Full_Name: Marc Patermann
OS: SLES 11 SP1 x86_64
Submission from: (NULL) (188.8.131.52)
thankfully Howard tracked down the issue I posted on openldap-technical as
"provider crash on high replication load" down to:
"I believe the problem is much simpler - the backglue code doesn't install a
handler for the Abandon operation, and this is preventing syncprov's abandon
handler from running, so it never gets to clean up when a consumer connection
By now you should submit a bug report to the ITS so we can track this further.
It should only require a small patch to the backglue code."
The provider has three databases, which are glued together under one root
Each of the three subordinates is replicated by about 60 consumers.
On heavy replication load (all 180 consumers do initial sync), I can crash the
server with on little mod on an object in one of the subordinate databases.
If I can provide any further information on this, please let me know.
A patch for backglue.c is in git master, please test. Thanks.
-- Howard Chu
CTO, Symas Corp. http://www.symas.com
Director, Highland Sun http://highlandsun.com/hyc/
Chief Architect, OpenLDAP http://www.openldap.org/project/