We have a master ejabberd with a several slaves clustered together. We had some mnesia locking problems due to the mod_offline message queue so we disabled the mod_offline on all ejabberd's in the cluster and restarted. We rebooted the server and we're now seeing uncontrolled memory growth and the servers run out of memory every 12 hrs.
We saw this error in the erl dump file.
=erl_crash_dump:0.1
Tue Oct 26 00:40:17 2010
Slogan: eheap_alloc: Cannot allocate 10875019720 bytes of memory (of type "old_heap").
System version: Erlang R13B04 (erts-5.7.5) [source] [64-bit] [smp:8:8] [rq:8] [async-threads:0] [hipe] [kernel-poll:true]
Compiled: Thu
I don't know why disabling
I don't know why disabling mod_offline can cause memory growth. If you have erl_crash dumps, you can investigate them with the crash dump viewer, find what process grows so big: erl -s webtool
Workaround: instead of disabling offline storage, you can set the option access_max_user_messages