qlever: Timeout with path inside UNION

On the Plazi dataset[^1], the following query gives a timeout:

PREFIX publication: <http://publication.plazi.org/id/>
PREFIX treatment: <http://treatment.plazi.org/id/>
PREFIX trt: <http://plazi.org/vocab/treatment#>
PREFIX dc: <http://purl.org/dc/elements/1.1/>
PREFIX dwc: <http://rs.tdwg.org/dwc/terms/>
PREFIX dwcFP: <http://filteredpush.org/ontologies/oa/dwcFP#>
PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>
SELECT DISTINCT ?tn ?tc ?tcauth ?kingdom ?tnrank
WHERE {
  <http://taxon-concept.plazi.org/id/Plantae/Calophyllum_brasiliense_Cambess_> trt:hasTaxonName ?tn .
  ?tc trt:hasTaxonName ?tn ;
      dwc:scientificNameAuthorship ?tcauth ;
      a dwcFP:TaxonConcept .

  ?tn a dwcFP:TaxonName .
  ?tn dwc:rank ?tnrank .
  { ?tn dwc:kingdom ?kingdom . } UNION { ?tn trt:hasParentName* ?k . ?k dwc:rank "kingdom" ; dwc:kingdom ?kingdom . }
}

Specifically, it returns a 429 with "exception": "Operation timed out. Last operation: TransitivePath ?tn <http://plazi.org/vocab/treatment#hasParentName> ?k"

Interestingly, replacing the UNION with either of its branches works fine (i.e. replacing the last line with ?tn dwc:kingdom ?kingdom . or ?tn trt:hasParentName* ?k . ?k dwc:rank "kingdom" ; dwc:kingdom ?kingdom . both work)

[^1]: The data can be found at https://git.ld.plazi.org/plazi/treatments-rdf, but a single file containing most of the data relevant for the above query is https://git.ld.plazi.org/plazi/treatments-rdf/src/branch/main/data/3B/0C/87/3B0C8780DDFAFF17CC35F904FB422245.ttl

About this issue

  • Original URL
  • State: open
  • Created 2 years ago
  • Comments: 24 (6 by maintainers)

Most upvoted comments

@RobinTF it’s now public. Sorry for the mistake.

@retog @nleanba

Okay, there’s definitely a discrepancy of our configurations:

Here is your execution tree: Plazi Error Tree

And here’s ours: AD working tree

Written in square brackets are the size estimates, which are 100% accurate for all IndexScans as you can see they’re way off. A little bit is expected as @nleanba reported that the datasets aren’t quite up-to-date but this seems to be more than just a little bit.

I also ran QLever with address sanitizer and a debugger with your dataset to see if there’s a buffer overflow somewhere, but I found nothing.

So that leads me to three possible conclusions:

  1. We’re loading completely different datasets which are the reason why we can’t reproduce this issue
  2. We’re loading the same (or sufficiently similar) datasets, but you have a data corruption issue on your end
  3. The QLever binary you’re using is corrupted, or creating with a compiler that creates a binary containing the bugs.

So what we need is a step-by-step guide how you create your index and start QLever, so we can try and reproduce that. Also since we’re running our server without docker, maybe you can try compiling QLever natively to see if that helps? If we can’t reproduce this on our end I’m afraid we won’t make any progress here I’m afraid.