Arvados

A user who needs to run the same pipeline on many samples in different collections should have a simple mechanism to combine the output from those pipelines into a single collection.

Possible implementation:

For now, this should take the form of a crunch job that can be run locally on a VM, launches a new pipeline instance for each input collection, waits for all pipelines to complete and then merges their output collections into a single output collection.

Make sure we think a bit about the most user friendly way to solve this problem. Maybe there's a better way than the proposed implementation?