Workflow [PR], commit [df6ac46]

Summary: ❌

Are those tests still flaky? This comment confuses me #75876 (comment)

Are those tests still flaky? This comment confuses me #75876 (comment)

I don't see any problems right now, but it still looks like worth to use separate caches, less contention for DROP CACHES, and later it can gain the benefit of filtering by cache_name in system.filesystem_cache

I am reluctant to use separate caches, because I remember clearly that as those tests reuse the same cache, it used to find many bugs in cache, because query to system.filesystem_cache triggered assertCacheCorrectness being called, which is only called on drop filesystem cache and read from system.filesystem_cache (calling it in other places can slow down tests too much)

(calling it in other places can slow down tests too much)

Though calling it with randomization in more hot places could solve it as well.

Better to do this explicitly I think, maybe just add drop filesystem cache after each test? Or not after each (to catch more possible issues) but after X tests (where X some random value)