No response from the HaDoop cluster then - stop/start map/reduce would solve the problem.

Note: HDFS has no such issue.
Is it a common problem (we use v.19)?

Thanks,
-Songting

There are some bugs in 0.19.1 and 0.19.0 that can cause this to happen.

Usually the one I've seen involves several task slots will fill up with "KILLED_UNCLEAN" tasks and the mapreduce side has to be restarted.
This has been fixed on the 0.19.2-dev branch for a long time, and 0.19.2 is due out very soon.

We've been using a cut off of the 0.19.2-dev branch here for a while in production with success.

On 7/6/09 12:23 PM, "Songting Chen" <[EMAIL PROTECTED]> wrote:

No response from the HaDoop cluster then - stop/start map/reduce would solve the problem.

Note: HDFS has no such issue.
Is it a common problem (we use v.19)?

Thanks,
-Songting

I believe reducing the kept job history can ameliorate this problem.

On Mon, Jul 6, 2009 at 12:39 PM, Scott Carey <[EMAIL PROTECTED]>wrote:

> There are some bugs in 0.19.1 and 0.19.0 that can cause this to happen.
>
> Usually the one I've seen involves several task slots will fill up with
> "KILLED_UNCLEAN" tasks and the mapreduce side has to be restarted.
> This has been fixed on the 0.19.2-dev branch for a long time, and 0.19.2 is
> due out very soon.
>
> We've been using a cut off of the 0.19.2-dev branch here for a while in
> production with success.
>
>
>
> On 7/6/09 12:23 PM, "Songting Chen" <[EMAIL PROTECTED]> wrote:
>
>
>
> No response from the HaDoop cluster then - stop/start map/reduce would
> solve the problem.
>
> Note: HDFS has no such issue.
> Is it a common problem (we use v.19)?
>
> Thanks,
> -Songting
>
>
--
Pro Hadoop, a book to guide you from beginner to hadoop mastery,
http://www.amazon.com/dp/1430219424?tag=jewlerymall
www.prohadoopbook.com a community for Hadoop Professionals