Fedora-directory-devel Digest, Vol 19, Issue 8Re: [Fedora-directory-devel] Please review: distributed numeric assignment plugin

[Howard Chu]

> Date: Mon, 22 Jan 2007 13:02:40 -0700
 > From: Richard Megginson <rmeggins at redhat.com>

>> > Rob Crittenden wrote:
>>> >> Pete Rowley wrote:
>>>> >>> New pre-operation plugin:
>>> >>
>>> >> In dna_get_next_value() what could cause a call to 
>>> >> slapi_search_internal_get_entry() to fail that you have to try 3 times?
>>> >>
>> > It's actually the mod operation that can fail, the original value is 
>> > deleted and the new value added in one operation with two mods, if the 
>> > original value has changed since the search the mod operation will 
>> > fail. It's a way to get an atomic increment.
> Ok.  So I think the errors should be treated a little differently.  If 
> the search fails, that's bad - probably a fatal error, or perhaps 
> someone deleted the configuration entry out from under you.  I think 
> that if the mod fails, you should check the error code, for something 
> like LDAP_TYPE_OR_VALUE_EXISTS, which means the mod->add failed because 
> attribute already has that value, or whatever specific error is returned 
> from the mod->delete value operation when the value doesn't exist.  
> Other errors are probably fatal.

You should only be able to get LDAP_NO_SUCH_ATTRIBUTE, not 
LDAP_TYPE_OR_VALUE_EXISTS from this operation. The delete of the old value 
must fail before it even tries to add the new value. This is the standard 
test-and-set operation in LDAP. (And of course LDAP_NO_SUCH_OBJECT if someone 
is really messing with you...)

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