Bug: Undirected query with DISTINCT fails unpredictably on Windows

[prrao87]

Kùzu version

v0.8.0

What operating system are you using?

Windows

What happened?

Arthur from G.V() ran the following undirected query with a DISTINCT keyword on the LDBC1 and LDBC10 datasets using the Java API. On MacOS, it runs pretty flawlessly, but on Windows, it randomly succeeds (many times it crashes Kùzu)

 MATCH ()-[r]-() RETURN DISTINCT r LIMIT 1000

The task would be to first reproduce the crash on Windows, and see how the behaviour differs from what's happening on macOS.

Are there known steps to reproduce?

Run the above query on a small LDBC1 or LDBC10 dataset on Windows, preferably through the Java API (though the issue is likely at the C++ layer).

[benjaminwinger]

Does this occur if you run with just one thread?

[prrao87]

cc @arthur-bigeard-gdotv

[arthur-bigeard-gdotv]

I'll check first thing on Monday (few hours ahead of you guys so already in weekend mode :D). Just to confirm, is there a specific way to configure a Kuzu connection to be single threaded?
In the context of G.V() Kuzu operations happen in a Spring Boot app, for this example as the result of running a query - Spring Boot itself is multi threaded so it feels safe to assume we would be running in a multi-threaded environment in this case.

[prrao87]

@arthur-bigeard-gdotv seems that this could be linked to a data race issue on windows. That's why testing with single thread makes sense.

You should be able to force single threaded behaviour when you instantiate the Kùzu database object. I'm not sure about the exact argument but the Java docs should show this.

[arthur-bigeard-gdotv]

Got it, i can test that easily, I'll let you know ASAP

…

On Fri, 14 Feb 2025, 18:01 Prashanth Rao, ***@***.***> wrote: @arthur-bigeard-gdotv <https://github.com/arthur-bigeard-gdotv> seems that this could be linked to a data race issue on windows. That's why testing with single thread makes sense. You should be able to force single threaded behaviour when you instantiate the Kùzu database object. I'm not sure about the exact argument but the Java docs should show this. — Reply to this email directly, view it on GitHub <#4903 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BJSFPA5LDT2L7HFZ5UZN5FT2PYVP3AVCNFSM6AAAAABXC5G7YCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMNJZHE2TEMRWGI> . You are receiving this because you were mentioned.Message ID: ***@***.***> [image: prrao87]*prrao87* left a comment (kuzudb/kuzu#4903) <#4903 (comment)> @arthur-bigeard-gdotv <https://github.com/arthur-bigeard-gdotv> seems that this could be linked to a data race issue on windows. That's why testing with single thread makes sense. You should be able to force single threaded behaviour when you instantiate the Kùzu database object. I'm not sure about the exact argument but the Java docs should show this. — Reply to this email directly, view it on GitHub <#4903 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BJSFPA5LDT2L7HFZ5UZN5FT2PYVP3AVCNFSM6AAAAABXC5G7YCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMNJZHE2TEMRWGI> . You are receiving this because you were mentioned.Message ID: ***@***.***>