[Openstack-operators] Cinder Juno to Kilo upgrade and DB encodings

Arne Wiebalck Arne.Wiebalck at cern.ch
Fri Sep 18 07:07:47 UTC 2015


On 17 Sep 2015, at 18:11, Mathieu Gagné <mgagne at internap.com<mailto:mgagne at internap.com>> wrote:

On 2015-09-17 4:06 AM, Arne Wiebalck wrote:
Hi,

During our Cinder upgrade on CentOS7 from Juno to Kilo, we ran into this bug:
https://bugs.launchpad.net/cinder/+bug/1455726

As there is no fix available from what I see, what we came up with as a “solution”
is to explicitly set the character and the collation in all existing tables in the database
before the upgrade:

—>
alter database cinder CHARACTER SET utf8 COLLATE utf8_unicode_ci;
SET foreign_key_checks = 0;
ALTER TABLE `backups` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `cgsnapshots` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `consistencygroups` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `encryption` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `iscsi_targets` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `migrate_version` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `quality_of_service_specs` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `quota_classes` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `quota_usages` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `quotas` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `reservations` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `services` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `snapshot_metadata` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `snapshots` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `transfers` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `volume_admin_metadata` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `volume_glance_metadata` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `volume_metadata` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `volume_type_extra_specs` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `volume_types` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
ALTER TABLE `volumes` CONVERT TO CHARACTER SET utf8 COLLATE utf8_unicode_ci;
SET foreign_key_checks = 1;
<—

Note that in our case the databases default character set was already utf8 everywhere, while the collation
was utf8_general_ci. With that conversion the upgrade seems to work fine in our tests.

Before we retry the upgrade: do if people here think that this a reasonable approach or will this cause
other issues? Are there alternative approaches?


In our case, we ran this command before upgrading:

ALTER DATABASE cinder CHARACTER SET utf8 COLLATE utf8_general_ci;

See this thread about the same problem:
http://lists.openstack.org/pipermail/openstack/2015-August/013599.html

And the proposed solution:
http://lists.openstack.org/pipermail/openstack/2015-August/013601.html

I suspect that people using Puppet to deploy OpenStack are encountering
this issue due to this change:
https://review.openstack.org/#/c/175991/

We thought this change wouldn't affect anyone but it looks it does.


Thanks a lot for the pointers, Mathieu. I’ll give that a try.

Cheers,
 Arne

—
Arne Wiebalck
CERN IT






-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-operators/attachments/20150918/46d8ee4a/attachment-0001.html>


More information about the OpenStack-operators mailing list