah, misunderstanding sorry, you were talking about disk images :)
> >>
> >>     
> > I'm talking about this config:
> > disk        = [
> >                   'phy:/dev/vm/vm01,xvda1,w',
> >                   'phy:/dev/vm/vm01-swap,xvda2,w',
> >                   'phy:/dev/daten/devel_debian_amd64,xvda3,w',
> >               ]
> >   
> 
> file: is definitely unsafe; its IO gets buffered in the dom0 pagecache,
> which means the guests writes aren't really writes.  I believe phy: has
> similar problems, whereas tap:aio: implemented direct IO.  But someone
> more storagey can confirm.
> 

I though phy: submits direct bio's bypassing dom0 pagecache.. 

-- Pasi