Artem

@ktk no, issue still exists. There is some old draft PR but it is out of date - https://github.com/ad-freiburg/qlever/pull/1534 Still need to maintain a fork to change prefix size.

@JervenBolleman maybe try to re-build the image on that machine with `-DADDITIONAL_COMPILER_FLAGS="-march=native"` in https://github.com/ad-freiburg/qlever/blob/master/Dockerfile#L36

This issue is still present and occurs very often even under light load for us at artresearch.net I think it is enough to have few concurrent requests to hit it....

It looks like it somehow related to pinned patterns, or at least with them it occurs very very often. On our dev system we don't pin often used predicates and...

@hannahbast I think the problem is in eager vs lazy. In our case we have ASK and SELECT queries with exactly same patterns, can it be that they cache key...

For this particular query can't one rewrite VALUES into explicit UNIONs + also move `?subject a .` to outside to reduce join size. E.g something like: ```sparql SELECT ?clazz (COUNT(?subject)...

@JervenBolleman I typically just construct a massive query with N unions, qlever handles it quite well. In your example, I think, it will evaluate ``` GRAPH { ?subject ?target ....