The schemaUpdateNow_fix_and_test bug

[Nadezhda Ivanova]

----- Original Message -----
> From: Andrew Bartlett <abartlet at samba.org>
> To: Nadezhda Ivanova <nadezhda.ivanova at postpath.com>
> Cc: samba-technical at lists.samba.org <samba-technical at lists.samba.org>
> Sent: 11 November 2008 23:34:30 o'clock GMT+0200 Europe;Athens
> Subject: Re: The schemaUpdateNow_fix_and_test bug

> > On Tue, 2008-11-11 at 11:11 +0200, Nadezhda Ivanova wrote:
> > Hi Andrew,
> > 
> > Thanks for taking the time to try this again. I have been working on
> > that and it seems that the hanging is muuuuch worse if we send the
> > request via LDAP instead of using ldbmodify directly. 
> 
> Then why was the test written with ldapmodify on the local ldb?  Why 
> not
> use ldapmodify over ldap:// so as to provide a reproducible test case?
> 
Because we had no way to establish this before trying with ldbmodify without ldap. And it still is informative, is it not, by giving us a base for comparison?

> > What I have tried so far is to add a flag so that the addition of
> > INTEGER and CASE_SENSITIVE are only added at bootstrapping, that is,
> > that is not done if the indexes are generated because a
> > schemaUpdateNow is invoked. I did this because you mentioned, and it
> > is written in several places in the code, that we only need this at
> > startup while we do not yet have a schema. It speeded up things, but
> > not too much, besides it does not work - if I restart Samba
> > afterwards, I get a message that the schema head is missing...
> > So what I plan to do now is to remove the fix, then make a profile
> > using google tools. I know valgrind is mostly used, but I have more
> > experience with google tools and I did not find how to make a cpu
> > profile with valgrind. This will hopefully point out where we get 
> the
> > delay. Then I will make a python script to execute whjat we are 
> doing
> > via ldap, so that you can try it this way. If you still have no
> > problem, we can start comparing system configurations. What do you
> > think?
> 
> That sounds reasonable.  Please use the ldb engine to handle this (in
> python), rather than native python-ldap, as this must be reproduced in
> 'make test', or else we will chase system configurations forever. 

Not sure what you mean by ldb engine...

> 
> Andrew Bartlett
> 
> -- 
> Andrew Bartlett
> http://samba.org/~abartlet/
> Authentication Developer, Samba Team           http://samba.org
> Samba Developer, Red Hat Inc.

Nadya Ivanova