force native order when wrapping ByteBuffer

[AlexanderSaydakov]

Fixes #8032

force native order when wrapping ByteBuffer since Druid might have it set incorrectly

[AlexanderSaydakov]

This was already forced every time a ByteBuffer from Druid is wrapped for use with Datasketches library, except this one instance that was missed by mistake.

[jihoonson]

Would you please add a unit test?

[jihoonson]

There is a helper method called AggregatiohnTestHelper.runRelocateVerificationTest() which could facilitate writing unit tests for this.

[himanshug]

this is unfortunate (discussed in #6381 (comment) as well ) . I wish there could be a better solution to handle this.

[AlexanderSaydakov]

Would you please add a unit test?

Unfortunately I don't have a unit test. The bug leads to sporadic failures. I could not find a small use case to reproduce it.

[himanshug]

@AlexanderSaydakov were you able to verify that it does fix #8032 , sounds like it was not reproducible ?

[AlexanderSaydakov]

I managed to get a similar failure in a micro-quickstart environment using the same 20 sketches from a test: https://github.com/apache/incubator-druid/blob/master/extensions-core/datasketches/src/test/resources/quantiles/doubles_sketch_data.tsv
The proposed change fixes that case. But I could not find a scenario that would fail in a test in development environment. I am almost sure that the proposed change should fix #8032

@quenlang would it be possible for you to give this a try and let us know if this change fixes your problem? Thank you.

[himanshug]

I see, but anyways, this fixes some bug if not #8032 .

[clintropolis]

👍

[quenlang]

@AlexanderSaydakov I had a try with this patch, the query worked. Thank you very much!