"You can ... register it in WritableComparator". Does this mean call
WritableComparator.define(), passing the user provided Comparator
implementation? Is that what you're proposing Ben?
JobConf.setOutputKeyComparatorClass() requires a WritableComparator,
which means the user would need to subclass that rather than just
implement the Comparator interface.
Alan.
Benjamin Reed wrote:
Exactly. Just to be clear. We are talking about how to implement the ordering.
Pig developers would just need to implement the Comparator class as outlined
by Alan.
ben
On Tuesday 06 November 2007 03:53:20 Andrzej Bialecki wrote:
Benjamin Reed wrote:
Using the compareTo method is just the default. We can define our own
WritetableComparator subclass to hook into the Hadoop sorting.
You can either register it in WritableComparator (though this is less
flexible - the comparator will be used in every case, no matter what job
runs in the JVM), or set it specifically on a per-job basis using
JobConf.setOutputKeyComparatorClass().
