Pratyush Banerjee wrote:
Hi All,
I have been using hadoop on a 50 machine cluster for sometime now and
just wondered why the mapred.tasktracker.expiry.interval defaulted to
10 minutes.
If I want to reduce it to 1 min i.e. 6000 msec, should that cause any
problems.
mapred.tasktracker.expiry.interval is the amount of time the JT waits
before declaring the TT as lost i.e kill all the tasks that were
executed at the tracker and re-execute them if needed. This time should
be sufficiently large to distinguish between network jitters/lags and a
dead host.
Also do i need to modify mapred.task.timeout accordingly to suit my
reduced mapred.tasktracker.expiry.interval ?
This is the time for the tasktracker to declare a task as dead/stuck.
These are two independent parameters.
Amar
thanks and regards,
Pratyush
