On the test instance I noticed an ever increasing use of disk storage, for storing the original, raw documents.

While it needs further investigation, I assume this is partly true to do re-ingesting documents with the same ID. Overwriting existing state, but keeping duplicates in the storage backend. Currently the strategy there is that we don't clean up.

The question is: should be change this strategy? One opinion I read about this, in the context of Mastodon IIRC, was that storage is cheaper than API calls cleaning up storage. On the other side, constantly increasing storage (and cost) just for being lazy also feels wrong.