The thing is we only see that error once every so often. Additional, since Cassandra is not logging a shutdown message then it must be a violent termination, which leaves no traces in the system logs. It's possible that there is something wrong with the hardware, but the OS side I don't see what would be killing it. The fact that it happens after compaction, usually of the hints leads me to think that there is ghost hiding there that we had the great luck to find.

On Wed, Oct 12, 2011 at 8:57 AM, Jonathan Ellis <jbellis@gmail.com> wrote:
I'm comfortable with saying that there's some problem with your
environment that hasn't been identified yet, because we see many many
people running 0.7.9 and it just does not die randomly.

Again, the exception you see in the log is consistent with being
killed externally (and not consistent with dying from a Cassandra
bug).