[Libraries] [Collections] [Backend] Soft Delete collections

[pomegranited]

Currently, only "enabled" collections are displayed in the frontend, so "disabled" collections can be considered to be "soft deleted".

Implement APIs and views to support "soft deleting" collections, and add Django Admin pages for our collection models.

openedx-learning

Python API:

• Add "Delete Collection" method(s) which allow users to soft-delete, re-enable, or permanently delete a collection.
• Permanently deleting a collection should also remove all its associated entities -- please update the model if needed, and ping Dave Ormsbee for confirmation.

Add Collections model Django Admin, using basic configuration options to:

• List Collection fields: title, key, enabled, created, modified
• List search fields: title, key
• List filter fields: enabled
• Editable Collection fields: title, description, enabled

Don't worry about making a Collection's components add/removable here (because there can be lots), we're just worried about Collection editing for now.

edx-platform

• Refactor how the collection events are triggered, so that wherever collection or its components are modified or deleted, the search index gets updated. Use django model signals wherever possible.
• Ensure that the search index is updated for each component in a soft/hard deleted or restored Collection.
  Use async tasks since there may be a lot of components to update.

Add library collection views:

• "Soft Delete Collection" view: calls oel Soft Delete API method; event handlers will do the rest.
• "Re-enable Collection" view: calls oel Re-Enable API method; event handlers will do the rest.

Out of scope

1. Soft-deleted collections are not automatically cleaned up -- Django Admins must do this manually.

[bradenmacdonald]

@sdaitzman @lizc577 CC @jmakowski1123 Would one of you be able to answer @pomegranited's question here? When a collection is deleted, is it "gone gone", or should we kind of disable it and hide it, but keep the ability to potentially un-deleted it again in the future?

Just straight-up deleting it is the easiest to implement. Keeping it around but enabled may have issues if someone later tries to import a collection that has the same name/ID/key (though we can find a way to handle that).

[sdaitzman]

@bradenmacdonald @pomegranited I'll defer to @jmakowski1123 since this is more a product question, but here are some general thoughts. We can also talk more at our meeting later today.

If there's limited incremental cost in retaining collections for e.g. 90 days, I think it would make sense to implement that way. Even if collections are only recoverable by administrators/not through the UI initially, the ability to recover deleted content is important. Some trash/recovery workflow is common/borderline expected in a lot of authoring software with similar UI. At some point in the future, it would make sense to design and introduce a restore workflow. The existing "publish library" screen and publishing workflow are very simple right now, and we could add restoring/archiving content to that workflow in a future version. Smaller details like a temporary "Undo" button could also improve the experience.

If implementing soft-delete will add scope or more significant cost/delay, I think I'm okay with holding off for the libraries MVP. I'd prioritize shipping and getting real user feedback over a restore-collection workflow or recovery option.

[bradenmacdonald]

@pomegranited How much work will soft delete be? We discussed this on a call, and if you can find a relatively simple way to implement soft delete (so that e.g. an admin could un-delete it if needed), let's go with that for now. The main problem to solve is that if someone later imports (or creates) a new collection with the same key, we need to not throw an error (either by permanently deleting the soft-deleted collection or changing its key?).

[pomegranited]

Hiya @bradenmacdonald @sdaitzman Soft-deleting is simple to implement, and a Django-Admin-only UI for managing the details seems sufficient for now.

The main problem to solve is that if someone later imports (or creates) a new collection with the same key, we need to not throw an error (either by permanently deleting the soft-deleted collection or changing its key?).

Our current solution for creating collections handles auto-assigning and de-duplicating the key for new collections, so this won't be an issue unless we change the create flow.

I think we can implement this with 3 tickets:

1. [Backend] Delete Collection API + views + refactor event handlers (2 PRs)
2. [Backend] Manage Collections in Django Admin (1 PR)
3. [Frontend] Delete Collection with Undo Toast (1 PR)

openedx-learning

• Delete Collection API: allow users to soft-delete, re-enable, or permanently delete a collection.
  Permanently deleting a collection currently requires manually removing all its associated entities -- we should update the model to allow cascade delete.
• Add Collections to Django Admin: List/View/Edit collections; sort by title/key/created/modified; filter by enabled/disabled; search by collection key/title/description; add bulk actions to enable/disable/permanently delete selected collections

edx-platform:

• Search index refactor: trigger LIBRARY_COLLECTION_UPDATED on Collection model post_save; and remove any extraneous manual triggering of LIBRARY_COLLECTION_UPDATED.
  If collection was enabled/disabled, raise CONTENT_OBJECT_ASSOCIATIONS_CHANGED for each component in the collection too.
  We do this so that if Collections are disabled via our views or Django Admin, the search index always gets updated.
• Search index: update LIBRARY_COLLECTION_UPDATED handler; if collection was disabled, delete its document from the search index.
  (The CONTENT_OBJECT_ASSOCIATIONS_CHANGED handler already stores only enabled collections for an object, so no work to do here, but see below.)
• "Soft Delete Collection" view: calls oel Soft Delete API method; handlers do the rest.
• "Re-enable Collection" view: calls oel Re-Enable API method; handlers do the rest.

frontend-authoring:

• Delete Collection: Add action to the ... menu , with Undo Toast handler.

Out of scope

1. Soft-deleted collections are not automatically cleaned up -- Django Admins must do this manually.
2. There's a potential performance concern with updating the search index -- deleting a collection from the index is quick, but we also need to remove it from all of the content objects the collection contains. Currently we handle these search index updates synchronously for responsiveness in the MFE (cf discussion). But when a deleted collection contains many content objects, we may need allow for asynchronous reindexing of its content objects.

[bradenmacdonald]

@pomegranited Yeah I'm more worried about imports in the future (where we need to keep the key the same as whatever's in the imported data file) than creating collections. We can solve that problem down the road; we're not implementing imports for MVP.

[Backend] Manage Collections in Django Admin (1 PR)

I wouldn't spend much time on this. We don't actually have a requirement to allow soft-deleting collections via the admin or any other means. It's just if we have the option of implementing it as soft-delete or hard-delete on the backend, for now we'll go with soft-delete, but we don't have to implement the whole delete-then-recover workflow in the admin UI.

when a deleted collection contains many content objects, we may need allow for asynchronous reindexing of its content objects.

Good point. I think that's fine. We can just make all collection deletions async in terms of updating the items.

[pomegranited]

Thanks @bradenmacdonald .

I wouldn't spend much time on [Django Admin].

Cool -- that drops us down to two tickets for this change. I've updated the description here to describe the backend changes needed.

We can just make all collection deletions async in terms of updating the items.

That's the problem -- currently the CONTENT_OBJECT_ASSOCIATIONS_CHANGED event handler updates the search index synchronously, because each individual update is small. If we refactor the event handlers like I've proposed, we'll be sending a CONTENT_OBJECT_ASSOCIATIONS_CHANGED event for each component in a soft or hard-deleted collection, and they'll all be handled synchronously too. I don't currently have a way to send events asynchronously (or to indicate in the event data that this event should be handled asynchronously).

Still ok to leave that part out of scope here?