Hello,
I've notice in an earlier 1.2.x that if I had a compaction throughput
throttle some of the nodes would give an out of memory error only if I was
inserting data for more than 10 hours continuosly. The work around was to
switch off compaction throttling.
This was in a test environment
Having (2) 1.2.5 nodes constantly crashing due to OutOfHeap errors.
It always happens when the same large compaction is about to finish (they
re-run the same compaction after restarting.)
An indicator is CMS GC time of 3-5s (and the many related problems felt
throughout the rest of the cluster)
