Support Utf8View for Avro by kumarlokesh · Pull Request #7434 · apache/arrow-rs

kumarlokesh · 2025-04-22T15:07:44Z

Which issue does this PR close?

Closes #7262.

Rationale for this change

What changes are included in this PR?

This PR adds support for Utf8View to the arrow-avro reader, enabling the creation of StringViewArray instead of regular StringArray for string data.

Added Utf8View variant to Codec enum and with_utf8view method for codec conversion
Updated Decoder enum with a StringView variant to handle StringViewArray creation
Added a configuration option use_utf8view to ReadOptions struct
Added benchmarks
Created an example application demonstrating how to use the feature

Benchmark output:

array_creation/string_array_10_chars                                                                            
                        time:   [9.5921 µs 9.6520 µs 9.7296 µs]
                        change: [-7.4725% -5.8906% -4.2024%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 2 outliers among 20 measurements (10.00%)
  1 (5.00%) high mild
  1 (5.00%) high severe
array_creation/string_view_10_chars                                                                            
                        time:   [10.053 µs 10.128 µs 10.207 µs]
                        change: [-3.1158% -0.9311% +1.0989%] (p = 0.43 > 0.05)
                        No change in performance detected.
Found 2 outliers among 20 measurements (10.00%)
  2 (10.00%) high mild
array_creation/string_array_100_chars                                                                            
                        time:   [16.242 µs 16.372 µs 16.576 µs]
                        change: [-1.3374% +0.4015% +2.4089%] (p = 0.71 > 0.05)
                        No change in performance detected.
Found 2 outliers among 20 measurements (10.00%)
  1 (5.00%) high mild
  1 (5.00%) high severe
array_creation/string_view_100_chars                                                                            
                        time:   [13.027 µs 13.290 µs 13.592 µs]
                        change: [-16.763% -12.893% -9.4265%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 2 outliers among 20 measurements (10.00%)
  1 (5.00%) high mild
  1 (5.00%) high severe
array_creation/string_array_1000_chars                                                                           
                        time:   [76.218 µs 77.540 µs 78.691 µs]
                        change: [-8.4455% -6.3045% -4.1053%] (p = 0.00 < 0.05)
                        Performance has improved.
array_creation/string_view_1000_chars                                                                            
                        time:   [45.964 µs 46.399 µs 46.770 µs]
                        change: [-8.7387% -3.9085% -0.5912%] (p = 0.07 > 0.05)
                        No change in performance detected.

string_operations/string_array_value_10_chars                                                                            
                        time:   [546.50 ns 565.27 ns 585.60 ns]
                        change: [-16.303% -10.503% -5.2608%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 1 outliers among 20 measurements (5.00%)
  1 (5.00%) high mild
string_operations/string_view_value_10_chars                                                                            
                        time:   [973.27 ns 992.09 ns 1.0101 µs]
                        change: [-14.527% -5.2942% +1.9214%] (p = 0.34 > 0.05)
                        No change in performance detected.
Found 3 outliers among 20 measurements (15.00%)
  1 (5.00%) low mild
  2 (10.00%) high mild
string_operations/string_array_value_100_chars                                                                            
                        time:   [548.03 ns 559.51 ns 573.23 ns]
                        change: [+1.1381% +3.0271% +5.0339%] (p = 0.01 < 0.05)
                        Performance has regressed.
string_operations/string_view_value_100_chars                                                                            
                        time:   [987.22 ns 1.0068 µs 1.0248 µs]
                        change: [-8.6431% +0.4588% +6.7220%] (p = 0.93 > 0.05)
                        No change in performance detected.
string_operations/string_array_value_1000_chars                                                                            
                        time:   [545.70 ns 556.69 ns 571.33 ns]
                        change: [-13.942% -8.4282% -3.7589%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 3 outliers among 20 measurements (15.00%)
  1 (5.00%) high mild
  2 (10.00%) high severe
string_operations/string_view_value_1000_chars                                                                            
                        time:   [947.71 ns 960.19 ns 974.77 ns]
                        change: [+0.9469% +4.6877% +8.4201%] (p = 0.01 < 0.05)
                        Change within noise threshold.
Found 1 outliers among 20 measurements (5.00%)
  1 (5.00%) high mild

avro_reader/string_array_10_chars                                                                           
                        time:   [149.00 µs 152.79 µs 155.90 µs]
                        change: [+2.5473% +5.4637% +8.5654%] (p = 0.00 < 0.05)
                        Performance has regressed.
avro_reader/string_view_10_chars                                                                           
                        time:   [141.75 µs 142.96 µs 144.64 µs]
                        change: [-14.307% -5.2067% +1.3228%] (p = 0.34 > 0.05)
                        No change in performance detected.
Found 1 outliers among 20 measurements (5.00%)
  1 (5.00%) high mild
avro_reader/string_array_100_chars                                                                           
                        time:   [173.20 µs 177.79 µs 183.34 µs]
                        change: [-4.6247% +1.1333% +5.8015%] (p = 0.72 > 0.05)
                        No change in performance detected.
avro_reader/string_view_100_chars                                                                           
                        time:   [174.64 µs 178.95 µs 183.48 µs]
                        change: [-10.183% -0.2004% +7.0467%] (p = 0.97 > 0.05)
                        No change in performance detected.
Found 1 outliers among 20 measurements (5.00%)
  1 (5.00%) high mild
avro_reader/string_array_1000_chars                                                                           
                        time:   [404.78 µs 415.51 µs 428.94 µs]
                        change: [-12.609% -2.9688% +4.7088%] (p = 0.64 > 0.05)
                        No change in performance detected.
avro_reader/string_view_1000_chars                                                                           
                        time:   [392.38 µs 403.15 µs 416.30 µs]
                        change: [+4.5005% +11.040% +18.788%] (p = 0.00 < 0.05)
                        Performance has regressed.
Found 2 outliers among 20 measurements (10.00%)
  1 (5.00%) high mild
  1 (5.00%) high severe

From the above benchmark results, we can draw following conclusions:

1. Array Creation Performance
1.1 Small Strings (10 chars):
StringArray: 9.65 µs
StringViewArray: 10.13 µs
StringArray is about 5% faster for small strings

1.2 Medium Strings (100 chars):
StringArray: 16.37 µs
StringViewArray: 13.29 µs
StringViewArray is about 19% faster

1.3 Large Strings (1000 chars):
StringArray: 77.54 µs
StringViewArray: 46.40 µs
StringViewArray is about 40% faster

2. String Value Access Performance
Value Access (all string sizes):
StringArray: ~560 ns
StringViewArray: ~980 ns
StringArray is consistently about 43% faster for value access

Are there any user-facing changes?

alamb

Thank you @kumarlokesh -- this looks really nice

To make reviewing this PR easier, is there any way you can break it into multiple smaller PRs? Specifically,

Improve the comments
Support for Map
Support for Utf8View?

Also it is not clear to me how much of a conflict this PR would have with:

#6965

alamb · 2025-04-28T17:52:44Z

arrow-avro/src/codec.rs

This is really nice -- thank you

alamb · 2025-04-28T17:53:45Z

arrow-avro/src/codec.rs

this seems to be a new feature (not just string view, but also map support)

Have you had a chance to review this PR?

Avro codec enhancements #6965 ?

alamb · 2025-04-28T17:55:19Z

arrow-avro/src/compression.rs

Thank you for these comments

kumarlokesh · 2025-04-28T20:22:50Z

@alamb moved improvements to comments to another PR #7449

kumarlokesh · 2025-05-18T19:22:50Z

@alamb updated this PR to remove "2. Support for Map". The sole focus of this PR is "3. Support for Utf8View".

In terms of scope of work between between #7434 and #6965, don't see a major overlap between what's intended. Therefore, I think the code conflict between the two should be minimal, and fairly trivial to resolve.

Thank you @kumarlokesh -- this looks really nice

To make reviewing this PR easier, is there any way you can break it into multiple smaller PRs? Specifically,

Improve the comments

Support for Map

Support for Utf8View?

Also it is not clear to me how much of a conflict this PR would have with:

Avro codec enhancements #6965

alamb

Thank you @kumarlokesh -- I think this looks really nicely documented and tested.

I had a few API questions / suggestions but I don't think they are necessary.

It seems to me that this PR does not change any APIs (only adds new ones) so I think we could merge it before the next major release. What do you think?

Thanks again for pushing this along and breaking it up into smaller chunks

alamb · 2025-05-19T20:17:01Z

arrow-avro/src/codec.rs

+    }
+
+    #[test]
+    fn test_string_with_utf8view_enabled() {


alamb · 2025-05-19T20:17:21Z

arrow-avro/src/lib.rs

 ///
 /// Provides types for parsing and representing Avro schema definitions.
-mod schema;
+pub mod schema;


is it needed to make these module public?

alamb · 2025-05-19T20:18:06Z

arrow-avro/src/reader/record.rs

+    /// * `use_utf8view` - If true, use StringViewArray instead of StringArray for string data
+    pub fn try_new_with_options(
+        data_type: &AvroDataType,
+        use_utf8view: bool,


Shall we use ReadOptions here rather than the single field? That would make updating these easier in the future

alamb · 2025-05-19T20:18:35Z

arrow-avro/src/reader/mod.rs

+    /// into Arrow's StringViewArray instead of the standard StringArray.
+    ///
+    /// Default: false
+    pub use_utf8view: bool,


I think it would be nice if this field was not pub -- that way we could add new options to ReadOptions in the future without it being a breaking API change

alamb · 2025-05-19T20:34:18Z

Marking as next-major-release as a precaution to avoid merging until we convince ourselves this PR has no breaking changes

kumarlokesh · 2025-05-20T05:44:50Z

Thank you @kumarlokesh -- I think this looks really nicely documented and tested.

I had a few API questions / suggestions but I don't think they are necessary.

It seems to me that this PR does not change any APIs (only adds new ones) so I think we could merge it before the next major release. What do you think?

Thanks again for pushing this along and breaking it up into smaller chunks

@alamb thanks for the review! Yes, I think this PR could be merged before the next major release.

Addressed review comments here cf50b5b

alamb · 2025-05-21T11:13:04Z

arrow-avro/benches/avro_reader.rs

+
+    let mut file = temp_file.reopen()?;
+
+    file.write_all(b"AVRO")?;


I wondered why this benchmark isn't using a avro writer and then I realized it is because there is no writer in avro-arrow 👍

alamb · 2025-05-21T11:14:51Z

arrow-avro/src/codec.rs

    Binary,
    /// String data represented as UTF-8 encoded bytes, corresponding to Arrow's StringArray
    Utf8,
+    /// String data represented as UTF-8 encoded bytes with an optimized view representation,


This is not pub it seems: https://docs.rs/arrow-avro/latest/arrow_avro/?search=Codec

Though now I look it seems there is nothing that is pub: https://docs.rs/arrow-avro/latest/arrow_avro/all.html 🤔

alamb · 2025-05-21T11:16:08Z

Let's keep thing moving and merge this PR.

@tustvold do you remember what the plan / status of the arrow-avro crate is? Did we mark it as experimental or something so we could change the APIs between major releases?

alamb · 2025-05-21T11:16:22Z

Thanks again @kumarlokesh

github-actions bot added the arrow Changes to the arrow crate label Apr 22, 2025

kumarlokesh mentioned this pull request Apr 22, 2025

Support Utf8View for avro #7262

Closed

alamb reviewed Apr 28, 2025

View reviewed changes

alamb mentioned this pull request Apr 28, 2025

Weekly Plan: Andrew Lamb 2025-04-28 apache/datafusion#15880

Closed

26 tasks

kumarlokesh mentioned this pull request Apr 28, 2025

Improve comments for avro #7449

Merged

kumarlokesh marked this pull request as draft April 28, 2025 20:17

kumarlokesh force-pushed the support-utf8-view-for-avro branch from 91a31ce to 78f8c5b Compare April 29, 2025 08:57

This was referenced May 5, 2025

Weekly Plan: Andrew Lamb 2025-05-05 apache/datafusion#15943

Closed

Weekly Plan: Andrew Lamb 2025-05-12 apache/datafusion#16022

Closed

kumarlokesh force-pushed the support-utf8-view-for-avro branch from 78f8c5b to 13d05d5 Compare May 18, 2025 16:22

Support Utf8View for Avro

ae4fe0b

kumarlokesh force-pushed the support-utf8-view-for-avro branch from 13d05d5 to ae4fe0b Compare May 18, 2025 16:27

kumarlokesh marked this pull request as ready for review May 18, 2025 16:49

kumarlokesh requested a review from alamb May 18, 2025 19:22

alamb mentioned this pull request May 19, 2025

Weekly Plan: Andrew Lamb 2025-05-19 apache/datafusion#16101

Closed

18 tasks

alamb approved these changes May 19, 2025

View reviewed changes

alamb added the next-major-release the PR has API changes and it waiting on the next major version label May 19, 2025

code review improvements

cf50b5b

alamb removed the next-major-release the PR has API changes and it waiting on the next major version label May 21, 2025

alamb approved these changes May 21, 2025

View reviewed changes

alamb reviewed May 21, 2025

View reviewed changes

alamb merged commit 36b2a27 into apache:main May 21, 2025
24 checks passed

kumarlokesh deleted the support-utf8-view-for-avro branch May 21, 2025 14:22

alamb mentioned this pull request May 23, 2025

Fix CI on main due to logical conflict #7542

Merged


		let mut file = temp_file.reopen()?;

		file.write_all(b"AVRO")?;

Conversation

kumarlokesh commented Apr 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Which issue does this PR close?

Rationale for this change

What changes are included in this PR?

Are there any user-facing changes?

Uh oh!

alamb left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kumarlokesh commented Apr 28, 2025

Uh oh!

kumarlokesh commented May 18, 2025

Uh oh!

alamb left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

alamb commented May 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kumarlokesh commented May 20, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

alamb commented May 21, 2025

Uh oh!

Uh oh!

alamb commented May 21, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

kumarlokesh commented Apr 22, 2025 •

edited

Loading

alamb commented May 19, 2025 •

edited

Loading