Created by: stephenroller
Patch Description As a mild convenience, we want the indices of datasets to be built automatically when they're first used. However, if we build them on every worker, we can easily flood NFS and the process is slow. However, if we build them on the primary worker, then it only takes a couple minutes to handle several TB.
This PR causes the indexes to be built on the main worker, while the rest take a break.
Testing steps Used in internal ablation