> Very useful to know about this possibility and glad you have a better
> idea about the root cause. Thank you very much for sharing this information.
>
> May I suggest, you contribute this as a FAQ entry in a suitable spot here ?
> http://wiki.glassfish.java.net/Wiki.jsp?page=GlassFish
It does not qualify as a "FAQ" strictly speaking :o) and I'm too new to Glassfish to know to which extent the problem is generic (or specific to some weird configuration of my OS).
Moreover, I'm afraid the investigations has not lead us to a full understanding yet. OK the oom-killer kills glassfish processes, but why?
We devised a Java memory-gobbler test utility which continually allocates memory, and we witnessed its memory grow the full memory size then use the full virtual memory and swap. The oom-killer kicked in only when the whole virtual memory neared exhaustion.
Contrast this with our glassfish cluster: the oom-killer kills the DAS and a node whereas there's still much memory left (and no swapping at all).
The info I found on the Web (and yes, the link you posted is about this subject), also tell that the oom-killer may be triggered as soon as the so-called "LOW" memory fills up, event though there's plenty of free room in the "HIGH" memory.
We have not prooved yet, whether glassfish would consume LOW memory faster than our memory gobbler...
Subject still under investigations so far.
[Message sent by forum member 'jduprez' (jduprez)]
http://forums.java.net/jive/thread.jspa?messageID=348737