[Data] Use iterator in write ops instead of accumulating all of the blocks in memory instead#57108
Merged
alexeykudinkin merged 5 commits intoray-project:masterfrom Oct 2, 2025
Merged
Conversation
…n mem instead Signed-off-by: Alexey Kudinkin <ak@anyscale.com>
Contributor
There was a problem hiding this comment.
Code Review
This pull request aims to optimize memory usage in write_datasink by using an iterator. While the change correctly introduces an iterator, it doesn't fully achieve the memory-saving goal because it still accumulates all write results in memory before processing. I've provided a review comment with a suggested code change to process the results incrementally, which should properly address the underlying memory issue.
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
Signed-off-by: Alexey Kudinkin <ak@anyscale.com>
Signed-off-by: Alexey Kudinkin <ak@anyscale.com>
Signed-off-by: Alexey Kudinkin <ak@anyscale.com>
Contributor
|
/gemini review |
Contributor
There was a problem hiding this comment.
Code Review
This pull request refactors the write operations to use an iterator-based approach for handling write results, which is a great improvement for memory efficiency. Instead of accumulating all block references and getting their results at once with ray.get(), the new implementation processes them incrementally. This avoids holding all results in memory, which is particularly beneficial for large datasets.
The changes include:
- Modifying
Dataset.write_datasinkto use_execute_to_iteratorand process write results in a streaming fashion. - Introducing a
WriteResult.combineclass method to aggregate per-bundle write results. - Relocating the
gen_datasink_write_resulthelper function to a more appropriate module.
The implementation is solid. I've suggested a minor enhancement to use a generator expression, which further improves memory usage by avoiding an intermediate list of results. Overall, this is a valuable optimization.
Comment on lines
+4928
to
+4936
| iter_, stats = self._write_ds._execute_to_iterator() | ||
| write_results = [] | ||
|
|
||
| for bundle in iter_: | ||
| res = ray.get(bundle.block_refs) | ||
| # Generate write result report | ||
| write_results.append(_gen_datasink_write_result(res)) | ||
|
|
||
| combined_write_result = WriteResult.combine(*write_results) |
Contributor
There was a problem hiding this comment.
Great work on switching to an iterator to process write results incrementally! This significantly improves memory efficiency.
To further enhance this and make it even more memory-friendly, you can use a generator expression instead of creating an intermediate write_results list. This avoids materializing all WriteResult objects in memory before combining them.
Also, the stats variable returned by _execute_to_iterator() is not used, so it can be replaced with _ to signal that it's intentionally ignored.
iter_, _ = self._write_ds._execute_to_iterator()
write_results_iter = (
_gen_datasink_write_result(ray.get(bundle.block_refs))
for bundle in iter_
)
combined_write_result = WriteResult.combine(*write_results_iter)
goutamvenkat-anyscale
approved these changes
Oct 2, 2025
eicherseiji
pushed a commit
to eicherseiji/ray
that referenced
this pull request
Oct 6, 2025
…locks in memory instead (ray-project#57108) <!-- Thank you for your contribution! Please review https://github.com/ray-project/ray/blob/master/CONTRIBUTING.rst before opening a pull request. --> <!-- Please add a reviewer to the assignee section when you create a PR. If you don't have the access to it, we will shortly find a reviewer and assign them to your PR. --> ## Changes Execute as a typical iteration writing and releasing resulting blocks incrementally rather than holding on to them until the whole op completes. <!-- Please give a short summary of the change and the problem this solves. --> ## Related issue number <!-- For example: "Closes ray-project#1234" --> ## Checks - [ ] I've signed off every commit(by using the -s flag, i.e., `git commit -s`) in this PR. - [ ] I've run pre-commit jobs to lint the changes in this PR. ([pre-commit setup](https://docs.ray.io/en/latest/ray-contribute/getting-involved.html#lint-and-formatting)) - [ ] I've included any doc changes needed for https://docs.ray.io/en/master/. - [ ] I've added any new APIs to the API Reference. For example, if I added a method in Tune, I've added it in `doc/source/tune/api/` under the corresponding `.rst` file. - [ ] I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/ - Testing Strategy - [ ] Unit tests - [ ] Release tests - [ ] This PR is not tested :( <!-- CURSOR_SUMMARY --> --- > [!NOTE] > Write ops now consume results via an iterator and aggregate per-bundle WriteResult using a new combine API, avoiding accumulating all blocks in memory. > > - **Write Path (Dataset.write_datasink)**: > - Switch to streaming execution (`_execute_to_iterator`) to process write results incrementally instead of `ray.get` on all blocks. > - Accumulate per-bundle results via `_gen_datasink_write_result` and aggregate with `WriteResult.combine`; log and callback use the combined result. > - **APIs/Utilities**: > - Move write-result aggregation helper to `ray.data.datasource.datasink` as `_gen_datasink_write_result`. > - Add `WriteResult.combine(...)` for aggregating multiple write results. > - **Code Cleanup**: > - Remove `gen_datasink_write_result` from `planner/plan_write_op.py` and related imports; adjust imports in `dataset.py` accordingly. > > <sup>Written by [Cursor Bugbot](https://cursor.com/dashboard?tab=bugbot) for commit 4c5a373. This will update automatically on new commits. Configure [here](https://cursor.com/dashboard?tab=bugbot).</sup> <!-- /CURSOR_SUMMARY --> --------- Signed-off-by: Alexey Kudinkin <ak@anyscale.com> Signed-off-by: Seiji Eicher <seiji@anyscale.com>
dstrodtman
pushed a commit
to dstrodtman/ray
that referenced
this pull request
Oct 6, 2025
…locks in memory instead (ray-project#57108) <!-- Thank you for your contribution! Please review https://github.com/ray-project/ray/blob/master/CONTRIBUTING.rst before opening a pull request. --> <!-- Please add a reviewer to the assignee section when you create a PR. If you don't have the access to it, we will shortly find a reviewer and assign them to your PR. --> ## Changes Execute as a typical iteration writing and releasing resulting blocks incrementally rather than holding on to them until the whole op completes. <!-- Please give a short summary of the change and the problem this solves. --> ## Related issue number <!-- For example: "Closes ray-project#1234" --> ## Checks - [ ] I've signed off every commit(by using the -s flag, i.e., `git commit -s`) in this PR. - [ ] I've run pre-commit jobs to lint the changes in this PR. ([pre-commit setup](https://docs.ray.io/en/latest/ray-contribute/getting-involved.html#lint-and-formatting)) - [ ] I've included any doc changes needed for https://docs.ray.io/en/master/. - [ ] I've added any new APIs to the API Reference. For example, if I added a method in Tune, I've added it in `doc/source/tune/api/` under the corresponding `.rst` file. - [ ] I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/ - Testing Strategy - [ ] Unit tests - [ ] Release tests - [ ] This PR is not tested :( <!-- CURSOR_SUMMARY --> --- > [!NOTE] > Write ops now consume results via an iterator and aggregate per-bundle WriteResult using a new combine API, avoiding accumulating all blocks in memory. > > - **Write Path (Dataset.write_datasink)**: > - Switch to streaming execution (`_execute_to_iterator`) to process write results incrementally instead of `ray.get` on all blocks. > - Accumulate per-bundle results via `_gen_datasink_write_result` and aggregate with `WriteResult.combine`; log and callback use the combined result. > - **APIs/Utilities**: > - Move write-result aggregation helper to `ray.data.datasource.datasink` as `_gen_datasink_write_result`. > - Add `WriteResult.combine(...)` for aggregating multiple write results. > - **Code Cleanup**: > - Remove `gen_datasink_write_result` from `planner/plan_write_op.py` and related imports; adjust imports in `dataset.py` accordingly. > > <sup>Written by [Cursor Bugbot](https://cursor.com/dashboard?tab=bugbot) for commit 4c5a373. This will update automatically on new commits. Configure [here](https://cursor.com/dashboard?tab=bugbot).</sup> <!-- /CURSOR_SUMMARY --> --------- Signed-off-by: Alexey Kudinkin <ak@anyscale.com> Signed-off-by: Douglas Strodtman <douglas@anyscale.com>
liulehui
pushed a commit
to liulehui/ray
that referenced
this pull request
Oct 9, 2025
…locks in memory instead (ray-project#57108) <!-- Thank you for your contribution! Please review https://github.com/ray-project/ray/blob/master/CONTRIBUTING.rst before opening a pull request. --> <!-- Please add a reviewer to the assignee section when you create a PR. If you don't have the access to it, we will shortly find a reviewer and assign them to your PR. --> ## Changes Execute as a typical iteration writing and releasing resulting blocks incrementally rather than holding on to them until the whole op completes. <!-- Please give a short summary of the change and the problem this solves. --> ## Related issue number <!-- For example: "Closes ray-project#1234" --> ## Checks - [ ] I've signed off every commit(by using the -s flag, i.e., `git commit -s`) in this PR. - [ ] I've run pre-commit jobs to lint the changes in this PR. ([pre-commit setup](https://docs.ray.io/en/latest/ray-contribute/getting-involved.html#lint-and-formatting)) - [ ] I've included any doc changes needed for https://docs.ray.io/en/master/. - [ ] I've added any new APIs to the API Reference. For example, if I added a method in Tune, I've added it in `doc/source/tune/api/` under the corresponding `.rst` file. - [ ] I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/ - Testing Strategy - [ ] Unit tests - [ ] Release tests - [ ] This PR is not tested :( <!-- CURSOR_SUMMARY --> --- > [!NOTE] > Write ops now consume results via an iterator and aggregate per-bundle WriteResult using a new combine API, avoiding accumulating all blocks in memory. > > - **Write Path (Dataset.write_datasink)**: > - Switch to streaming execution (`_execute_to_iterator`) to process write results incrementally instead of `ray.get` on all blocks. > - Accumulate per-bundle results via `_gen_datasink_write_result` and aggregate with `WriteResult.combine`; log and callback use the combined result. > - **APIs/Utilities**: > - Move write-result aggregation helper to `ray.data.datasource.datasink` as `_gen_datasink_write_result`. > - Add `WriteResult.combine(...)` for aggregating multiple write results. > - **Code Cleanup**: > - Remove `gen_datasink_write_result` from `planner/plan_write_op.py` and related imports; adjust imports in `dataset.py` accordingly. > > <sup>Written by [Cursor Bugbot](https://cursor.com/dashboard?tab=bugbot) for commit 4c5a373. This will update automatically on new commits. Configure [here](https://cursor.com/dashboard?tab=bugbot).</sup> <!-- /CURSOR_SUMMARY --> --------- Signed-off-by: Alexey Kudinkin <ak@anyscale.com>
joshkodi
pushed a commit
to joshkodi/ray
that referenced
this pull request
Oct 13, 2025
…locks in memory instead (ray-project#57108) <!-- Thank you for your contribution! Please review https://github.com/ray-project/ray/blob/master/CONTRIBUTING.rst before opening a pull request. --> <!-- Please add a reviewer to the assignee section when you create a PR. If you don't have the access to it, we will shortly find a reviewer and assign them to your PR. --> ## Changes Execute as a typical iteration writing and releasing resulting blocks incrementally rather than holding on to them until the whole op completes. <!-- Please give a short summary of the change and the problem this solves. --> ## Related issue number <!-- For example: "Closes ray-project#1234" --> ## Checks - [ ] I've signed off every commit(by using the -s flag, i.e., `git commit -s`) in this PR. - [ ] I've run pre-commit jobs to lint the changes in this PR. ([pre-commit setup](https://docs.ray.io/en/latest/ray-contribute/getting-involved.html#lint-and-formatting)) - [ ] I've included any doc changes needed for https://docs.ray.io/en/master/. - [ ] I've added any new APIs to the API Reference. For example, if I added a method in Tune, I've added it in `doc/source/tune/api/` under the corresponding `.rst` file. - [ ] I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/ - Testing Strategy - [ ] Unit tests - [ ] Release tests - [ ] This PR is not tested :( <!-- CURSOR_SUMMARY --> --- > [!NOTE] > Write ops now consume results via an iterator and aggregate per-bundle WriteResult using a new combine API, avoiding accumulating all blocks in memory. > > - **Write Path (Dataset.write_datasink)**: > - Switch to streaming execution (`_execute_to_iterator`) to process write results incrementally instead of `ray.get` on all blocks. > - Accumulate per-bundle results via `_gen_datasink_write_result` and aggregate with `WriteResult.combine`; log and callback use the combined result. > - **APIs/Utilities**: > - Move write-result aggregation helper to `ray.data.datasource.datasink` as `_gen_datasink_write_result`. > - Add `WriteResult.combine(...)` for aggregating multiple write results. > - **Code Cleanup**: > - Remove `gen_datasink_write_result` from `planner/plan_write_op.py` and related imports; adjust imports in `dataset.py` accordingly. > > <sup>Written by [Cursor Bugbot](https://cursor.com/dashboard?tab=bugbot) for commit 4c5a373. This will update automatically on new commits. Configure [here](https://cursor.com/dashboard?tab=bugbot).</sup> <!-- /CURSOR_SUMMARY --> --------- Signed-off-by: Alexey Kudinkin <ak@anyscale.com> Signed-off-by: Josh Kodi <joshkodi@gmail.com>
justinyeh1995
pushed a commit
to justinyeh1995/ray
that referenced
this pull request
Oct 20, 2025
…locks in memory instead (ray-project#57108) <!-- Thank you for your contribution! Please review https://github.com/ray-project/ray/blob/master/CONTRIBUTING.rst before opening a pull request. --> <!-- Please add a reviewer to the assignee section when you create a PR. If you don't have the access to it, we will shortly find a reviewer and assign them to your PR. --> ## Changes Execute as a typical iteration writing and releasing resulting blocks incrementally rather than holding on to them until the whole op completes. <!-- Please give a short summary of the change and the problem this solves. --> ## Related issue number <!-- For example: "Closes ray-project#1234" --> ## Checks - [ ] I've signed off every commit(by using the -s flag, i.e., `git commit -s`) in this PR. - [ ] I've run pre-commit jobs to lint the changes in this PR. ([pre-commit setup](https://docs.ray.io/en/latest/ray-contribute/getting-involved.html#lint-and-formatting)) - [ ] I've included any doc changes needed for https://docs.ray.io/en/master/. - [ ] I've added any new APIs to the API Reference. For example, if I added a method in Tune, I've added it in `doc/source/tune/api/` under the corresponding `.rst` file. - [ ] I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/ - Testing Strategy - [ ] Unit tests - [ ] Release tests - [ ] This PR is not tested :( <!-- CURSOR_SUMMARY --> --- > [!NOTE] > Write ops now consume results via an iterator and aggregate per-bundle WriteResult using a new combine API, avoiding accumulating all blocks in memory. > > - **Write Path (Dataset.write_datasink)**: > - Switch to streaming execution (`_execute_to_iterator`) to process write results incrementally instead of `ray.get` on all blocks. > - Accumulate per-bundle results via `_gen_datasink_write_result` and aggregate with `WriteResult.combine`; log and callback use the combined result. > - **APIs/Utilities**: > - Move write-result aggregation helper to `ray.data.datasource.datasink` as `_gen_datasink_write_result`. > - Add `WriteResult.combine(...)` for aggregating multiple write results. > - **Code Cleanup**: > - Remove `gen_datasink_write_result` from `planner/plan_write_op.py` and related imports; adjust imports in `dataset.py` accordingly. > > <sup>Written by [Cursor Bugbot](https://cursor.com/dashboard?tab=bugbot) for commit 4c5a373. This will update automatically on new commits. Configure [here](https://cursor.com/dashboard?tab=bugbot).</sup> <!-- /CURSOR_SUMMARY --> --------- Signed-off-by: Alexey Kudinkin <ak@anyscale.com>
landscapepainter
pushed a commit
to landscapepainter/ray
that referenced
this pull request
Nov 17, 2025
…locks in memory instead (ray-project#57108) <!-- Thank you for your contribution! Please review https://github.com/ray-project/ray/blob/master/CONTRIBUTING.rst before opening a pull request. --> <!-- Please add a reviewer to the assignee section when you create a PR. If you don't have the access to it, we will shortly find a reviewer and assign them to your PR. --> ## Changes Execute as a typical iteration writing and releasing resulting blocks incrementally rather than holding on to them until the whole op completes. <!-- Please give a short summary of the change and the problem this solves. --> ## Related issue number <!-- For example: "Closes ray-project#1234" --> ## Checks - [ ] I've signed off every commit(by using the -s flag, i.e., `git commit -s`) in this PR. - [ ] I've run pre-commit jobs to lint the changes in this PR. ([pre-commit setup](https://docs.ray.io/en/latest/ray-contribute/getting-involved.html#lint-and-formatting)) - [ ] I've included any doc changes needed for https://docs.ray.io/en/master/. - [ ] I've added any new APIs to the API Reference. For example, if I added a method in Tune, I've added it in `doc/source/tune/api/` under the corresponding `.rst` file. - [ ] I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/ - Testing Strategy - [ ] Unit tests - [ ] Release tests - [ ] This PR is not tested :( <!-- CURSOR_SUMMARY --> --- > [!NOTE] > Write ops now consume results via an iterator and aggregate per-bundle WriteResult using a new combine API, avoiding accumulating all blocks in memory. > > - **Write Path (Dataset.write_datasink)**: > - Switch to streaming execution (`_execute_to_iterator`) to process write results incrementally instead of `ray.get` on all blocks. > - Accumulate per-bundle results via `_gen_datasink_write_result` and aggregate with `WriteResult.combine`; log and callback use the combined result. > - **APIs/Utilities**: > - Move write-result aggregation helper to `ray.data.datasource.datasink` as `_gen_datasink_write_result`. > - Add `WriteResult.combine(...)` for aggregating multiple write results. > - **Code Cleanup**: > - Remove `gen_datasink_write_result` from `planner/plan_write_op.py` and related imports; adjust imports in `dataset.py` accordingly. > > <sup>Written by [Cursor Bugbot](https://cursor.com/dashboard?tab=bugbot) for commit 4c5a373. This will update automatically on new commits. Configure [here](https://cursor.com/dashboard?tab=bugbot).</sup> <!-- /CURSOR_SUMMARY --> --------- Signed-off-by: Alexey Kudinkin <ak@anyscale.com>
Aydin-ab
pushed a commit
to Aydin-ab/ray-aydin
that referenced
this pull request
Nov 19, 2025
…locks in memory instead (ray-project#57108) <!-- Thank you for your contribution! Please review https://github.com/ray-project/ray/blob/master/CONTRIBUTING.rst before opening a pull request. --> <!-- Please add a reviewer to the assignee section when you create a PR. If you don't have the access to it, we will shortly find a reviewer and assign them to your PR. --> ## Changes Execute as a typical iteration writing and releasing resulting blocks incrementally rather than holding on to them until the whole op completes. <!-- Please give a short summary of the change and the problem this solves. --> ## Related issue number <!-- For example: "Closes ray-project#1234" --> ## Checks - [ ] I've signed off every commit(by using the -s flag, i.e., `git commit -s`) in this PR. - [ ] I've run pre-commit jobs to lint the changes in this PR. ([pre-commit setup](https://docs.ray.io/en/latest/ray-contribute/getting-involved.html#lint-and-formatting)) - [ ] I've included any doc changes needed for https://docs.ray.io/en/master/. - [ ] I've added any new APIs to the API Reference. For example, if I added a method in Tune, I've added it in `doc/source/tune/api/` under the corresponding `.rst` file. - [ ] I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/ - Testing Strategy - [ ] Unit tests - [ ] Release tests - [ ] This PR is not tested :( <!-- CURSOR_SUMMARY --> --- > [!NOTE] > Write ops now consume results via an iterator and aggregate per-bundle WriteResult using a new combine API, avoiding accumulating all blocks in memory. > > - **Write Path (Dataset.write_datasink)**: > - Switch to streaming execution (`_execute_to_iterator`) to process write results incrementally instead of `ray.get` on all blocks. > - Accumulate per-bundle results via `_gen_datasink_write_result` and aggregate with `WriteResult.combine`; log and callback use the combined result. > - **APIs/Utilities**: > - Move write-result aggregation helper to `ray.data.datasource.datasink` as `_gen_datasink_write_result`. > - Add `WriteResult.combine(...)` for aggregating multiple write results. > - **Code Cleanup**: > - Remove `gen_datasink_write_result` from `planner/plan_write_op.py` and related imports; adjust imports in `dataset.py` accordingly. > > <sup>Written by [Cursor Bugbot](https://cursor.com/dashboard?tab=bugbot) for commit 4c5a373. This will update automatically on new commits. Configure [here](https://cursor.com/dashboard?tab=bugbot).</sup> <!-- /CURSOR_SUMMARY --> --------- Signed-off-by: Alexey Kudinkin <ak@anyscale.com> Signed-off-by: Aydin Abiar <aydin@anyscale.com>
Future-Outlier
pushed a commit
to Future-Outlier/ray
that referenced
this pull request
Dec 7, 2025
…locks in memory instead (ray-project#57108) <!-- Thank you for your contribution! Please review https://github.com/ray-project/ray/blob/master/CONTRIBUTING.rst before opening a pull request. --> <!-- Please add a reviewer to the assignee section when you create a PR. If you don't have the access to it, we will shortly find a reviewer and assign them to your PR. --> ## Changes Execute as a typical iteration writing and releasing resulting blocks incrementally rather than holding on to them until the whole op completes. <!-- Please give a short summary of the change and the problem this solves. --> ## Related issue number <!-- For example: "Closes ray-project#1234" --> ## Checks - [ ] I've signed off every commit(by using the -s flag, i.e., `git commit -s`) in this PR. - [ ] I've run pre-commit jobs to lint the changes in this PR. ([pre-commit setup](https://docs.ray.io/en/latest/ray-contribute/getting-involved.html#lint-and-formatting)) - [ ] I've included any doc changes needed for https://docs.ray.io/en/master/. - [ ] I've added any new APIs to the API Reference. For example, if I added a method in Tune, I've added it in `doc/source/tune/api/` under the corresponding `.rst` file. - [ ] I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/ - Testing Strategy - [ ] Unit tests - [ ] Release tests - [ ] This PR is not tested :( <!-- CURSOR_SUMMARY --> --- > [!NOTE] > Write ops now consume results via an iterator and aggregate per-bundle WriteResult using a new combine API, avoiding accumulating all blocks in memory. > > - **Write Path (Dataset.write_datasink)**: > - Switch to streaming execution (`_execute_to_iterator`) to process write results incrementally instead of `ray.get` on all blocks. > - Accumulate per-bundle results via `_gen_datasink_write_result` and aggregate with `WriteResult.combine`; log and callback use the combined result. > - **APIs/Utilities**: > - Move write-result aggregation helper to `ray.data.datasource.datasink` as `_gen_datasink_write_result`. > - Add `WriteResult.combine(...)` for aggregating multiple write results. > - **Code Cleanup**: > - Remove `gen_datasink_write_result` from `planner/plan_write_op.py` and related imports; adjust imports in `dataset.py` accordingly. > > <sup>Written by [Cursor Bugbot](https://cursor.com/dashboard?tab=bugbot) for commit 4c5a373. This will update automatically on new commits. Configure [here](https://cursor.com/dashboard?tab=bugbot).</sup> <!-- /CURSOR_SUMMARY --> --------- Signed-off-by: Alexey Kudinkin <ak@anyscale.com> Signed-off-by: Future-Outlier <eric901201@gmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Changes
Execute as a typical iteration writing and releasing resulting blocks incrementally rather than holding on to them until the whole op completes.
Related issue number
Checks
git commit -s) in this PR.method in Tune, I've added it in
doc/source/tune/api/under thecorresponding
.rstfile.Note
Write ops now consume results via an iterator and aggregate per-bundle WriteResult using a new combine API, avoiding accumulating all blocks in memory.
_execute_to_iterator) to process write results incrementally instead ofray.geton all blocks._gen_datasink_write_resultand aggregate withWriteResult.combine; log and callback use the combined result.ray.data.datasource.datasinkas_gen_datasink_write_result.WriteResult.combine(...)for aggregating multiple write results.gen_datasink_write_resultfromplanner/plan_write_op.pyand related imports; adjust imports indataset.pyaccordingly.Written by Cursor Bugbot for commit 4c5a373. This will update automatically on new commits. Configure here.