Flink: Fix timestamp-micros conversion in AvroToRowDataConverters#17194
Conversation
|
is this the same one as #17078? We need a test to prevent regression which fails before the patch, and succeeds after the patch. Thanks, |
Signed-off-by: Sergei Nikolaev <kinolaev@gmail.com>
054ce4f to
a466c3d
Compare
|
You are right @pvary, I'm sorry - this is the same as #17078. The test I have removed the changes to v1.20/v2.0. I'll keep this PR open in case you prefer to merge this one instead of #17078 for any reason. |
|
Merged to main. Please create the backport PR. |
|
I think it's worth backporting the fix to the 1.11.x and perhaps 1.10.x branches. @pvary, do I need to create new PRs for this? |
I don't expect new releases in 1.10.x. |
|
Here is the backport PR #17255 to 1.11.x, all checks have passed. |
Avro timestamp-micros values were converted to TimestampData by multiplying the microsecond remainder by 1_000_000 instead of 1000, producing an out-of-range nanoOfMillisecond and throwing IllegalArgumentException. Tests didn't catch it because microsecond remainder was 0. I added a non-zero microsecond component to DataGenerators' shared timestamp fixture.
Closes #17075.