How do I get workers to retain a large object needed over several task graphs?

Distributed
3

I did attempt scatter for a similar purpose and I noticed that

  1. Same object seems to be transferred several times to a worker for a single compute call.
  2. The object definitely is re-transferred across compute calls.
    Available options for collective communication in Dask - #9 by vij

I am curious if there is a mechanism to specify during the scatter that the object needs to be preserved in the a local cache for future use and when the map calls are made the worker checks for the required key in local-cache before fetching it from other workers.