JoshRosen opened a new pull request #25745: [SPARK-29033][SQL][WIP URL: https://github.com/apache/spark/pull/25745 <!-- Thanks for sending a pull request! Here are some tips for you: 1. If this is your first time, please read our contributor guidelines: https://spark.apache.org/contributing.html 2. Ensure you have added or run the appropriate tests for your PR: https://spark.apache.org/developer-tools.html 3. If the PR is unfinished, add '[WIP]' in your PR title, e.g., '[WIP][SPARK-XXXX] Your PR title ...'. 4. Be sure to keep the PR description updated to reflect all changes. 5. Please write your PR title to summarize what this PR proposes. 6. If possible, provide a concise example to reproduce the issue for a faster review. --> ### What changes were proposed in this pull request? <!-- Please clarify what changes you are proposing. The purpose of this section is to outline the changes and how this PR fixes the issue. If possible, please consider writing useful notes for better and faster reviews in your PR. See the examples below. 1. If you refactor some codes with changing classes, showing the class hierarchy will help reviewers. 2. If you fix some SQL features, you can provide some references of other DBMSes. 3. If there is design documentation, please add the link. 4. If there is a discussion in the mailing list, please add the link. --> Spark 2.x has two separate implementations of the "create named struct" expression: regular `CreateNamedStruct` and `CreateNamedStructUnsafe`. The former produces `GenericInternalRows`, while the latter produces `UnsafeRow`s. Both two expressions both extend the `CreateNameStructLike` trait. The "unsafe" version was added in SPARK-9373 / #7689 to support structs in `GenerateUnsafeProjection` (this was fairly early in the Tungsten effort, circa mid-2015 / Spark 1.5.x). This PR changes Spark so the UnsafeRow-based codepath is always used and removes the GenericRow-based path. For ease-of-review, I've broken this into two commits: - The first commit changes `CreateNamedStruct` to use the `CreateNamedStructUnsafe` implementation (producing `UnsafeRow`) and deletes `CreateNamedStructUnsafe`. - The second commit removes the `CreateNameStructLike` trait (since at that point it only has a single implementation). ### Why are the changes needed? <!-- Please clarify why the changes are needed. For instance, 1. If you propose a new API, clarify the use case for a new API. 2. If you fix a bug, you can clarify why it is a bug. --> The old `CreateNamedStruct` code path allocated a fresh `GenericInternalRow` on every invocation, incurring object allocation and primitive-boxing performance overheads. I suspect that this can be a significant performance problem in code which uses `Dataset`s: Spark's `ExpressionEncoder` uses `CreateNamedStruct` for serializing / deserializing case classes, so the `GenericInternalRow` performance problems can impact typed Dataset operations (e.g. `.map()`). ### Does this PR introduce any user-facing change? <!-- If yes, please clarify the previous behavior and the change this PR proposes - provide the console output, description and/or an example to show the behavior difference if possible. If no, write 'No'. --> There is no expected user-facing behavioral change. However, there **is** an binary-compatibility-breaking change due to the deletion of `CreateNamedStructUnsafe` and `CreateNamedStructLike`. I suspect that these were originally intended to be internal / private classes and therefore propose that this is an acceptable breaking change in Spark 3.x (since it seems pretty unlikely that Spark users would be directly using those internal interfaces). I think there's also code-simplification benefits: if we only have a single implementation of `CreateNamedStruct` then we're removing the possibility for bugs in case code authors pattern-match on concrete classes instead of the `CreateNamedStructLike` trait. ### How was this patch tested? <!-- If tests were added, say they were added here. Please make sure to add some test cases that check the changes thoroughly including negative and positive cases if possible. If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future. If tests were not added, please describe why they were not added and/or why it was difficult to add. --> I believe this is covered by Spark's existing tests. `CreateNamedStructUnsafe` is already used in `GenerateUnsafeProjection` and thus already had some pre-existing test coverage. ⚠️ I've marked this PR as `WIP` pending some benchmarking and time to think about possible corner-cases. #### Performance benchmark Here's a (somewhat hacky) end-to-end performance benchmark exercising this codepath in `spark-shell`: ```scala val start = System.currentTimeMillis() spark.sql("select struct(id, id, id, id, id, id, id, id, id) from range(1000 * 1000 * 50)").rdd.count val end = System.currentTimeMillis() println(end - start) println ``` On my laptop, this takes ~21 seconds after.
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
