Skip to content

Conversation

@pan3793
Copy link
Member

@pan3793 pan3793 commented Jan 17, 2022

What changes were proposed in this pull request?

Update document "Running multiple versions of the Spark Shuffle Service" to mention the workaround for YARN-11053

Why are the changes needed?

User may stuck when they following the current document to deploy multi-versions Spark Shuffle Service on YARN because of YARN-11053

Does this PR introduce any user-facing change?

User document changes.

How was this patch tested?

image

@github-actions github-actions bot added the DOCS label Jan 17, 2022
@pan3793 pan3793 changed the title [SPARK-37925] Update document to mention the workaround for YARN-11053 [SPARK-37925][DOC] Update document to mention the workaround for YARN-11053 Jan 17, 2022
@itholic
Copy link
Contributor

itholic commented Jan 17, 2022

Could you also include the screen-capture for generated web documentation in the PR description if possible ??

@pan3793
Copy link
Member Author

pan3793 commented Jan 17, 2022

@itholic updated.

@AmplabJenkins
Copy link

Can one of the admins verify this patch?

@pan3793
Copy link
Member Author

pan3793 commented Jan 19, 2022

@itholic could you please take a look again?

@itholic
Copy link
Contributor

itholic commented Jan 21, 2022

@xkrogen Could you happen to take a look at this since you initially writing the "Running multiple versions of the Spark Shuffle Service" section ?? (#31936)

also cc @tgravescs @dongjoon-hyun as a reviewer of the previous PR, FYI.

Copy link
Contributor

@xkrogen xkrogen left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for filing this @pan3793 !

It seems that this was broken by YARN-9075, which means it is only broken for 3.3.0 and 3.3.1. Given that it is only broken for such a narrow subset of versions, maybe we can move the workaround instructions to a separate part of the documentation (e.g. a separate section lower down), so it doesn't clutter up the main documentation for this feature?

Comment on lines 931 to 934
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Rather than setting it to some arbitrary class name, it would be better to set it to an empty string. In this way, the ApplicationClassLoader will be configured to use the default system classes (ref). This is probably more desirable behavior than completely overriding the system classes config, and is more similar to the changes from YARN-11053.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Let's not add this as it is a workaround for older YARN. It does not need to be a part of Spark docs

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Let's clean up the sentence a little:
"Note that versions of YARN prior to 3.3.2/3.4.0 have an issue[...] which requires setting yarn.nodemanager.aux-services.<service-name>.system-classes as a workaround. See issue for details."

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

does this affect all 2.9.0 til 3.3.2? it would be good to be explicit about the versions it affects.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

As I mentioned above, this should only affect versions 3.3.0 and 3.3.1

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for comments, updated.

Comment on lines 931 to 934
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Let's not add this as it is a workaround for older YARN. It does not need to be a part of Spark docs

@pan3793
Copy link
Member Author

pan3793 commented Feb 1, 2022

Updated based on comments, would you please take a look again? @srowen

@pan3793 pan3793 requested a review from srowen February 1, 2022 21:17
this. In addition to setting up separate classpaths, it's necessary to ensure the two versions
advertise to different ports. This can be achieved using the `spark-shuffle-site.xml` file described
above. For example, you may have configuration like:
this. Notes that YARN 3.3.0/3.3.1 have an issue which requires setting
Copy link
Member

@srowen srowen Feb 1, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Notes -> Note
as a workaround, see -> as a workaround. See

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

updated.

@pan3793 pan3793 requested a review from srowen February 2, 2022 05:19
@srowen
Copy link
Member

srowen commented Feb 5, 2022

Merged to master

@srowen srowen closed this in 74ebef2 Feb 5, 2022
@pan3793 pan3793 deleted the SPARK-37925 branch May 30, 2022 14:51
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants