A more standard approach could be converting the JobContext parameter of
the getSplits() method, into a Job object, which then allows retrieving the
counters, e.g.:
public List<InputSplit> getSplits(JobContext job) throws IOException
{
...
Job work = new Job(job.getConfiguration());
Counters counters = work.getCounters();
}
Would that be correct?
On Fri, May 17, 2013 at 5:33 PM, Public Network Services <
[email protected]> wrote:
> Hi...
>
> I need to communicate some proprietary number (long) values from the
> getSplits() method of a custom InputFormat class to the Hadoop driver class
> (used to launch the job), but the JobContext object passed to the
> getSplits() method has no access to a Counters object.
>
> From the source code, it seems that the Configuration object of the
> launched job is passed around, so the JobContext object of getSplits() has
> direct access to it via getConfiguration().
>
> So, what about using a loop like
>
> Job job = ... // The launched job
> Configuration conf = job.getConfiguration();
> while (!job.isComplete()) {
> // Read the values from the configuration
> }
>
> from the driver class, which presumably runs in the same framework that
> creates the splits?
>
> The getSplits() method of the custom InputFormat would set each of the
> values once.
>
> All this does seem like a hack, so I would like some expert advice before
> starting implementation. That is,
>
> 1. Will it work?
> 2. Is there a better method?
>
> Thanks!
>
>
