What does "Lost Task Tracker" Mean?

[Albert Chern <al...@gmail.com>]

Hello,

One error I seem to be getting a lot is "Lost Task Tracker".  The two
reduce tasks that a particular machine is working on will fail with
this error, which then propagates back to all the map tasks that it
has completed, causing them to reexecute.  What exactly does this
message mean?  I would assume that it means contact with the task
tracker has been lost entirely, but the machine is still reported as
part of the cluster (although none of the reexecuted map tasks seem to
be assigned to it).

Thanks,
Albert

Re: Re: What does "Lost Task Tracker" Mean?

[Albert Chern <al...@gmail.com>]

Hey Doug,

Thanks for the help.  I figured out what was causing it, though.  It
was because some of the machines were running out of temporary disk
space for the reduce operation.  This was with version 0.7.

On 11/27/06, Doug Cutting <cu...@apache.org> wrote:
> Albert Chern wrote:
> > One error I seem to be getting a lot is "Lost Task Tracker".
>
> What version of Hadoop are you running?
>
> > What exactly does this message mean?
>
> It means that a heartbeat from that tasktracker has not been registered
> with the jobtracker for a period.  There were bugs which made this more
> common in some past releases.
>
> Doug
>

Re: What does "Lost Task Tracker" Mean?

[Doug Cutting <cu...@apache.org>]

Albert Chern wrote:
> One error I seem to be getting a lot is "Lost Task Tracker".

What version of Hadoop are you running?

> What exactly does this message mean?

It means that a heartbeat from that tasktracker has not been registered 
with the jobtracker for a period.  There were bugs which made this more 
common in some past releases.

Doug