This box has just one JBoss and standard linux services. I didnt
change
default measurement intervals. So OOME cannot be due to excessive number
of metrics being collected, can it?.
That's a very minimal setup. The number of metrics won't be involved here.
Im pretty sure it relates somehow to my changes of the RHQ server
address. If agent cannot connect to the
server - does it keep metrics in memory to send them to the server
later? Or maybe failed connection retry leaks some resource? (socket,
memory or file handler)
If the server goes down, the agent will spool messages to disk and when
the server comes back, those messages will be loaded into memory and
sent to the server. The spool is upper-bound-limited through the
configuration (I think its something like 100MB - any new messages to
get spooled will push out the oldest messages).
We've seen instances in the past where failed connection retries causes
leaks, but that was a while ago and I think we fixed those in previous
releases. Search the bugzilla issues if you want to see if you can find
something similar.