Very different query plans for very similar queries, involving blank nodes

Support
1

Query 1 and 2 are similar but Query 1 explicitly uses blank nodes. The query details are attached with their query plans.

I also tried with a query using property paths - performance was pretty bad. I manually killed the execution after 15 secs (300x worse than Query2).

At the times that the OPTIONALs are called in each of the queries - the variables should be bound.

I can modify my queries, but why is this failing?

Andrea
Query1.txt (8.4 KB)
Query2.txt (7.0 KB)
Query3.txt (7.6 KB)

2

Hi Andrea,

We need to see the query profiler output to tell for sure (you can get it by adding the --profile flag to query explain). Note that you can still manually kill a query being profiled, the profiler will still collect the partial timings.

Re: queries, Query1 and Query3 (which both don't terminate) are essentially equivalent. There's not difference between ?s :p/:q ?o and ?s :p [ :q ?o ]. But Query2 is not and is substantially simpler because 3 out of 5 OPTIONALs are simple triple patterns and not BGPs of two triple patterns. Query2 can also potentially return different results from 1 and 3 (depends on the data).

Please do the profiling and then we can tell where the problem is.

Best,
Pavel

3

Pavel, Thanks for the reply. I refactored the code that I was using and rebuilt the databases.

The problem has disappeared ... so, clearly, it was my problem and not yours.

Thanks for the reply.

Andrea