Skip to content

[SPARK-18788][SPARKR] Add API for getNumPartitions #16668

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 6 commits into from
Closed

[SPARK-18788][SPARKR] Add API for getNumPartitions #16668

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
34f9aa5
getNumPartitions
felixcheung Jan 21, 2017
7c057fc
rename getNumPartitions for RDD
felixcheung Jan 21, 2017
ad1bd14
fix generic
felixcheung Jan 21, 2017
bab7466
more broken calls
felixcheung Jan 22, 2017
0353978
improve
felixcheung Jan 26, 2017
c05f786
remove note
felixcheung Jan 26, 2017
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions R/pkg/NAMESPACE
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -94,6 +94,7 @@ exportMethods("arrange",
"freqItems",
"gapply",
"gapplyCollect",
"getNumPartitions",
"group_by",
"groupBy",
"head",
Expand Down
25 changes: 25 additions & 0 deletions R/pkg/R/DataFrame.R
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3406,3 +3406,28 @@ setMethod("randomSplit",
}
sapply(sdfs, dataFrame)
})

#' getNumPartitions
#'
#' Return the number of partitions
#' Note: in order to compute the number of partition the SparkDataFrame has to be converted into a
#' RDD temporarily internally.
#'
#' @param x A SparkDataFrame
#' @family SparkDataFrame functions
#' @aliases getNumPartitions,SparkDataFrame-method
#' @rdname getNumPartitions
#' @name getNumPartitions
#' @export
#' @examples
#'\dontrun{
#' sparkR.session()
#' df <- createDataFrame(cars, numPartitions = 2)
#' getNumPartitions(df)
#' }
#' @note getNumPartitions since 2.1.1
Copy link
Member

@HyukjinKwon HyukjinKwon Jan 21, 2017

Choose a reason for hiding this comment

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

@felixcheung, should this be since 2.2.0? Just curious.

Copy link
Member Author

@felixcheung felixcheung Jan 21, 2017

Choose a reason for hiding this comment

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

I debated about this quite a bit - generally it should but we merged createDataFrame(..., numPartitions) to 2.1 and it felt important to have a getNumPartition in the same release too.

setMethod("getNumPartitions",
signature(x = "SparkDataFrame"),
function(x) {
getNumPartitionsRDD(toRDD(x))
Copy link
Contributor

@shivaram shivaram Jan 21, 2017

Choose a reason for hiding this comment

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

As discussed in the JIRA I worry that this will be a very expensive operation for large data frames. Specifically instead of create an RRDD, can we do some operations on the Scala side which might be cheaper ?

cc @yhuai @cloud-fan who know more about DataFrame on the SQL side

Copy link
Member Author

@felixcheung felixcheung Jan 22, 2017

Choose a reason for hiding this comment

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

Right, we agreed.
The conversion, especially into RRDD, is in particular concerning. From what I can see though this df.rdd.getNumPartitions is the recommended practice, which seems to be all over pyspark. (granted, DataFrame to RDD in pyspark is likely slightly more efficient)

An alternative, is we could wrap all of this on the JVM side - at least that should save us the around trip to RRDD.

But agreed, is there a more efficient way this could be exposed in DataFrame/Dataset directly instead?

Copy link
Contributor

@cloud-fan cloud-fan Jan 23, 2017

Choose a reason for hiding this comment

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

shall we add the getNumPartitions to DataFrame/Dataset at scala side?

Copy link
Member Author

@felixcheung felixcheung Jan 23, 2017

Choose a reason for hiding this comment

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

That would be great!

Copy link
Contributor

@cloud-fan cloud-fan Jan 24, 2017

Choose a reason for hiding this comment

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

are you going to do it here? Or do we need to send a new PR for the scala side changes?

Copy link
Contributor

@cloud-fan cloud-fan Jan 24, 2017

Choose a reason for hiding this comment

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

isn't just calling rdd.numPartitions? we need to materialize the RDD inside DataFrame anyway, but it's cheap at scala side.

Copy link
Member Author

@felixcheung felixcheung Jan 24, 2017

Choose a reason for hiding this comment

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

ah, that we could do easily. is that something ok for Spark 2.1.1? If yes, I could go ahead with changes here for Scala, Python and R.

Copy link
Contributor

@cloud-fan cloud-fan Jan 24, 2017

Choose a reason for hiding this comment

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

you said this filled a hole for Spark 2.1, what's this hole? is this Spark R only?

Copy link
Member Author

@felixcheung felixcheung Jan 24, 2017

Choose a reason for hiding this comment

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

sorry, I should clarify. Yes, for R only - since SparkR only has DataFrame APIs and no (publicly supported) RDD APIs, users are left without a way to check number of partitions.

Copy link
Contributor

@cloud-fan cloud-fan Jan 24, 2017

Choose a reason for hiding this comment

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

maybe we can add this slow implementation to Spark 2.1, and improve it in Spark 2.2

})
2 changes: 1 addition & 1 deletion R/pkg/R/RDD.R
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -313,7 +313,7 @@ setMethod("checkpoint",
#' @rdname getNumPartitions
#' @aliases getNumPartitions,RDD-method
#' @noRd
setMethod("getNumPartitions",
setMethod("getNumPartitionsRDD",
signature(x = "RDD"),
function(x) {
callJMethod(getJRDD(x), "getNumPartitions")
Expand Down
10 changes: 5 additions & 5 deletions R/pkg/inst/tests/testthat/test_sparkSQL.R
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -196,18 +196,18 @@ test_that("create DataFrame from RDD", {
expect_equal(dtypes(df), list(c("name", "string"), c("age", "int"), c("height", "float")))
expect_equal(as.list(collect(where(df, df$name == "John"))),
list(name = "John", age = 19L, height = 176.5))
expect_equal(getNumPartitions(toRDD(df)), 1)
expect_equal(getNumPartitions(df), 1)

df <- as.DataFrame(cars, numPartitions = 2)
expect_equal(getNumPartitions(toRDD(df)), 2)
expect_equal(getNumPartitions(df), 2)
df <- createDataFrame(cars, numPartitions = 3)
expect_equal(getNumPartitions(toRDD(df)), 3)
expect_equal(getNumPartitions(df), 3)
# validate limit by num of rows
df <- createDataFrame(cars, numPartitions = 60)
expect_equal(getNumPartitions(toRDD(df)), 50)
expect_equal(getNumPartitions(df), 50)
# validate when 1 < (length(coll) / numSlices) << length(coll)
df <- createDataFrame(cars, numPartitions = 20)
expect_equal(getNumPartitions(toRDD(df)), 20)
expect_equal(getNumPartitions(df), 20)

df <- as.DataFrame(data.frame(0))
expect_is(df, "SparkDataFrame")
Expand Down