stores
The problem is that `object_store.is_private` would check in which
object store the dataset is stored, and that fails if we haven't written
to the object store yet (which is why extended metadata is a
workaround).
Fixes https://github.com/galaxyproject/galaxy/issues/17208
In the extended_metadata + remote tool_evaluation_strategy the
commandline is templated as part of the job, so we need to give the test
more time to start running before we delete outputs (as then we can't
template the command line anymore).
Add a sleep so we can delay output purging until command line is
templated. If we don't sleep we generate a command line where the
output path is just `''`. That needs another fix!
I think I clung to the idea of having two ways to do this too long. I did a bunch of testing with the old way (exposing integer references based on numeric database primary keys). Having a config option that if changed would break everything exisiting is also a symptom of maybe me clinging too hard for too long.
The result of dropping this option is a much cleaner API schema, simpler API objects, and better typing throughout. We can also be more certain of how things entering the Vault are stored - I think being more strucutured about this is good.
Ultimately, the future facing stuff (e.g. OAuth 2.0) is going to require UUIDs so that I can store things like refresh tokens in the store before the object has been fully created.
Allow attaching a device ID on object stores that indicates they produce equivalent paths. Implement an API endpoint that allows swapping datsets between equivalent object stores if they have not been copied by new users.
Per @bgruening's request - s3 -> aws_s3 and swift -> generic_s3.
I think we only construct object stores with the one method - so this should be backward compatible.
mvdbeek