Re: index doesn't work?

[Zhang Weiwu <zhangweiwu@realss.com>]

Quanah Gibson-Mount wrote:
> --On Wednesday, November 07, 2007 10:37 PM +0800 Zhang Weiwu
> <zhangweiwu@realss.com> wrote:
>
>> My question is: The 4 minute search time is not reasonable (Excel can do
>> much faster with same number of records), and not acceptable for our use
>> (because checking modification history is daily routine). How can I
>> improve it? What other information I didn't provider to have asked this
>> question properly? What else (besides admin manual which only described
>> indexing a few pages) should I read?
>
> Sounds like your index may be corrupt, or you haven't properly tuned
> the DB_CONFIG file for the accesslog DB.  Alternately, when you ran
> slapindex, you failed to specify which database to index.  See the -b
> and -n options to slapindex.
Thank you for the hint. As I explained my index is produced with -b
option. When I try to tweak settings in DB_CONFIG I found tutorials
about it <http://k12linux.mesd.k12.or.us/ldap/performance.html> which
suggest me this:

    The size of dbcachesize should be big enough to hold your largest
    index file (*.dbb).

Unfortunately I do not have any *.dbb file. Besides log files I only get
several other files that looks like index files (they ends in .bdb)

-rw-rw---- 1 openldap openldap 126M 2007-11-09 10:35 dn2id.bdb
-rw-rw---- 1 openldap openldap 1,7G 2007-11-09 10:35 id2entry.bdb
-rw-rw---- 1 openldap openldap 2,1M 2007-11-09 10:35 reqDN.bdb
-rw-rw---- 1 openldap openldap  13M 2007-06-03 13:28 reqStart.bdb
-rw-rw---- 1 openldap openldap 660K 2007-11-09 10:35 reqType.bdb

Since the biggest file (id2entry.bdb) is at 1,7G, does that suggest I
should have dbcachesize of that size? I cannot understand why the cache
can be so large, after all there are only 50,000 records in it. After
all the query result of 'ldapsearch reqStart=* reqstart' is only 6.8MB.
But anyway I added these configuration:

dbcachesize 1740000000
cachesize 6000

Which is very close to my biggest .bdb file size / number of records.
The query time is reduced from 4 minutes to 3 minutes, still far from
user expectation (less than 10 seconds). Tested query is like this:

(reqDN=uid=paul_smith,ou=contacts,ou=xxx,dc=xxx,dc=xxx) 

I am going to try a lot more experiments, but would also like to hear if
there are something looking significantly wrong in my system? Thank!