[SSSD] [PATCH] IFP: cached objects

Jakub Hrozek jhrozek at redhat.com
Tue May 26 15:40:21 UTC 2015


On Tue, May 26, 2015 at 05:26:06PM +0200, Pavel Březina wrote:
> On 05/26/2015 05:22 PM, Jakub Hrozek wrote:
> >On Thu, May 21, 2015 at 11:44:08AM +0200, Pavel Březina wrote:
> >>https://fedorahosted.org/sssd/ticket/2338
> >>
> >>Depends on IFP: users and groups
> >
> >I haven't tested the code yet at all, but let's start the review :-)
> >
> >> From b08f37a1bf9593c74cdec8d829f0d125b9c66c30 Mon Sep 17 00:00:00 2001
> >>From: =?UTF-8?q?Pavel=20B=C5=99ezina?= <pbrezina at redhat.com>
> >>Date: Mon, 18 May 2015 12:05:38 +0200
> >>Subject: [PATCH 1/4] IFP: Implement
> >>  org.freedesktop.sssd.infopipe.Cache[.Object]
> >
> >This patch need to be split into one that just adds the functionality
> >and one that adds the sysdb upgrade.
> >
> >But mostly, I'm not sure we need to index this attribute, do you think
> >the lookups for cached objects would be so frequent? Keep in mind that
> >indexing adds some penalty on the cache size but also building the
> >indexes is not totally for free, so we should carefully consider the
> >indexes..
> >
> >Additionally, we were talking with Sumit about sysdb upgrades for
> >indexing attributes with the context being:
> >     https://fedorahosted.org/sssd/ticket/2597
> >
> >What Sumit proposed to investigate if we could get away of not bumping
> >the version (or adding a .z version to the sysdb string) if all we
> >change is indexes..do you have time to test that? (newer indexes with
> >older client etc..)
> >
> >I would prefer a different name for the new attribute, cache is too generic,
> >can we prefix the name?
> 
> All of the above was agreed on the design page. I don't oppose to removing
> the index nor renaming the attribute, can we discuss it in the design
> thread? We can add index sometime later if we notice it is a bottleneck, no
> problem there.

Sure. Discussing if the cache upgrade is needed at all might be even
better in a separate thread. Sorry if it was discussed, I forgot that...

I'm not too opposed to adding the index. In LDB, it's another record in
the form of:
    indexed_attribute:DN

Plus every time a cached attribute is modified, added or deleted, this
index record must be changed accordingly as well.

Please note the indexing itself is not the problem, I would just like
to avoid two sysdb upgrades if we can and avoid the upgrade at all, even
better.



More information about the sssd-devel mailing list