Slow DESCRIBE

[ktk]

We use DESCRIBES on SHACL shapes so one can potentially get a lot of results back. For that particular query, the first run takes quite some time: https://qlever.cs.uni-freiburg.de/lindas/rer85n

[joka921]

I just had a brief look at this.

1. This entitity is connected to rather long rdf collections (<el1> <el2> <el3>) which expand to triples with blank nodes according to the RDF standards. QLever's implementation of the SHACL DESCRIBE currently follows all blank nodes recursively, and the maximal depth of these blank nodes is the limiting factor here ( a single large chain of blank nodes gives the worst ratio of computation time to output result as in your example).

2. The first easy fix would be to not follow rdf collections or to limit the number of hops.

3. The more involved fix would be to make the DESCRIBE implementation more efficient by making it cheaper per hop. This can be done by e.g. some kind of caching, especially for datasets that are rather small, but have rather long reification chains.