Arvados

h1. Collections as regular Arvados objects

* Collections are assigned regular UUIDs

* Collections have a regular @owner_uuid@ instead of the system user

* Manifest hash goes into the @portable_data_hash@ field

* Add mutable @name@, @description@, @properties@ fields

* Continue to use manifest text hash (@portable_data_hash@) for crunch

** Choosers set the value to the manifest hash, record the selected collection uuid on link_uuid

** Task output is manifest fragment or manifest hash+reader token stored in Keep

** Crunch-job continues to set job output to manifest hash

** Add optional @output_name@ and @output_owner@ fields to job submission

** When job completes, collection is created under @output_owner@, given @output_name@, and the collection uuid is recorded under new @output_uuid@ field

(Side note: job should not be marked as successful/completed until all its output and logs have been uploaded.)

h2. Migration

* Create collection objects from name links