[Openstack-operators] Slow db migration in havana

Michael Still mikal at stillhq.com
Tue Jun 11 22:11:11 UTC 2013


Nikola -- I think it there are performance improvements we can do then
that's a good idea.

I'm a little worried about deployers who track trunk deploying this
change -- we probably have a window in the order of days before it
becomes an issue. I'm also unsure about changing a migration once its
landed.

I'm wondering if we should revert this migration and replace it with
one which does nothing? If we do that, how do we handle people who
might already have run this migration?

Perhaps tweaking this migration is ok as long as its functionally
identical -- early deployers just suck up some pain, and later
deployments run a bit faster. Is that possible?

Michael

On Tue, Jun 11, 2013 at 11:34 PM, Nikola Đipanov <ndipanov at redhat.com> wrote:
> On 11/06/13 12:29, Michael Still wrote:
>> Hi.
>>
>> I just wanted to make sure people are aware of a slow db migration
>> which landed in havana today. I discovered this while testing a CI
>> script I am working on which tests DB migrations.
>>
>> The migration is number 186, which landed with this review
>> https://review.openstack.org/#/c/29251/11 -- it rewrites the block
>> device mapping tables, and will therefore affect people making heavy
>> use of volumes.
>>
>
> Hey Michael,
>
> I believe there may be a few low-hanging fruit fixes that can be done to
> that migration to improve it speed-wise.
>
> Since it would be easier for you to test those - I'll ping you as soon
> as I have a patch put together so you can try it out, if that's OK with you.
>
> Cheers,
>
> Nikola
>
>> Here's a log of the test run:
>>
>> http://openstack.stillhq.com/ci/If30afdb59d4c4268b97d3d10270df2cc729a0c4c/11/sqlalchemy_migration_nova_user_001/log.html
>>
>> The interesting log entry is at:
>>
>> http://openstack.stillhq.com/ci/If30afdb59d4c4268b97d3d10270df2cc729a0c4c/11/sqlalchemy_migration_nova_user_001/log.html#524
>>
>> This migration takes just under 20 minutes with my test database of
>> 45k block device mappings and 35k shadow block device mappings, on an
>> otherwise unloaded cloud instance.
>>
>> Cheers,
>> Michael
>>
>> PS: remember when I asked for test datasets? If you'd like to donate a
>> copy of a nova database to add to this test suite, please mail me
>> privately so we can negotiate.
>>
>



More information about the OpenStack-operators mailing list