[Users] [vdsm] Migration failed (previous migrations succeded)

[Gianluca Cecchi]

On Mon, Feb 3, 2014 at 1:43 PM, Michal Skrivanek
<michal.skrivanek at redhat.com> wrote:
>
> On Feb 1, 2014, at 18:25 , Itamar Heim <iheim at redhat.com> wrote:
>
>> On 01/31/2014 09:30 AM, Sven Kieske wrote:
>>> Hi,
>>>
>>> is there any documentation regarding all
>>> allowed settings in the vdsm.conf?
>>>
>>> I didn't find anything related in the rhev docs
>>
>> that's a question for vdsm mailing list - cc-ing…
>
> the vdsm.conf has a description for each parameter…just search for all containing "migration":)
> we do want to expose some/most/all of them in UI/REST eventually, the "migration downtime" setting is there now in 3.4, but others are missing
>
> Thanks,
> michal
>>
>>>
>>> Am 30.01.2014 21:43, schrieb Itamar Heim:
>>>> On 01/30/2014 10:37 PM, Markus Stockhausen wrote:
>>>>>>> Von: Itamar Heim [iheim at redhat.com]
>>>>>>> Gesendet: Donnerstag, 30. Januar 2014 21:25
>>>>>>> An: Markus Stockhausen; ovirt-users
>>>>>>> Betreff: Re: [Users] Migration failed (previous migrations succeded)
>>>>>>>
>>>>>>>
>>>>>>> Now I' getting serious problems. During the migration the VM was
>>>>>>> doing a rather slow download at 1,5 MB/s. So the memory changed
>>>>>>> by 15 MB per 10 seconds. No wonder that a check every 10 seconds
>>>>>>> was not able to see any progress. Im scared what will happen if I
>>>>>>> want to migrate a medium loaded system runing a database.
>>>>>>>
>>>>>>> Any tip for a parametrization?
>>>>>>>
>>>>>>> Markus
>>>>>>>
>>>>>>
>>>>>> what's the bandwidth? default is up to 30MB/sec, to allow up to 3 VMs to
>>>>>> migrate on 1Gb without congesting it.
>>>>>> you could raise that if you have 10GB, or raise the bandwidth cap and
>>>>>> reduce max number of concurrent VMs, etc.
>>>>>
>>>>> My migration network is IPoIB 10GBit. During our tests only one VM
>>>>> was migrated.  Bandwidth cap or number of concurrent VMs has not
>>>>> been changed after default install.
>>>>>
>>>>> Is migration_max_bandwidth in /etc/vdsm/vdsm.conf still the right place?
>>>>
>>>> probably
>>>>
>>>>> And what settings do you suggest?
>>>>


Not tried myself the change of values, but in a previous thread
(actually for limiting and not speeding up migration ;-) these two
parameters were described and to be put in each /etc/vdsm/vdsm.conf

max_outgoing_migrations
(eg 1 for allowing only one migration at a time)

migration_max_bandwidth
(unit is in MBytes/s, vdsm default is 32MiBps and it is for single
migration, not overall)

I think it is necessary to follow this workflow for every host
- put host into maintenance
- stop vdsmd service
- change values
- start vdsmd service
- activate host

Gianluca