mr: tasks crash & tasks assign to old nodes

[Stefan Groschupf <sg...@media-style.com>]

Hi,
I notice 2 problems, but wasn't able to find the source until yet.
Does may someone notice the same problem and may already know the  
source of the problem?

First I notice that in case the local hard-drive is full the reduce  
job crashs without any reexecution on a other node.
Second I notice that the jobtracker assign somehow tasks to nodes  
that already down.
I had just restart one of my tasktrackers and than I had seen that  
the jobtrackers know the old (just the heartbeat was older) and the  
new tasktracker.
Not a problem, but when I process a job I notice that somehow the old  
tasktracker got a task assign but of curse wasn't processing it.

Any ideas what the sources of this problems could be, I would be  
interested to helpout fixing this problems.

Stefan