[Date Prev][Date Next]
Re: (ITS#8433) slapd deadlocks on shutdown in 3+ MMR configuration with active writes
- To: openldap-its@OpenLDAP.org
- Subject: Re: (ITS#8433) slapd deadlocks on shutdown in 3+ MMR configuration with active writes
- From: firstname.lastname@example.org
- Date: Thu, 09 Jun 2016 15:47:08 +0000
- Auto-submitted: auto-generated (OpenLDAP-ITS)
--On Wednesday, June 08, 2016 11:39 PM +0000 email@example.com wrote:
> Thread 3 is inside OpenSSL, it thinks there are 433 bytes to read on this
> socket and it's waiting for them to arrive. There's not much we can do in
> slapd about this.
Going back through my notes on when we hit this before with customers, the
following are the steps necessary to trigger these deadlocks:
a) Must have 2+ replication agreements per node. This will not occur in 2
MMR node environments
b) starttls is used in all the replication agreements
c) There needs to be fairly consistent write traffic
After some point of time, slapd will then deadlock on shutdown. Or, if
allowed to run long enough, it will grind completely to a halt (2-3 weeks,
depending on traffic).
This started after we upgraded to openssl 1.0.1j. It is continuing to
occur with openssl 1.0.2h (just to ntoe).
Release and QA Engineer