Re: How indexes used with search

[Howard Chu <hyc@symas.com>]

Emmanuel LÃcharny wrote:
> Le 9/26/12 2:01 PM, Howard Chu a Ãcrit :
> > Roman Rybalko wrote:
> > > 25.09.2012 14:57, Roman Rybalko ÐÐÑÐÑ:
> > > > Also it's not clear for me how objectID set is determined when there
> > > > is search on "one" scope? The same when "sub" scope cover not the full
> > > > database contents but a part of the object tree (suffix: cn=test,
> > > > search base: cn=test1,cn=test)? I thought that search with base
> > > > covering not the full backend suffix (but a partial tree) should work
> > > > faster, but seems this is not true, even overhead introduced by
> > > > additional objectID set. Right?
> > > Howard, can you explain it please?
> >
> > The scope is just another set, yes. It must be ANDed with the index
> > sets. Searches with subtree scope over the entire database use a
> > shortcut, instead of reading a scope set we just use a fake set
> > representing all IDs. Searches over any other scope cannot take any
> > shortcut.
>
> As discussed with Howard last month, OneLevel scope can define a set of
> candidate IDs too, as soon as you can extract the children of the base
> entry from the Rdn index. Not sure OpenLDAP support that atm. Howard ?

The actual answer depends on which backend you're using. There's little point 
in discussing it when you can just read the source code to get the details. 
back-bdb uses explicit scope indices. back-hdb and back-mdb use only an rdn index.

--
  -- Howard Chu
  CTO, Symas Corp.           http://www.symas.com
  Director, Highland Sun     http://highlandsun.com/hyc/
  Chief Architect, OpenLDAP  http://www.openldap.org/project/