This reverts commit 718f1d09083104be061e35bf0600ba11034b2a45.
I tried plumbing the subset selection through (via get_subset_external_pipeline_result), but there's a couple calls that only passes through solids_to_execute and not solid_selection, and both are used to generate the original snapshot id.
If you think it's worth it, I can keep at it, but we're not saving gRPC calls since we have to call out to subset the pipeline (and generate the valid snapshot id) anyway. Not sure if the main part of the optimization was not calling to gRPC or if the difference between constructing the plan and constructing the pipeline subset snapshot is significant.
No worries. If there’s a test you’re adding to backfills for the thing you’re working on that caught this, that would break if I ever tried to I revert this later, that would be cool.
I can do this too later but we could also split out the backfill path from the non-backfill path, I mostly cared about the ‘retry a single run’ case