Re: Trouble with "Connection Refused" errors and timeouts on server with high Thread Backload

[Quanah Gibson-Mount <quanah@symas.com>]

--On Monday, August 05, 2019 10:08 PM +0000 erich_2323 
<erich_2323@protonmail.com> wrote:

> Thanks for the reply Quanah.
>
> I am using MDB as by the database backend.
>
> objectClass: olcDatabaseConfig
> objectClass: olcMdbConfig
> olcDatabase: {2}mdb

Ok.  First, I would strongly advise updating to the latest OpenLDAP release 
(2.4.48).  Second, what is the mdb_stat -eaf output for the master's 
database?  Third, I would take advantage of the rtxnsize setting in newer 
OpenLDAP releases to ensure you're not fragmenting your database with these 
searches. Fourth, I would not use the patch as-is in that ITS, but backport 
what was actually committed to OpenLDAP master, as the patch in the ITS has 
some negative impacts on performance that the actual committed code 
addressed.

Additional things you may wish (a) ensure you have proper indexing for the 
filter you're using for the search (eq on entryCSN + whatever other 
components make up your search), (b) Your olcConnMaxPending of 100 only 
allows for 100 pending operations across all clients, it sounds like you 
may need to increase that (you say you have many read-only replicas, but 
don't provide a number).

Hope that helps!

Regards,
Quanah

--

Quanah Gibson-Mount
Product Architect
Symas Corporation
Packaged, certified, and supported LDAP solutions powered by OpenLDAP:
<http://www.symas.com>