Shared Chunk Cache for all DatasetArrays, CacheWeight for AlfuCache

[fm3]

• The cache for remote dataset array contents can now have a configured size in bytes. New config option datastore.cache.imageArrayChunks.maxSizeBytes. Default is 2 GB, consider increasing for production.
• For this, one cache object is shared between the dataset arrays
• Switches from akka cache to Caffeine directly (akka uses caffeine internally as well), allowing to add cache item weighing (weight is now set to bytes per chunk for array chunks)
• Unified some caching usages / naming (no more distinction between AlfuCache and AlfuFoxCache)

TODO

• use shared cache also in Precomputed + Neuroglacner classes
• include vault path in cache key
• how to clear cache for individual datasets?
• null pointer exception for loading volume annotation buckets

URL of deployed dev instance (used for testing):

• https://cacheweight.webknossos.xyz

Steps to test:

• Load some datasets, should still show
• annotate some, should still show
• use reload button and look for logging (cache chunks should be removed, but only for the reloaded layer)

Issues:

• fixes Share chunkContents cache across multiple arrays #6630
• contributes to Unify backend-internal caching #6969

---

• Updated changelog
• Considered common edge cases
• Needs datastore update after deployment

[frcroth]

Maybe also change PR name?

[frcroth]

Not totally happy with this being placed here, since the definition of the key is somewhere else. Maybe move this predicate to where the key is defined and only use it here?

[fm3]

Hmm, yes I see what you mean. However, the same would be the case for the other two predicate functions here. So we could either extract each of those so that there is only one place for each cache, where cache key building happens. This would lead to less distributed knowledge on how the keys work. However, I also like all of these methods being together here, since it is less distributed knowledge about the fact that these caches should be cleared dataset-wise or layer-wise. So it’s some fragmentation either way. If you’re ok with it, I’d leave it as it is for the moment.