[HUDI-3214] Optimize auto partition in spark

• Type: Improvement
• Status: Open
• Priority: Critical
• Resolution: Unresolved
• Component/s: spark, writer-core
• Labels: pull-request-available
• Story Points: 1
• Assignee: Yann Byron
• Reporter: Yann Byron

Reviewers

Shiyan Xu

• Created: 11/Jan/22 14:32
• Updated: 10/Mar/23 01:50

Description

recently, if partition's value has the format like "pt1=xxxx/pt2=yyyy/pt3=zzzz" which split by slash, Hudi will partition automatically. The directory of this table will have multi partition structure.

I think it's unpredictable. So create this umbrella task to optimize auto partition in order to make the behavior more reasonable.

Also, in hudi 0.8, schama will hold `pt1`, `pt2`, `pt3`, but not in 0.9+.

There are a few of sub tasks:

• add a flag to control whether enable auto-partition, to make the default behavior reasonable..
• achieve a new key generator designed specifically for this scenario.
• solve the bug about the different schema when enable hoodie.file.index.enable or not in this case.

 

Test Codes: 

import org.apache.hudi.QuickstartUtils._
import scala.collection.JavaConversions._
import org.apache.spark.sql.SaveMode._
import org.apache.hudi.DataSourceReadOptions._
import org.apache.hudi.DataSourceWriteOptions._
import org.apache.hudi.config.HoodieWriteConfig._

val tableName = "hudi_trips_cow"
val basePath = "file:///tmp/hudi_trips_cow"
val dataGen = new DataGenerator
val inserts = convertToStringList(dataGen.generateInserts(10))

val df = spark.read.json(spark.sparkContext.parallelize(inserts, 2))
val newDf = df.withColumn("partitionpath", regexp_replace($"partitionpath", "(.*)(\\/){1}(.*)(\\/){1}", "continent=$1$2country=$3$4city="))

newDf.write.format("hudi").
options(getQuickstartWriteConfigs).
option(PRECOMBINE_FIELD_OPT_KEY, "ts").
option(RECORDKEY_FIELD_OPT_KEY, "uuid").
option(PARTITIONPATH_FIELD_OPT_KEY, "partitionpath").
option(TABLE_NAME, tableName).
mode(Overwrite).
save(basePath) 

Issue Links

• relates to: HUDI-3065 spark auto partition discovery does not work from 0.9.0
• links to: GitHub Pull Request #5009