DM-39188: Export raws only if they are not already in the central repo

[hsinfang]

I hesitate about this because adding a query with the central repo means time spent waiting for the database query, and it probably won't scale well. If we can agree just don't export raws in all cases, it seems better just to remove it. Do we really want to wait for DM-36051.?

[hsinfang]

Change of mind after some Slack discussions. We can just not export raws at all.

[kfindeisen]

The addition of second_local_repo looks fine in itself, but I don't see any tests that actually run interface and second_interface (or at least their export_outputs method?) in parallel. Then again, testing for parallelism issues is hard, since the test usually needs to put the system under a fair bit of stress...

[hsinfang]

Thanks for catching that. I'm adding it to test_export_outputs but indeed they aren't really run in parallel and I'm not sure how to enforce that in the current framework.

[kfindeisen]

Would something go horribly wrong if you used multiprocessing from inside a test?

[hsinfang]

As a multiprocessing newbie.... if I just use it from inside a test to do a dummy test, it can pass. But using the middleware interface gave me TypeError: cannot pickle 'lsst.afw.cameraGeom._cameraGeom.Camera' object .
Will see whether I can avoid sending non-pickleble objects...

[kfindeisen]

Sorry, I forgot that multiprocessing needs everything to be picklable. 😞 Since the actual framework never tries to share MiddlewareInterface objects, it's probably not worth designing them for it...