Use Iceberg-Rust for parsing the ManifestList and Manifests #2004
Conversation
Fokko
changed the title
|
@roeap Thanks, your patch fixed correctly passing through the |
There was a problem hiding this comment.
looks like we're waiting for apache/iceberg-rust#1328 to finalize, which is dependent on apache/iceberg-rust#1482 :)
tests/utils/test_manifest.py
Outdated
| @@ -357,7 +357,7 @@ def test_write_empty_manifest() -> None: | |||
|
|
|||
| @pytest.mark.parametrize("format_version", [1, 2]) | |||
| def test_write_manifest( | |||
| generated_manifest_file_file_v1: str, generated_manifest_file_file_v2: str, format_version: TableVersion | |||
| generated_manifest_file_file_v1: str, generated_manifest_file_file_v2: str, format_version: TableVersion, test_schema: Schema | |||
There was a problem hiding this comment.
nit use test_partition_spec too
|
@Fokko This is awesome! thanks for your work on this one. |
|
@yogevyuval Thanks for asking, and yes, I do expect performance impact since just a part of the deserialization is cythonized. With this change, much more is pushed into Rust. This will also reduce the GIL pressure since we don't have to build the readers anymore :) I left this PR small to focus on the essentials, but once this is in, we can also clean up a LOT of code 👍 |
Interesting, we have quite a bit of concurrency going on so curious to see the results. Once apache/iceberg-rust#1328 gets merged and release i'll try and create some basic benchmarks and share the results |
While working on apache#2004 I've noticed some small discrepancies that I think would be good to address in a separate PR.
While working on apache#2004 I've noticed some small discrepancies that I think would be good to address in a separate PR.
Rationale for this change
This replaces the Cython implementation for reading Avro with Iceberg-Rust. This would greatly simplify the PyIceberg project since we don't have to publish Python wheels anymore.
Are these changes tested?
Are there any user-facing changes?