[SPARK-35754][CORE] Add config to put migrating blocks on disk only

[q2w]

What changes were proposed in this pull request?

This PR adds a config which makes block manager decommissioner to migrate block data on disk only.

Why are the changes needed?

While migrating block data, if enough memory is not available on peer block managers existing blocks are dropped. After this PR migrating blocks won't drop any existing blocks.

Does this PR introduce any user-facing change?

No

How was this patch tested?

UT in BlockManagerSuite

[q2w]

@Ngone51 Please have a look.

[dongjoon-hyun]

cc @holdenk

[dongjoon-hyun]

ok to test

[dongjoon-hyun]

Could you check the UT failure, @q2w ? If it passes in your environment, please re-trigger GitHub Action.

[info] *** 1 TEST FAILED ***
[error] Failed: Total 3061, Failed 1, Errors 0, Passed 3060, Ignored 7, Canceled 1
[error] Failed tests:
[error] 	org.apache.spark.storage.BlockManagerDecommissionUnitSuite

[HyukjinKwon]

@q2w, can you rebase and create a new PR? Seems like GA in this PR is somehow messed up.

[SparkQA]

Test build #140375 has finished for PR 32902 at commit c462df9.

• This patch fails Spark unit tests.
• This patch merges cleanly.
• This patch adds no public classes.

[mridulm]

This would be inconsistent with the expectation of the storage level set on the RDD/DF.
Instead, what we can do is prefer disk in case it is available as an option (if storageLevel.useDisk == true)

[q2w]

This would be inconsistent with the expectation of the storage level set on the RDD/DF.
Instead, what we can do is prefer disk in case it is available as an option (if storageLevel.useDisk == true)

@mridulm That's correct. In the current flow of decommissioning,the blocks being offloaded can replace existing cached blocks on peer block manager if enough memory is not available. This way we are saving one cached block by loosing another.
If this change is in, we can avoid loosing cached blocks to place offloading cached blocks.
This PR came out as a result of discussion on another related PR. #32677

[github-actions]

We're closing this PR because it hasn't been updated in a while. This isn't a judgement on the merit of the PR in any way. It's just a way of keeping the PR queue manageable.
If you'd like to revive this PR, please reopen it and ask a committer to remove the Stale tag!