All work falls to final reducer in index_hadoop batch job

Hey all,

When I launch a batch ingest job with index_hadoop, it seems that the last reducer is the only one that actually writes the partitions to deep storage.

This makes it OOM often and take a signifigantly long time (3+ hrs).

I assume this is not expected behavior. Is there something that would cause all the work to get pushed to the final reduce stage?

Thanks,

Walt