Cache programmatic API - invalidate entries whose keys match a predicate

[mkouba]

Currently, you would need to make use of CaffeineCache, then iterate over the key set and finally remove an entry if it matches a predicate/condition.

[quarkus-bot]

✔️ The latest workflow run for the pull request has completed successfully.

It should be safe to merge provided you have a look at the other checks in the summary.

[gwenneg]

Nice improvement, thanks! This could use a new section in the cache doc.

LGTM

[mkouba]

This could use a new section in the cache doc.

Yes, the Caching using the programmatic API could describe all io.quarkus.cache.Cache methods.

[karesti]

@mkouba @gwenneg while I understand the motivation behind this method, for caching servers such as Redis or Infinispan looks like a dangerous idea to me (all the keys need to be read, apply the predicate and then send remove operations). I would like us to consider the idea of deprecating this method in the SPI

[karesti]

@cescoffier @geoand

[mkouba]

@mkouba @gwenneg while I understand the motivation behind this method, for caching servers such as Redis or Infinispan looks like a dangerous idea to me (all the keys need to be read, apply the predicate and then send remove operations). I would like us to consider the idea of deprecating this method in the SPI

Hm, I don't think we should deprecate the method but instead throw an UnsupportedOperationException in Redis or Infinispan impls and add a note to the method javadoc. It's perfectly valid for inmemory impls such as Caffeine (which was BTW the only impl we had at the time the method was added ;-).

[karesti]

@mkouba sounds like the method was promoted because we had it in the library implementation we were using. I believe this method does not belong in a Caching SPI, but this is not my call

[mkouba]

@mkouba sounds like the method was promoted because we had it in the library implementation we were using. I believe this method does not belong in a Caching SPI, but this is not my call

Well, that depends on how the cache is used. And it's actually not part of an SPI but a part of public user-facing API.

[karesti]

@mkouba looks like if it is also part of an SPI now, since we need to implement the interface to be used by the interceptors and chose a different cache provider

[karesti]

but as I said earlier, it's not my call :)