[rdf4j-dev] Shacl performance

[Håvard Ottestad <hmottestad@xxxxxxxxx>]

Hi everyone, and Bart in particular

3.0.0-SNAPSHOT is considerably faster for single-shot validation. 

Validating 500 000 triples worth of DCAT data against a fairly comprehensive SHACL file runs in around 6 seconds on my aging laptop. 

This is achieved by generating SPARQL queries for common use cases like minCount 1, maxCount 1 and class. If you add your data in an IsolationLevels.NONE transaction with a MemoryStore, and don’t need RDFS reasoning, it will skip the entire change detection approach and just use your memory store directly. 

You can also call setPerformanceLogging(true) to log (INFO) performance of each SHACL shape. Recommend you disable caching and parallel to get a correct result. 

Håvard