On 21/04/15 14:55 +0000, Nikhil Komawar wrote:
Rally is great overall however, we need good EXPLAIN examples on real world data. Smaller deployments might benefit from a simple sample performance analysis however, larger data sets can have impacts on areas that you never expect.A spec means that we document the indices proposed in the code base, based on all of the use cases. The way I look at it, a patch is needed anyways and it (rally gate job) would get attention from reviewers when the patch is proposed.
Yes, I believe we need both. However, I'd probably just start with something smaller and see how it behaves before going with big data sets. I'm not saying we don't need tests with proper data sets, I'm saying that I'd probably start with smaller ones. As Mike already mentioned in his email, there's an impact in writes and we can see that from Rally tests, AFAICT. The spec can come later, IMHO. Cheers, Flavio
________________________________________ From: Flavio Percoco <fla...@redhat.com> Sent: Tuesday, April 21, 2015 10:48 AM To: OpenStack Development Mailing List (not for usage questions) Subject: Re: [openstack-dev] [glance] Why no DB index on sort parameters On 21/04/15 14:39 +0000, Nikhil Komawar wrote:This is a good idea. We recently removed a unique constraint that may result into some queries being very slow especially those that involve "name" property. I would recommend sketching out a spec that identifies potential full table scans especially for queries that join over image_properties table. We should discuss there what other use cases look like rather than smaller feedback on the ML.More thatn a spec, I'd be interested in seeing the patch with the change up and the results reported in Rally. I guess we'll need a spec anyway, although I'd probably be ok with a good bug report here. /me *shrugs* FlavioThanks, -Nikhil ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ From: Mike Bayer <mba...@redhat.com> Sent: Tuesday, April 21, 2015 9:45 AM To: openstack-dev@lists.openstack.org Subject: Re: [openstack-dev] [glance] Why no DB index on sort parameters On 4/21/15 2:47 AM, Ajaya Agrawal wrote: Hi All, I see that glance supports arbitrary sort parameters and the default is "created_at" while listing images. Is there any reason why we don't have index over these fields? If we have an index over these fields then we would avoid a full table scan to do sorting. IMO at least the created_at field should have an index on it. just keep in mind that more indexes will place a performance penalty on INSERT statements, particularly at larger volumes. I have no idea if that is important here but something to keep in mind. Cheers, Ajaya __________________________________________________________________________ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev__________________________________________________________________________ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev-- @flaper87 Flavio Percoco __________________________________________________________________________ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
-- @flaper87 Flavio Percoco
pgpL5YJyWI2GL.pgp
Description: PGP signature
__________________________________________________________________________ OpenStack Development Mailing List (not for usage questions) Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev