Re: Xen PVM: Strange lockups when running PostgreSQL load

[Stefan Bader <stefan.bader@canonical.com>]

[-- Attachment #1.1: Type: text/plain, Size: 1368 bytes --]

On 19.10.2012 17:08, Jan Beulich wrote:
>>>> On 19.10.12 at 16:57, Stefan Bader <stefan.bader@canonical.com> wrote:
>> On 19.10.2012 16:49, Jan Beulich wrote:
>>
>>>> No, I assumed that you saying native kernel did prior to ticket lock 
>>>> conversion,
>>>
>>> I indeed did say that. Just go look back at the old code.
>>
>> No, sorry, but no. Just that old code did it does not help me here. I di 
>> believe
>> you there. Still it seems there is something wrong doing it there in a pv 
>> guest.
> 
> PV guest may be a little broad - did you observe such a problem
> with a recent kernel of ours?

3.2 (ok, older) and 3.5 up to now, certainly can and should retry with at least
3.6. And on the host side it was on Amazon EC2 and locally with a 4.1.2 host.

> 
>> Though I cannot give a satisfying explanation, yet.
> 
> But that's going to be necessary in order to judge whether the
> brute force approach you take is really the only alternative (and
> whether it's fixing the problem rather than just making it less
> likely).

I know. Unfortunately that means digging quite deeply in the not-so-well-known.
And it does not help either that the backtraces in the dump seem not completely
trustworthy. Like vcpus that the host says are polling on a certain eventchannel
do not seem to be sitting on the hypercall.
> 
> Jan
> 



[-- Attachment #1.2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 897 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

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