Midpoint nodes crash with OutOfMemoryError when recompute (or other iterative tasks, not sure) runs. It may happen on larger volumes of identities (sixty thousands in my case) in midpoint repository only.
Memory dump analysis from one of the incidents shows two worker threads retaining 90% of memory in midpoint.repo.cache.Cache.objects HashMaps. See attached MAT screenshot.
Please ask for password to the attached memory analyzer html report over private message.
Not sure how to fix this, repo cache needs explicit invalidation of entries which may be a root cause. My grandma always told me never to implement custom cache solutions with strong references.