[SPARK-16101][SQL] Refactoring CSV data source to be consistent with JSON data source by HyukjinKwon · Pull Request #13988 · apache/spark

HyukjinKwon · 2016-06-30T03:54:29Z

What changes were proposed in this pull request?

This PR refactors CSV data source to be consistent with JSON data source.

This PR removes classes CSVParser and introduces new classes UnivocityParser and UnivocityGenerator to be consistent with JSON data source (JacksonParser, JacksonGenerator). Also, CSVRelation is merged with CSVFileFormat just like JsonFileFormat.

This is a rough look of this change:

CSVOptions - reading/writing settings that can be created from options from CsvParser.

UnivocityGenerator - writing logics from CSVRelation and CsvParser

UnivocityParser - parsing logics in CSVTypeCast, CsvParser and CSVRelation

CSVFileFormat - CSVOutputWriterFactory and CsvOutputWriter in CSVRelation

This PR makes the methods in classes have consistent arguments with JSON ones.

UnivocityGenerator and JacksonGenerator

private[csv] classUnivocityGenerator(
schema: StructType,
writer: Writer,
options: CSVOptions=newCSVOptions(Map.empty[String, String])) {
...
defwrite ...
defclose ...
defflush ...

private[sql] classJacksonGenerator(
schema: StructType,
writer: Writer,
options: JSONOptions=newJSONOptions(Map.empty[String, String])) {
...
defwrite ...
defclose ...
defflush ...

UnivocityParser resembles JacksonParser.

private[csv] classUnivocityParser(
schema: StructType,
requiredSchema: StructType,
options: CSVOptions) extendsLogging {
defthis(schema: StructType, options: CSVOptions) =this(schema, schema, options)
...
defparse(input: String)

classJacksonParser(
schema: StructType,
columnNameOfCorruptRecord: String,
options: JSONOptions) extendsLogging {
...
defparse(input: String)

csv.CSVInferSchema and json.InferSchema

csv.CSVInferSchema.infer(csv: Dataset[String], caseSensitive: Boolean, options: CSVOptions)
json.InferSchema.infer(json: RDD[String], columnNameOfCorruptRecord: String , configOptions: JSONOptions)

This PR also makes the classes put in together in a consistent manner with JSON.
- CsvFileFormat
```
CsvFileFormatCsvOutputWriter
```
- JsonFileFormat
```
JsonFileFormatJsonOutputWriter
```

Also, this re-write existing CSV parsing logics to re-use the row, separate parsing mode logic/convering logics and etc.

How was this patch tested?

Existing tests should cover this.

HyukjinKwon · 2016-06-30T03:57:14Z

I still need to correct some nits and check the consistency with JSON data source but I opened this just to check if it breaks anything. I will submit some more commits soon. (and will also update the PR description to be in more details maybe).

SparkQA · 2016-06-30T05:23:37Z

Test build #61523 has finished for PR 13988 at commit 211bfb4.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

HyukjinKwon · 2016-07-01T01:23:00Z

Hi @rxin, I think the change in this PR might be still pretty big. Should I maybe make this separate into two PRs for both reading and writing parts?

SparkQA · 2016-07-01T03:10:04Z

Test build #61587 has finished for PR 13988 at commit b1050b5.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-07-01T03:22:27Z

Test build #61588 has finished for PR 13988 at commit 0d60c57.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

HyukjinKwon · 2016-07-07T03:46:11Z

(@rxin gentle ping..)

SparkQA · 2016-07-09T07:18:34Z

Test build #62011 has finished for PR 13988 at commit f0d1512.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-07-09T08:43:37Z

Test build #62014 has finished for PR 13988 at commit 97d7bd4.

This patch fails MiMa tests.
This patch merges cleanly.
This patch adds no public classes.

HyukjinKwon · 2016-07-09T08:47:09Z

retest this please

HyukjinKwon · 2016-07-09T09:26:39Z

I updated the PR description. I hope this is helpful for reviewing.

SparkQA · 2016-07-09T10:09:36Z

Test build #62015 has finished for PR 13988 at commit 97d7bd4.

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

SparkQA · 2016-07-09T10:26:25Z

Test build #62016 has finished for PR 13988 at commit 97d7bd4.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

deanchen · 2016-07-18T03:07:15Z

@HyukjinKwon we ran in to this issue where csv writes ints for DateType instead of date string. (https://issues.apache.org/jira/browse/SPARK-16597)

Actually, I opened another PR here, #13912. Maybe it is about that PR.

ah thanks, commented on that PR. Glad to see someone showing some love to Spark's csv datasource!

HyukjinKwon · 2016-07-27T00:24:11Z

@rxin Could you take a look please?

SparkQA · 2016-08-09T09:12:51Z

Test build #63425 has finished for PR 13988 at commit a634435.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-08-10T02:14:45Z

Test build #63482 has finished for PR 13988 at commit c143a01.

This patch passes all tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
- case class CachedData(plan: LogicalPlan, cachedRepresentation: InMemoryRelation)
- class CacheManager extends Logging
- trait DataSourceScanExec extends LeafExecNode with CodegenSupport
- case class RowDataSourceScanExec(
- case class FileSourceScanExec(
- case class ExternalRDD[T](
- case class ExternalRDDScanExec[T](
- case class LogicalRDD(
- case class RDDScanExec(
- trait FileRelation
- case class LocalTableScanExec(
- abstract class RowIterator
- trait LeafExecNode extends SparkPlan
- trait UnaryExecNode extends SparkPlan
- trait BinaryExecNode extends SparkPlan
- case class PlanLater(plan: LogicalPlan) extends LeafExecNode
- abstract class SparkStrategies extends QueryPlanner[SparkPlan]
- class UnsafeRowSerializer(
- case class ScalaUDAF(
- case class InMemoryRelation(
- case class InMemoryTableScanExec(
- trait RunnableCommand extends LogicalPlan with logical.Command
- case class ExecutedCommandExec(cmd: RunnableCommand) extends SparkPlan
- case class AlterTableRecoverPartitionsCommand(
- case class DataSourceAnalysis(conf: CatalystConf) extends Rule[LogicalPlan]
- class FindDataSourceTable(sparkSession: SparkSession) extends Rule[LogicalPlan]
- case class InsertIntoDataSourceCommand(
- case class InsertIntoHadoopFsRelationCommand(
- case class PartitionDirectory(values: InternalRow, path: Path)
- case class PartitionSpec(
- class CSVOutputWriterFactory(options: CSVOptions) extends OutputWriterFactory
- class CsvOutputWriter(
- case class JDBCPartition(whereClause: String, idx: Int) extends Partition
- class ResolveDataSource(sparkSession: SparkSession) extends Rule[LogicalPlan]
- case class PreprocessTableInsertion(conf: SQLConf) extends Rule[LogicalPlan]
- case class PreWriteCheck(conf: SQLConf, catalog: SessionCatalog)
- case class DebugExec(child: SparkPlan) extends UnaryExecNode with CodegenSupport
- class ExchangeCoordinator(
- case class MapPartitionsRWrapper(
- class IncrementalExecution(
- class ExecutionPage(parent: SQLTab) extends WebUIPage(\"execution\") with Logging
- class SQLHistoryListenerFactory extends SparkHistoryListenerFactory
- class SQLListener(conf: SparkConf) extends SparkListener with Logging
- class SQLHistoryListener(conf: SparkConf, sparkUI: SparkUI)
- class SQLTab(val listener: SQLListener, sparkUI: SparkUI)
- case class SparkPlanGraph(

SparkQA · 2016-08-23T03:26:36Z

Test build #64254 has finished for PR 13988 at commit 346c4a6.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-08-30T08:28:16Z

Test build #64635 has finished for PR 13988 at commit 52ca52b.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

HyukjinKwon · 2016-09-10T08:07:15Z

cc @hvanhovell Do you mind if I ask to review this please? I remember the initial proposal was reviewed by you. If this seems too big to review, I can split this into reading path and writing path.

HyukjinKwon · 2016-09-10T08:10:44Z

This is also loosely related with https://issues.apache.org/jira/browse/SPARK-15463. After this one is merged, we could resemble the implementation of JSON one easily rather then introducing another refactoring.

SparkQA · 2016-09-21T15:16:03Z

Test build #65716 has finished for PR 13988 at commit 015567e.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-09-28T10:28:54Z

Test build #66037 has finished for PR 13988 at commit a6d85b6.

This patch fails to build.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-09-28T11:44:29Z

Test build #66035 has finished for PR 13988 at commit df69c8d.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-09-28T12:43:38Z

Test build #66038 has finished for PR 13988 at commit ac94e67.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

HyukjinKwon · 2016-10-08T08:37:18Z

@hvanhovell If this change looks too big, I will split this into reading path and writing path if you confirm please.

SparkQA · 2016-10-14T04:02:37Z

Test build #66939 has finished for PR 13988 at commit 697f276.

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

SparkQA · 2016-10-14T04:27:29Z

Test build #66936 has finished for PR 13988 at commit 97eec87.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

SparkQA · 2016-10-22T17:15:53Z

Test build #67391 has finished for PR 13988 at commit 346b1d2.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

HyukjinKwon · 2016-11-28T14:45:41Z

I will try to split this into two PRs for read path and write path. Would that sound okay to you both @rxin and @hvanhovell?

…to SPARK-16101

HyukjinKwon · 2017-01-05T11:48:05Z

 schema.foreach(field => verifyType(field.dataType))
 }
 }
+


These below just came from CSVRelation.

HyukjinKwon · 2017-01-05T11:49:21Z

@@ -39,22 +38,43 @@ private[csv] object CSVInferSchema {
 * 3. Replace any null types with string type
 */
 def infer(


This argument change is kind of a important change to introduce similar functionalities with JSON. (e,g., creating a dataframe from RDD[String] or Dataset[String]).

HyukjinKwon · 2017-01-05T11:50:19Z

- case datum =>
- Try(datum.toDouble)
- .getOrElse(NumberFormat.getInstance(Locale.US).parse(datum).doubleValue())
+ private def makeSafeHeader(


This just came from CSVFileFormat.

HyukjinKwon · 2017-01-05T11:53:34Z


 val isCommentSet = this.comment != '\u0000'
+
+ def asWriterSettings: CsvWriterSettings = {


These just came from CSVParser.

HyukjinKwon · 2017-01-05T11:54:33Z

+ writerSettings.setHeaders(schema.fieldNames: _*)
+ private val gen = new CsvWriter(writer, writerSettings)
+
+ // A `ValueConverter` is responsible for converting a value of an `InternalRow` to `String`.


These below just mostly came from CSVRelation.

HyukjinKwon · 2017-01-05T11:54:56Z

+ private type ValueConverter = String => Any
+
+ var numMalformedRecords = 0
+ val row = new GenericInternalRow(requiredSchema.length)


Now, we reuse the single row.

Also, it separates numMalformedRecords when it calls parse (...) which looked weird before.

HyukjinKwon · 2017-01-05T11:55:43Z

+ * each element represents a column) and turns it into either one resulting row or no row (if the
+ * the record is malformed).
+ */
+ def parse(input: String): Option[InternalRow] = {


Here, I separate the parsing mode logics (withParseMode) and actual converting logics (parse).

Also, the argument change (matching it up to JacksonParser) is also important. We could avoid additional refactoring when introducing the same funtionalities with JacksonParser, (e.g., loading it from RDD[String] or Dataset[String], from_json and to_json functions).

For example, PR - 13300 introduces such refactoring.

SparkQA · 2017-01-05T12:14:02Z

Test build #70917 has finished for PR 13988 at commit 72e7dcc.

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

SparkQA · 2017-01-05T13:14:19Z

Test build #70919 has finished for PR 13988 at commit 34fc6ca.

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

SparkQA · 2017-01-05T13:28:12Z

Test build #70920 has finished for PR 13988 at commit 08e5fe7.

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

SparkQA · 2017-01-05T14:33:49Z

Test build #70923 has finished for PR 13988 at commit 4c6666d.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

HyukjinKwon · 2017-01-06T02:31:29Z

Hi @cloud-fan, do you mind if I ask to check whether it looks making sense?

HyukjinKwon · 2017-01-06T02:42:48Z

+ options: CSVOptions) extends Logging {
+ def this(schema: StructType, options: CSVOptions) = this(schema, schema, options)
+
+ val valueConverters = makeConverters(schema, options)


Some changes about converting here came from CSVTypeCast.

cloud-fan · 2017-01-06T13:24:47Z

can you split this into smaller PRs? it's really painful to review such a big refactor-only PR.

HyukjinKwon · 2017-01-06T14:31:42Z

Sure! Let me split this into reading and writing ones. Thank you for yout comments. Let me close this for now.

HyukjinKwon changed the title ~~[WIP][SPARK-16101][SQL] Refactoring CSV data source to be consistent with JSON data source~~[SPARK-16101][SQL][WIP] Refactoring CSV data source to be consistent with JSON data sourceJun 30, 2016

HyukjinKwon changed the title ~~[SPARK-16101][SQL][WIP] Refactoring CSV data source to be consistent with JSON data source~~[SPARK-16101][SQL] Refactoring CSV data source to be consistent with JSON data sourceJul 1, 2016

deanchen reviewed Jul 18, 2016
View reviewed changes

HyukjinKwon mentioned this pull request Aug 8, 2016
[TRIVIAL][SQL] Match the name of OrcRelation companion object to OrcFileFormat #14529
Closed

HyukjinKwon force-pushed the SPARK-16101 branch 2 times, most recently from a6d85b6 to ac94e67Compare September 28, 2016 10:25

HyukjinKwon force-pushed the SPARK-16101 branch from 97eec87 to 697f276Compare October 14, 2016 02:23

HyukjinKwon force-pushed the SPARK-16101 branch from 697f276 to 346b1d2Compare October 22, 2016 15:09

Refactoring CSV data source to be consistent with JSON data source
72e7dcc

HyukjinKwon force-pushed the SPARK-16101 branch from 346b1d2 to 72e7dccCompare January 5, 2017 10:36

HyukjinKwon added 3 commits January 5, 2017 19:39

Refactoring CSV data source to be consistent with JSON data source
5228538

Simpler and prettier
0a7eb8d

Merge branch 'SPARK-16101' ofhttps://github.com/HyukjinKwon/sparkin…
34fc6ca
…to SPARK-16101

HyukjinKwon commented Jan 5, 2017
View reviewed changes

Even clean up more
08e5fe7

HyukjinKwon commented Jan 5, 2017
View reviewed changes

Let's keep the existing behaviour
4c6666d

HyukjinKwon commented Jan 6, 2017
View reviewed changes

HyukjinKwon closed this Jan 6, 2017

HyukjinKwon deleted the SPARK-16101 branch January 2, 2018 03:44


		val isCommentSet = this.comment != '\u0000'

		def asWriterSettings: CsvWriterSettings = {

Uh oh!

Conversation

HyukjinKwon commented Jun 30, 2016• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What changes were proposed in this pull request?

How was this patch tested?

Uh oh!

HyukjinKwon commented Jun 30, 2016• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

SparkQA commented Jun 30, 2016

Uh oh!

HyukjinKwon commented Jul 1, 2016• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

SparkQA commented Jul 1, 2016

Uh oh!

SparkQA commented Jul 1, 2016

Uh oh!

HyukjinKwon commented Jul 7, 2016

Uh oh!

SparkQA commented Jul 9, 2016

Uh oh!

SparkQA commented Jul 9, 2016

Uh oh!

HyukjinKwon commented Jul 9, 2016

Uh oh!

HyukjinKwon commented Jul 9, 2016

Uh oh!

SparkQA commented Jul 9, 2016

Uh oh!

SparkQA commented Jul 9, 2016

Uh oh!

deanchenJul 18, 2016

Choose a reason for hiding this comment

Uh oh!

HyukjinKwonJul 18, 2016• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

deanchenJul 18, 2016

Choose a reason for hiding this comment

Uh oh!

HyukjinKwon commented Jul 27, 2016

Uh oh!

SparkQA commented Aug 9, 2016

Uh oh!

SparkQA commented Aug 10, 2016

Uh oh!

SparkQA commented Aug 23, 2016

Uh oh!

SparkQA commented Aug 30, 2016

Uh oh!

HyukjinKwon commented Sep 10, 2016

Uh oh!

HyukjinKwon commented Sep 10, 2016

Uh oh!

SparkQA commented Sep 21, 2016

Uh oh!

SparkQA commented Sep 28, 2016

Uh oh!

SparkQA commented Sep 28, 2016

Uh oh!

SparkQA commented Sep 28, 2016

Uh oh!

HyukjinKwon commented Oct 8, 2016

Uh oh!

SparkQA commented Oct 14, 2016

Uh oh!

SparkQA commented Oct 14, 2016

Uh oh!

SparkQA commented Oct 22, 2016

Uh oh!

HyukjinKwon commented Nov 28, 2016

Uh oh!

HyukjinKwonJan 5, 2017• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

HyukjinKwonJan 5, 2017• edited Loading Uh oh! There was an error while loading. Please reload this page.

HyukjinKwon commented Jun 30, 2016•
edited
Loading

HyukjinKwon commented Jun 30, 2016•
edited
Loading

HyukjinKwon commented Jul 1, 2016•
edited
Loading

HyukjinKwonJul 18, 2016•
edited
Loading

HyukjinKwonJan 5, 2017•
edited
Loading

HyukjinKwonJan 5, 2017•
edited
Loading

HyukjinKwonJan 5, 2017•
edited
Loading

HyukjinKwonJan 5, 2017•
edited
Loading

HyukjinKwonJan 5, 2017•
edited
Loading

HyukjinKwonJan 5, 2017•
edited
Loading