Discussion and ideas to archive products done with a previous version of a pipeline and allow to create new versions.

Objective, definitions

Until now, most of our pipelines changes could be labeled as "revisions" : scripts were improved but results were expected to remain the same.

We aim now to deal with "versions" changes where a script modifications can include complete workflow overload and different (hopefully better) results.

In the event of a pipeline version change, it is expected that pre-existing products could be re-created using the newer pipelines, while keeping the previous result as an "archived" state.

It is not expected for archived products to be advertised in the interface, but they *must* be accessible for the purpose of reproducible science.

Design

Archived products accessibility

Archived products are to be explicitly accessed by productids. Archived versions of a same productid should be distinct by version, keep their creation time and also present an archival time.

As stated before, while archived product access is mandatory, a user should not be exposed to "deprecated" versions in their day-to-day workflow. Archive presence will not be exposed in the current interface elements such as map/search[1]/workspace view.

[1] Maybe with a specific keyword?

Dependencies

MarsSI products have a dependency fields since the major DB rewrite.

What about an archived products dependencies? We cannot archive deps at the same time, they might be still up-to-date. Forget deps as they are used foremost for procesing logic?

What about an archived product that is a dependancy? It should trigger a archival of the other products too. But their own pipeline version do not change... Use mctime?

Implementation

Data

Files are to be moved to /data/archive/$TARGET/$INSTHOSTID/$INSTID/$VERSION (ex /data/archive/MARS/MRO/CTX/CTX_008716_1998_008782_1999/V1)

Create an archive job and trigger it from the database? Less efficient but cleaner and safer than managing DB on one hand, CPs on the other.

Database

Rework a bit the Product model:

• Create a BaseProduct abstract model: hold common fields
• Product keep the name, inherit BaseProduct and status/depends fields, most methods
• Create an ArchivedProduct modeI, BaseProduct + version fields

Web

Provide a set of search/display interface that allow for copy operations, like workspace?

ArchivedProducts/Screenshot_2023-07-06_at_14.59.38.jpg does not exist. Create it?