Enqueueing regional jobs uses too much memory about analysis-backend

[landonreed]

So if we stick with the simple solution for now, is this simply a matter of splitting these requests up into smaller chunks?

What do you think an appropriate cluster size is for these requests? And does it need to be anything more complex than a loop over the list of GridRequests?

[abyrd]

Leaving this ticket open because I think we should eventually merge the broker into the backend, it's way simpler.

[abyrd]

The process of enqueueing now uses a single request object. It is still done over an HTTP API, but it's O(1) in memory and time now. Closing, because we have a separate ticket for eliminating the HTTP interface and enqueueing directly.