[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: search for "dynamic" calculated attribute's value (bdb_equality_candidates: not indexed)



----- "Daniel" <deepee@gmx.net> wrote:

> Hi slapd-Gurus out there,
> 
> using the search filter "(test=TRUE)" results in: "candidates not 
> indexed". The cause for the below slapd trace output seems clear to
> me: 
> In fact there is no entry in my bdb backend existing with an attribute
> 
> description "test" present...
> 
> => bdb_filter_candidates
>         EQUALITY
> => bdb_equality_candidates (test)
> <= bdb_equality_candidates: (test) not indexed
> <= bdb_filter_candidates: id=-1 first=1 last=430
> <= bdb_list_candidates: id=-1 first=1 last=430
> <= bdb_filter_candidates: id=-1 first=1 last=430
> <= bdb_list_candidates: id=25 first=141 last=430
> <= bdb_filter_candidates: id=25 first=141 last=430
> bdb_search_candidates: id=25 first=141 last=430
> => test_filter
>     EQUALITY
> => access_allowed: search access to "dc=foo,dc=bar" "test" requested
> <= root access granted
> => access_allowed: search access granted by manage(=mwrscxd)
> <= test_filter 5
> bdb_search: 141 does not match filter
> 
> 
> In regard to the above slapd-trace I've the following question:
> Could someone please be so kind and tell me whether there exists a 
> possibility to search (filter) for "dynamic" (calculated) attribute 
> values, that do not get stored in a backend (in my case bdb)? Idea: An
> 
> overlay should calculate a dynamic attribute's value based on already
> 
> existing attribute values of an entry each time this dynamic attribute
> 
> is searched for using a filter-statement, e.g. 
> '(&(dynamic-attr-name=xyz)(|((...)(...))...)...)'.
> 
> First of all I had a quick look into slapo-collect's sources. The 
> collective attributes seem to be added dynamically to an entry but the
> 
> collected attributes are not "filterable". I think the reason is that
> 
> slapo-collect operates on the response/slapreply and that's to late in
> 
> the overlay's processing queue...
> 
> Then I thought of something similar to the operational attribute 
> "hasSubordinates" - this attribute's value is filterable and seems (at
> 
> least sounds) to be dynamically evaluated. On the other hand I'm not 
> sure whether "hasSuborinates" is dynamically calculated each time an 
> entry get's searched or accessed. Perhabs the value of
> "hasSubordinates" 
> is only be calculated in case the entrys' childrens get modified and 
> (thus seldomly) updated within the backend, which could be the reason
> to 
> be search- and filterable?!
> 
> Last but not least I tried to implement a small test-overlay using a 
> custom search-callback that updates the dynamic "test" attribute
> within 
> the bdb-backend: Imagine an (very simple/stupid) overlay that for 
> example set a custom attribute named "test" to "TRUE" in case an
> entry's 
> attributes "givenName" and "sn" both contain at least one value. ;-) I
> 
> know, that the filter "(&(givenName=*)(sn=*))" would do the trick but
> 
> the example is just a simple scenario for me to step into 
> overlay-processing details.
> 
> If I use attr_merge() within the search-callback to store the dynamic
> 
> test attribute's value "TRUE" or "FALSE" within the entry the search 
> succeeds without problems - I mean the filter (test=TRUE) get's 
> processed ok.
> 
> So my questions still left is:
> Do I always have to store/update the dynamic attribute's calculation 
> result value within each entry (in the backend) or is there another 
> posibility that allows filtering for the calculated attribute's value,
> 
> while getting rid of the store and update overhead within the
> backend?
> 
> Hopefully this is no hen egg problem concerning the internal
> processing 
> order of filters, candidates, entries, etc.

Using dynamically computed attrs in a search filter and pretending it to ber indexed is an oxymoron.  If you intend to allow it, you need to be prepared to intercepting the search operation and doing the candidate collection yourself.  I don't recall any overlay or so that does it and that could be used as an example, simply because it makes little sense.  In any case, the problem can be solved, but the solution may not be trivial and require a fair amount of coding.

p.


Ing. Pierangelo Masarati
OpenLDAP Core Team

SysNet s.r.l.
via Dossi, 8 - 27100 Pavia - ITALIA
http://www.sys-net.it
-----------------------------------
Office:  +39 02 23998309
Mobile:  +39 333 4963172
Fax:     +39 0382 476497
Email:   ando@sys-net.it
-----------------------------------