If the last operation is lazy, cache the result independently of lazy-result-max-cache-size

[hannahbast]

Since #1638, QLever has the runtime parameter lazy-result-max-cache-size, with a deliberately small default value of 5 MB. However, I would argue that the final result should be cached independently of the value of lazy-result-max-cache-size but just depending on the CACHE_MAX_SIZE... settings, even if the last operation is output-lazy.

Here is an example query that is quite expensive to compute (all street intersections in the United States), but where the final result has "only" a size of 11M x 5 ~ 440 MB. So no problem for the normal cache, but too large for the lazy cache.

https://qlever.cs.uni-freiburg.de/osm-planet/1DHcyB

When repeating the query, the result before the last operation is stored in the cache, but the lazy join at the end is repeated every time. This is also a good example for a query, where the intermediate results should not be cached (but the final result should be).

@RobinTF @joka921 What do you think?

[RobinTF]

@hannahbast I think this is a reasonable thing to do, especially because the final LIMIT is not applied for the last operation, so even with limit it is likely to be bigger.

[hannahbast]

@RobinTF Thanks for the feedback + I agree. Can you pinpoint the location in the code, where this change would have to be made? It looks like a small change to me, but maybe I am overlooking something?

[RobinTF]

The max size is defined here:

qlever/src/engine/Operation.cpp

Lines 232 to 234 in 2953f16

	auto maxSize =
	std::min(RuntimeParameters().get<"lazy-result-max-cache-size">(),
	cache.getMaxSizeSingleEntry());

You'd have to pass the "isRoot" bool to this function though