Re: [Xen-devel] [xen-unstable test] 128240: regressions - FAIL

[George Dunlap <george.dunlap@xxxxxxxxxx>]

On 10/01/2018 04:40 PM, Andrew Cooper wrote:
> On 01/10/18 16:35, Wei Liu wrote:
>> On Mon, Oct 01, 2018 at 04:19:07PM +0100, George Dunlap wrote:
>>> On 10/01/2018 04:17 PM, Wei Liu wrote:
>>>> On Mon, Oct 01, 2018 at 09:10:25AM -0600, Jan Beulich wrote:
>>>>>>>> On 01.10.18 at 16:33, <wei.liu2@xxxxxxxxxx> wrote:
>>>>>> On Mon, Oct 01, 2018 at 03:04:02AM -0600, Jan Beulich wrote:
>>>>>>>>>> On 30.09.18 at 23:59, <osstest-admin@xxxxxxxxxxxxxx> wrote:
>>>>>>>> flight 128240 xen-unstable real [real]
>>>>>>>> http://logs.test-lab.xenproject.org/osstest/logs/128240/ 
>>>>>>>>
>>>>>>>> Regressions :-(
>>>>>>>>
>>>>>>>> Tests which did not succeed and are blocking,
>>>>>>>> including tests which could not be run:
>>>>>>>>  test-amd64-amd64-migrupgrade 22 guest-migrate/src_host/dst_host fail 
>>>>>>>> REGR. vs. 
>>>>>> 128084
>>>>>>> At the first glance
>>>>>>>
>>>>>>> libxl: error: libxl_sched.c:232:sched_credit_domain_set: Domain 
>>>>>>> 1:Getting 
>>>>>> domain sched credit: Invalid argument
>>>>>>> libxl: error: libxl_create.c:1275:domcreate_rebuild_done: Domain 
>>>>>>> 1:cannot 
>>>>>> (re-)build domain: -3
>>>>>>> might indicate a problem resulting from the switch to credit2 as the 
>>>>>>> default
>>>>>>> scheduler. But "first glance" here really means what it says - I didn't 
>>>>>>> look
>>>>>>> (yet) at what exactly libxl tries to do there, in the hope that others 
>>>>>>> may
>>>>>>> know without much digging.
>>>>>> I think this is due to toolstack trying to set the same scheduler
>>>>>> parameters for the newly created guest.
>>>>>>
>>>>>> But in this test, the destination host is using a different scheduler
>>>>>> from the source host. Asking for credit scheduler on a credit2 host is
>>>>>> wrong.
>>>>>>
>>>>>> The relevant snippet in guest cfg (JSON) is:
>>>>>>
>>>>>>                 "sched_params": {
>>>>>>                     "sched": "credit",
>>>>>>                     "weight": 256,
>>>>>>                     "cap": 0
>>>>>>                 },
>>>>>>
>>>>>> I can't think of a method to fix it off the top of my head though.
>>>>> So is this something that was specified in the original config? Or
>>>>> is it just the current value which gets read and an attempt made
>>>>> to re-install. If there was no explicit setting in the guest config,
>>>>> shouldn't such a "default" setting be retained by not transferring
>>>>> any scheduler specifics during migration?
>>>>>
>>>> No setting in guest cfg. Those values are extracted from the hypervisor.
>>>> I think we may be able to not send default values to the remote end.
>>> Wait, the migration code reads the scheduler parameters -- even if these
>>> have not been explicitly set by the admin -- and sends them along with
>>> the migration stream?  And if the remote scheduler is different, the
>>> migration fails?
>>>
>>> That's not so good. :-)
>> But one can argue that the guest is specific configured that way so it's
>> parameters should be preserved. We normally analyse things on a case by
>> case basis.
> 
> If there isn't an obvious fix, then the switch of default scheduler
> needs reverting until there is a fix present.  This is currently
> blocking master.

Agreed.  I'd argue for ignoring failures to set scheduler parameters on
migrate, on the grounds that this will be less risk to the project as a
whole than reverting credit2 again.  But either way we should do
something quickly.

 -George

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel