Re: (b68fa5 don't flush entries until after cleanup callbacks)

Howard Chu wrote:

Hallvard B Furuseth wrote:

...

openldap-commit2devel@OpenLDAP.org writes:

...

commit b68fa5ecd73037ebd436a2663003f544d482f71e Author: Howard Chuhyc@openldap.org Date: Tue Nov 1 13:17:06 2011 -0700

  ITS#6981 don't flush entries until after cleanup callbacks

This sounds like it breaks REP_ENTRY_MUSTRELEASE's original purpose:

commit b92387bb1caf75bac5bf308d1780204a36ceb5dc
Author: Howard Chu<hyc@openldap.org>
Date:   Sun Apr 24 22:21:57 2005 +0000

ITS#3671 add REP_ENTRY_MUSTRELEASE flag, release the entry before
attempting to write the response on the network. (This should probably
be the default behavior anyway, since network hangs of this sort can
cause problems for any backend that has entry locks.)

Hm, good point. Sounds like this last commit should be reverted then.

...

I suppose the solution ought to be the opposite: Move the cleanup call too above send_ldap_ber(). Dunno how feasible that is.

No, that's no good either.

I guess a more reasonable fix is to see why pcache is mucking with response entries in op_cleanup instead of op_response.

Which is due to ITS#5112.