wbo4958 opened a new pull request, #37855:
URL: https://github.com/apache/spark/pull/37855
### What changes were proposed in this pull request?
``` scala
val df = spark.range(0, 100, 1, 50).repartition(4)
val v = df.rdd.mapPartitions { iter => {
Iterator.single(iter.length)
}.collect()
println(v.mkString(","))
```
The above simple code outputs `50,0,0,50`, which means there is no data in
partition 1 and partition 2.
The RoundRobin seems to ensure to distribute the records evenly *in the same
partition*, and not guarantee it between partitions.
Below is the code to generate the key
``` scala
case RoundRobinPartitioning(numPartitions) =>
// Distributes elements evenly across output partitions, starting
from a random partition.
var position = new
Random(TaskContext.get().partitionId()).nextInt(numPartitions)
(row: InternalRow) =>
{ // The HashPartitioner will handle the `mod` by the number of
partitions
position += 1
position
}
```
In this case, There are 50 partitions, each partition will only compute 2
elements. The issue for RoundRobin here is it always starts with position=2 to
do the Roundrobin.
See the output of Random
``` scala
scala> (1 to 200).foreach(partitionId => print(new
Random(partitionId).nextInt(4) + " ")) // the position is always 2.
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
2 2
```
Similarly, the below Random code also outputs the same value,
``` scala
(1 to 200).foreach(partitionId => print(new Random(partitionId).nextInt(2) +
" "))
(1 to 200).foreach(partitionId => print(new Random(partitionId).nextInt(4) +
" "))
(1 to 200).foreach(partitionId => print(new Random(partitionId).nextInt(8) +
" "))
(1 to 200).foreach(partitionId => print(new Random(partitionId).nextInt(16)
+ " "))
(1 to 200).foreach(partitionId => print(new Random(partitionId).nextInt(32)
+ " "))
```
Consider partition 0, the total elements are [0, 1], so when shuffle writes,
for element 0, the key will be (position + 1) = 2 + 1 = 3%4=3, the element 1,
the key will be (position + 1)=(3+1)=4%4 = 0
consider partition 1, the total elements are [2, 3], so when shuffle writes,
for element 2, the key will be (position + 1) = 2 + 1 = 3%4=3, the element 3,
the key will be (position + 1)=(3+1)=4%4 = 0
The calculation is also applied for other left partitions since the starting
position is always 2 for this case.
So, as you can see, each partition will write its elements to Partition [0,
3], which results in Partition [1, 2] without any data.
This PR changes the starting position of RoundRobin. The default position
calculated by `new Random(partitionId).nextInt(numPartitions)` may always be
the same for different partitions, which means each partition will output the
data into the same keys when shuffle writes, and some keys may not have any
data in some special cases.
<!--
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.
-->
### 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 PR can fix the data skew issue for the special cases.
### 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.
-->
Will add some tests and watch CI pass
--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
