You are viewing a plain text version of this content. The canonical link for it is here.
Posted to reviews@spark.apache.org by "hui730 (via GitHub)" <gi...@apache.org> on 2023/11/01 13:47:54 UTC
[PR] Keep the taskBinary in Stage to avoid deserializing it multiple times, and send binary to executor instead of task. [spark]
hui730 opened a new pull request, #43621:
URL: https://github.com/apache/spark/pull/43621
<!--
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.
7. If you want to add a new configuration, please read the guideline first for naming configurations in
'core/src/main/scala/org/apache/spark/internal/config/ConfigEntry.scala'.
8. If you want to add or modify an error type or message, please read the guideline first in
'core/src/main/resources/error/README.md'.
-->
### 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.
-->
My modification is a performance optimization, which keep the taskBinary in Stage to avoid deserializing it multiple times, and send binary to executor instead of task. Here are specific modifications：
a. Add stageID info to TaskDescription.
b. Add taskBinary info to taskSet. Remove taskBinary info to ShuffleMapTask and ResultTask.
c. add binary getter/setter of Task.
d. Store binary broadcast when submitTasks.
e. Add new Netty event message LaunchBinary, before executor run task, launch binary to new executor first.
### 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 taskBinary for each task of the same stage is the same. The current situation is that each task will bring a copy of the taskBinary's broadcast variables to the executor, and my modification will send this taskBinary to each executor that has executed the corresponding stage, reducing the repetition of broadcast variables and avoid deserializing binary multiple times.
### Does this PR introduce _any_ user-facing change?
<!--
Note that it means *any* user-facing change including all aspects such as the documentation fix.
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 possible, please also clarify if this is a user-facing change compared to the released Spark versions or within the unreleased branches such as master.
If no, write 'No'.
-->
No
### 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.
If benchmark tests were added, please run the benchmarks in GitHub Actions for the consistent environment, and the instructions could accord to: https://spark.apache.org/developer-tools.html#github-workflow-benchmarks.
-->
Because this modification is a modification of the Spark Core, and it also involves some parameter changes for unit testing, so I conducted all unit testing on all modules base on master branch.
### Was this patch authored or co-authored using generative AI tooling?
<!--
If generative AI tooling has been used in the process of authoring this patch, please include the
phrase: 'Generated-by: ' followed by the name of the tool and its version.
If no, write 'No'.
Please refer to the [ASF Generative Tooling Guidance](https://www.apache.org/legal/generative-tooling.html) for details.
-->
No
--
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.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
Re: [PR] Keep the taskBinary in Stage to avoid deserializing it multiple times, and send binary to executor instead of task. [spark]
Posted by "mridulm (via GitHub)" <gi...@apache.org>.
mridulm commented on PR #43621:
URL: https://github.com/apache/spark/pull/43621#issuecomment-1790008994
Each task has a reference to a broadcast'ed task binary `Array[Byte]` - not the object.
This `Array[Byte]` is pulled to each executor once - and for each subsequent task, this `Array[Byte]` is reused to deserialize and create a new task closure.
Explicit use of broadcast variables is documented to be read only - but with task closures, we cannot make a change in this behavior. If the instance is reused, task closures can start having side effects on other tasks in the same executor.
--
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.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
Re: [PR] Keep the taskBinary in Stage to avoid deserializing it multiple times, and send binary to executor instead of task. [spark]
Posted by "HyukjinKwon (via GitHub)" <gi...@apache.org>.
HyukjinKwon commented on PR #43621:
URL: https://github.com/apache/spark/pull/43621#issuecomment-1789867256
Mind filing a JIRA please?
--
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.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
Re: [PR] Keep the taskBinary in Stage to avoid deserializing it multiple times, and send binary to executor instead of task. [spark]
Posted by "mridulm (via GitHub)" <gi...@apache.org>.
mridulm commented on PR #43621:
URL: https://github.com/apache/spark/pull/43621#issuecomment-1791582645
The broadcast variable is read once - not n times, and the deserialization n times prevents side effect between tasks, which would be a behavior change if we move away from it.
On the performance aspect , as currently formulated, I would expect negligible (if any) difference - though would be happy to see numbers to the contrary to evaluate effectiveness !
Additionally, given the possibility of side effect, these benefits should be very compelling to entertain if it is worth going down this path.
--
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.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
Re: [PR] Keep the taskBinary in Stage to avoid deserializing it multiple times, and send binary to executor instead of task. [spark]
Posted by "hui730 (via GitHub)" <gi...@apache.org>.
hui730 commented on PR #43621:
URL: https://github.com/apache/spark/pull/43621#issuecomment-1790565946
> > My plan is to create a new binary from executor binary，use System.arraycopy().
>
> This is what is effectively happening currently, right ? The underlying serialzed task array array is immutable - and is repeatedly read to deserialize into the task closure.
>
> I want to make sure I understand the proposal, and how it is different from what Spark is currently doing.
Assuming that there are currently n tasks (same stage) running simultaneously in the executor.
the current situation is to read the remote broadcast, deserializing it, and then start running these n tasks. This step is serial.
In my modification, the step of reading the remote broadcast and deserializing it into array [byte], which is asynchronous. It is parallel to the launchTask. This can save time.
--
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.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
Re: [PR] Keep the taskBinary in Stage to avoid deserializing it multiple times, and send binary to executor instead of task. [spark]
Posted by "hui730 (via GitHub)" <gi...@apache.org>.
hui730 commented on PR #43621:
URL: https://github.com/apache/spark/pull/43621#issuecomment-1790095897
> Each task has a reference to a broadcast'ed task binary `Array[Byte]` - not the object. This `Array[Byte]` is pulled to each executor once - and for each subsequent task, this `Array[Byte]` is reused to deserialize and create a new task closure. So essentially we are trying to save on a jvm local deserialization cost.
>
> Explicit use of broadcast variables is documented to be read only - but with task closures, I would be very reluctant to assume this behavior. If the instance is reused, task closures can start having side effects on other tasks in the same executor.
Okay, I get what you mean. Especially in Hadoop.
My plan is to create a new binary from executor binary，use System.arraycopy().
How do you like it， sir？:)
--
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.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
Re: [PR] Keep the taskBinary in Stage to avoid deserializing it multiple times, and send binary to executor instead of task. [spark]
Posted by "mridulm (via GitHub)" <gi...@apache.org>.
mridulm commented on PR #43621:
URL: https://github.com/apache/spark/pull/43621#issuecomment-1790348164
> My plan is to create a new binary from executor binary，use System.arraycopy().
This is what is effectively happening currently, right ?
The underlying serialzed task array array is immutable - and is repeatedly read to deserialize into the task closure.
I want to make sure I understand the proposal, and how it is different from what Spark is currently doing.
--
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.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
Re: [PR] Keep the taskBinary in Stage to avoid deserializing it multiple times, and send binary to executor instead of task. [spark]
Posted by "hui730 (via GitHub)" <gi...@apache.org>.
hui730 commented on PR #43621:
URL: https://github.com/apache/spark/pull/43621#issuecomment-1789940175
> Mind filing a JIRA please?
Sure, I'll complete it today
--
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.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org
Re: [PR] Keep the taskBinary in Stage to avoid deserializing it multiple times, and send binary to executor instead of task. [spark]
Posted by "hui730 (via GitHub)" <gi...@apache.org>.
hui730 closed pull request #43621: Keep the taskBinary in Stage to avoid deserializing it multiple times, and send binary to executor instead of task.
URL: https://github.com/apache/spark/pull/43621
--
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.
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org