I have a problem with dialog replication on two active-active opensips servers, installed with the latest opensips version 2.3.2 on Debian stretch.
A b2bua front end with a load balancer pushes traphic to one of the two opensips servers. All three servers (front and two backends) are in a cluster, both backends share the dialogs. If I start stressing the platform with traphic the package memory on one or two workers start growing. During off hours without much trafic memory stays the same, but doesn't shrink again, till out-of-memory and everything crashes. Dialogs disappear on both servers like expected if a call is ended. Another strange thing is that early dialogs (state 1 or 2) are not shared on the cluster, after pickup a dialog is visible on both backends.
After reading some topics on the forum, Out Of Memory docs and doing some troubleshooting I could use some help. Any ideas what I could be doing wrong (if more info is needed, no problem).
Some settings on the two active-active backend servers:
children = 12
listen = bin:10.85.16.133:5555