This bug is created as a clone of upstream ticket:
We had been running a RedHat-supplied 389-ds-base implementation (126.96.36.199). This implementation caused TCP latency spikes of 1000ms and multiples under load (up to 100 searches/s, 10-30 concurrent connections), leading to TCP retransmissions, aborted connections and generally bad performance.
A later version, (188.8.131.52) from the 389 Directory Server project did not show any improvement.
An even later version which we are using currently in production (184.108.40.206) has fixed the latency issue. We suspect that a patch in 220.127.116.11 has brought the improvement (Ticket #48341: 0001-Ticket-48341-deadlock-on-connection-mutex.patch). With version 18.104.22.168, however, we are experiencing frequent crashes of loaded server (every couple of hours). We have not been able to reproduce the issue with artificial load patterns in our development environment.
Currently we are trying to upgrade to 22.214.171.124 (or by now: 126.96.36.199) but are waiting for a fix for shadow attributes.
As there had been no significant changes to the connection handling since 188.8.131.52, I would like to flag the issue even before upgrading to 1.3.5.x (with debuginfo enabled for core dumps).
What we are seeing so far is:
* crashes are segfaults
* thread #1 is crashing
* the crashing function is connection_table_move_connection_out_of_active_list (in ldap/servers/slapd/conntable.c)
* connection id is 0, fd is 0
* an error log line "connection - conn=0 fd=0 Attempt to release connection that is not acquired" is always the last logged line before the crash.
This pattern is *always* the same.
I will try to supply an up-to-date stack trace or a core dump once we have migrated to the current 1.3.5.x version
Cannot reproduce the problem in house.
Much more likely, fixed by NS 0.2.0. I think close with RHEL7.4 and 1.3.6.