You posted system specifics earlier; would you mind posting again? can't find them in the thread.
Sent from my iPhone On May 13, 2011, at 8:05 AM, Adi <adi.pan...@gmail.com> wrote: >>> Is there a reason for using OpenJDK and not Sun's JDK? > > The cluster we are seeing the problem in uses Sun's JDK java version > "1.6.0_21",Java(TM) SE Runtime Environment (build 1.6.0_21-b06),Java > HotSpot(TM) 64-Bit Server VM (build 17.0-b16, mixed mode) > > The standalone node where I tried to reproduce the issue uses OpenJDK and > this one does not see this issue as it is able to reuse JVMs. > > -Adi > > Also... I believe there were noted issues with the .17 JDK. I will look for >> a link and post if I can find. >> >> > >> Otherwise, the behaviour I have seen before. Hadoop is detaching from the >> JVM and stops seeing it. >> >> I think your problem lies in the JDK and not Hadoop. >> >> >> On May 12, 2011 at 8:12 PM, Adi <adi.pan...@gmail.com> wrote: >> >>>>> 2011-05-12 13:52:04,147 WARN >>>> org.apache.hadoop.mapreduce.util.ProcessTree: >>>>> Error executing shell command >>>>> org.apache.hadoop.util.Shell$ExitCodeException: kill -12545: No such >>>> process >>>> >>>> Your logs showed that Hadoop tried to kill processes but the kill >>>> command claimed they didn't exist. The next time you see this problem, >>>> can you check the logs and see if any of the PIDs that appear in the >>>> logs are in fact still running? >>>> >>>> A more likely scenario is that Hadoop's tracking of child VMs is >>>> getting out of sync, but I'm not sure what would cause that. >>>> >>>> >>> Yes those java processes are in fact running. And those error messages do >>> not always show up. Just sometimes. But the processes never get cleaned >> up. >>> >>> -Adi >>