[ejabberd] PEP being memory hungry?

Yao Ko koyao at raptr.com
Mon Sep 20 22:17:03 MSD 2010


I've turned on mod_pubsub_odbc on a cluster of about 6 nodes.  All was
well until a few days later, some nodes started running out of memory
and died.  This caused a cascade event and caused other nodes to fail
as well.  After a few days of investigation, we turned off pubsub and
everything came back to normal.

Here's the relevant config:

{mod_pubsub_odbc,   [
              {access_createnode, pubsub_createnode},
              {ignore_pep_from_offline, true}, % reduces resource
comsumption, but XEP incompliant
              {last_item_cache, false},
              {plugins, ["pep_odbc"]}  % pep requires mod_caps

I found a Google post about PEP being memory hungry:


relevant quote: "It appears that the issue is isolated to the PEP
plugin for pubsub, specifically with the entity capability cache, and
the ejabberd team is investigating it further."

Not sure how relevant the article is, since it dates back to 2008.

Is there anything that can be done to fix or mitigate the memory issue
with PEP?   Does the {max_fsm_queue, undefined} setting that I've set
in the config have anything to do with the memory problem?


More information about the ejabberd mailing list