We've continuously had concurrent modification exceptions for our regularly
occurring []s that run on thousands of shards, perhaps unnecessarily so.
These exceptions started after the last major [] framework refactoring,
which changed the default number of shards from 100 to essentially infinite. I
don't think infinite is the way to go, and 100 shards should be more than
sufficient for anything we're currently running.
-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=131175353
52bb8a5773