[openstack-dev] [nova] if by "archived" you mean, "wipes out your tables completely", then sure, it works fine

Sean Dague sean at dague.net
Fri Mar 13 11:07:37 UTC 2015

On 03/13/2015 05:54 AM, Thomas Herve wrote:
> [snip]
>> If we assume that all of our tables are filled up with zeroes for those
>> deleted columns, because that’s the default, this **wipes the whole table
>> clean**.
>> How do the tests pass? Well the tests are in test_db_api->ArchiveTestCase,
>> and actually, they don’t. But they don’t fail every time, because the test
>> suite here runs with a database that is almost completely empty anyway, so
>> the broken archival routine doesn’t find too many rows to blow away except
>> for the rows in “instance_types”, which it only finds sometimes because the
>> tests are only running it with a small number of things to delete and the
>> order of the tables is non-deterministic.
>> I’ve posted the bug report at https://bugs.launchpad.net/nova/+bug/1431571
>> where I started out not knowing much about how this worked except that my
>> tests were failing, and slowly stumbled my way to come to this conclusion. A
>> patch is at https://review.openstack.org/#/c/164009/, where we look at the
>> actual Python-side default. However I’d recommend that we just hardcode the
>> zero here, since that’s how our soft-delete columns work.
> Hi Mike,
> Thanks for the investigation. I was wondering when that behavior was introduced and it seems that http://git.openstack.org/cgit/openstack/nova/commit/?id=ecf74d4c0a5a8a4290ecac048fc437dafe3d40fc is the likely culprit, which would mean that only Kilo is affected. Can you confirm?
> Thanks,

Yes, that looks like the problematic patch. I'd rather actually revert
that patch instead.

Also, real tests would be nice to actually prevent future regression.


Sean Dague

More information about the OpenStack-dev mailing list