apoc.cypher.mapParallel2 version 5.5.0 to 5.13.0 migration and performance degradation

[alexanoid]

apoc.cypher.mapParallel2 version 5.5.0 to 5.13.0 migration and performance degradation

[alexanoid]

I tested again in a new environment, and it seems that 5.5 and 5.13 are woring approximately the same. So, I think there are no particular issues. But please do not delete apoc.cypher.mapParallel2 because at the moment, the new PARALLEL runtime cannot fully replace it. Details https://stackoverflow.com/questions/77509091/neo4j-parallel-runtime-query-returns-different-result

[conker84]

The overhead should be related to the rebind and something that changed into the database kernel. This feature, since the introduction of the parallel runtime is deprecated. Please consider migrating your query to the new runtime.

[alexanoid]

Hi @conker84 , thanks for your answer. I already tried parallel runtime in 5.13.0 Unfortunately, in my tests it works much slower than the approach with apoc.cypher.mapParallel2 like in the query above :(

[conker84]

Could you provide a flame graph taken with the Async Profiler of the two executions (5.5 and 5.13)?
This would help us.

[alexanoid]

Yes, I can try, but I have never used this profiler before. I hope it works on Windows 10, which is the system where I run my tests. There, I can run Neo4j of different versions in Docker. Perhaps you could provide instructions on what exactly I need to do?

[alexanoid]

If possible, please refrain from deprecating this feature as it operates much faster than the new parallel runtime. Just ensure it functions the same as in version 5.5.0.