On Thu, Sep 15, 2011 at 1:16 PM, jiang licht <[EMAIL PROTECTED]> wrote:
> Lower priority jobs then have to either wait for availability of reducer
> resources and idle, or get smaller number of slots than configured...
This last part ("or get smaller number...") isn't right, is it? The number
of reducer tasks is a function of the Pig script design. While any
particular job can be forced to wait until it receives enough slots to
complete processing, the cluster state shouldn't force a change in script