Attachments usually don't come through on ASF mailing lists.
Also, Mahout trunk requires Hadoop 0.20.0
On Aug 3, 2009, at 6:05 PM, Mark Desnoyer wrote:
Hi,
Has anybody had a problem running KMeans leaking TCP connections,
which eventually causes the OS to issue a "Too Many Files Open"
error? In other words, it looks like a file descriptor leak.
I'm running a four machine cluster of Hadoop 18.3, with the current
version of Mahout from the trunk. Each k-means iteration is causing
an extra 40 or so TCP connections to be created that get orphaned in
CLOSE_WAIT. I've attached the output of lsof from the middle of a run.
Any help would be appreciated!
Cheers,
Mark
