Work with a Hudi dataset - Amazon EMR
Initialize a Spark session for HudiWrite to a Hudi datasetUpsert dataDelete a record Read from a Hudi dataset
Services or capabilities described in Amazon Web Services documentation might vary by Region. To see the differences applicable to the China Regions, see Getting Started with Amazon Web Services in China (PDF).

Work with a Hudi dataset

Hudi supports inserting, updating, and deleting data in Hudi datasets through Spark. For more information, see Writing Hudi tables in Apache Hudi documentation.

The following examples demonstrate how to launch the interactive Spark shell, use Spark submit, or use Amazon EMR Notebooks to work with Hudi on Amazon EMR. You can also use the Hudi DeltaStreamer utility or other tools to write to a dataset. Throughout this section, the examples demonstrate working with datasets using the Spark shell while connected to the master node using SSH as the default hadoop user.

When running spark-shell, spark-submit, or spark-sql using Amazon EMR 6.7.0 or later, pass the following commands.

Note

Amazon EMR 6.7.0 uses Apache Hudi 0.11.0-amzn-0, which contains significant improvements over previous Hudi versions. For more information, see the Apache Hudi 0.11.0 Migration Guide. The examples on this tab reflect these changes.

To open the Spark shell on the primary node

  1. Connect to the primary node using SSH. For more information, see Connect to the primary node using SSH in the Amazon EMR Management Guide.

  2. Enter the following command to launch the Spark shell. To use the PySpark shell, replace spark-shell with pyspark.

    spark-shell --jars /usr/lib/hudi/hudi-spark-bundle.jar \
--conf "spark.serializer=org.apache.spark.serializer.KryoSerializer" \    
--conf "spark.sql.catalog.spark_catalog=org.apache.spark.sql.hudi.catalog.HoodieCatalog"  \
--conf "spark.sql.extensions=org.apache.spark.sql.hudi.HoodieSparkSessionExtension"

When running spark-shell, spark-submit, or spark-sql using Amazon EMR 6.6.x or earlier, pass the following commands.

Note

  • Amazon EMR 6.2 and 5.31 and later (Hudi 0.6.x and later) can omit the spark-avro.jar from the configuration.

  • Amazon EMR 6.5 and 5.35 and later (Hudi 0.9.x and later) can omit spark.sql.hive.convertMetastoreParquet=false from the configuration.

  • Amazon EMR 6.6 and 5.36 and later (Hudi 0.10.x and later) must include the HoodieSparkSessionExtension config as described in the Version: 0.10.0 Spark Guide: 

    --conf  "spark.sql.extensions=org.apache.spark.sql.hudi.HoodieSparkSessionExtension" \
To open the Spark shell on the primary node

  1. Connect to the primary node using SSH. For more information, see Connect to the primary node using SSH in the Amazon EMR Management Guide.

  2. Enter the following command to launch the Spark shell. To use the PySpark shell, replace spark-shell with pyspark.

    spark-shell \
--conf "spark.serializer=org.apache.spark.serializer.KryoSerializer" \
--conf "spark.sql.hive.convertMetastoreParquet=false" \
--jars /usr/lib/hudi/hudi-spark-bundle.jar,/usr/lib/spark/external/lib/spark-avro.jar

To use Hudi with Amazon EMR Notebooks, you must first copy the Hudi jar files from the local file system to HDFS on the master node of the notebook cluster. You then use the notebook editor to configure your EMR notebook to use Hudi.

To use Hudi with Amazon EMR Notebooks

  1. Create and launch a cluster for Amazon EMR Notebooks. For more information, see Creating Amazon EMR clusters for notebooks in the Amazon EMR Management Guide.

  2. Connect to the master node of the cluster using SSH and then copy the jar files from the local filesystem to HDFS as shown in the following examples. In the example, we create a directory in HDFS for clarity of file management. You can choose your own destination in HDFS, if desired.

    hdfs dfs -mkdir -p /apps/hudi/lib
    hdfs dfs -copyFromLocal /usr/lib/hudi/hudi-spark-bundle.jar /apps/hudi/lib/hudi-spark-bundle.jar

  3. Open the notebook editor, enter the code from the following example, and run it.

    %%configure
{ "conf": {
            "spark.jars":"hdfs:///apps/hudi/lib/hudi-spark-bundle.jar",
            "spark.serializer":"org.apache.spark.serializer.KryoSerializer",
            "spark.sql.catalog.spark_catalog": "org.apache.spark.sql.hudi.catalog.HoodieCatalog",
            "spark.sql.extensions":"org.apache.spark.sql.hudi.HoodieSparkSessionExtension"
          }}

To use Hudi with Amazon EMR Notebooks, you must first copy the Hudi jar files from the local file system to HDFS on the master node of the notebook cluster. You then use the notebook editor to configure your EMR notebook to use Hudi.

To use Hudi with Amazon EMR Notebooks

  1. Create and launch a cluster for Amazon EMR Notebooks. For more information, see Creating Amazon EMR clusters for notebooks in the Amazon EMR Management Guide.

  2. Connect to the master node of the cluster using SSH and then copy the jar files from the local filesystem to HDFS as shown in the following examples. In the example, we create a directory in HDFS for clarity of file management. You can choose your own destination in HDFS, if desired.

    hdfs dfs -mkdir -p /apps/hudi/lib
    hdfs dfs -copyFromLocal /usr/lib/hudi/hudi-spark-bundle.jar /apps/hudi/lib/hudi-spark-bundle.jar
    hdfs dfs -copyFromLocal /usr/lib/spark/external/lib/spark-avro.jar /apps/hudi/lib/spark-avro.jar

  3. Open the notebook editor, enter the code from the following example, and run it.

    { "conf": {
            "spark.jars":"hdfs:///apps/hudi/lib/hudi-spark-bundle.jar,hdfs:///apps/hudi/lib/spark-avro.jar",
            "spark.serializer":"org.apache.spark.serializer.KryoSerializer",
            "spark.sql.hive.convertMetastoreParquet":"false"
          }}

Initialize a Spark session for Hudi

When you use Scala, you must import the following classes in your Spark session. This needs to be done once per Spark session.

import org.apache.spark.sql.SaveMode
import org.apache.spark.sql.functions._
import org.apache.hudi.DataSourceWriteOptions
import org.apache.hudi.DataSourceReadOptions
import org.apache.hudi.config.HoodieWriteConfig
import org.apache.hudi.hive.MultiPartKeysValueExtractor
import org.apache.hudi.hive.HiveSyncConfig
import org.apache.hudi.sync.common.HoodieSyncConfig

Write to a Hudi dataset

The following examples show how to create a DataFrame and write it as a Hudi dataset.

Note

To paste code samples into the Spark shell, type :paste at the prompt, paste the example, and then press CTRL + D.

Each time you write a DataFrame to a Hudi dataset, you must specify DataSourceWriteOptions. Many of these options are likely to be identical between write operations. The following example specifies common options using the hudiOptions variable, which subsequent examples use.

Note

Amazon EMR 6.7.0 uses Apache Hudi 0.11.0-amzn-0, which contains significant improvements over previous Hudi versions. For more information, see the Apache Hudi 0.11.0 Migration Guide. The examples on this tab reflect these changes.

// Create a DataFrame
val inputDF = Seq(
 ("100", "2015-01-01", "2015-01-01T13:51:39.340396Z"),
 ("101", "2015-01-01", "2015-01-01T12:14:58.597216Z"),
 ("102", "2015-01-01", "2015-01-01T13:51:40.417052Z"),
 ("103", "2015-01-01", "2015-01-01T13:51:40.519832Z"),
 ("104", "2015-01-02", "2015-01-01T12:15:00.512679Z"),
 ("105", "2015-01-02", "2015-01-01T13:51:42.248818Z")
 ).toDF("id", "creation_date", "last_update_time")

//Specify common DataSourceWriteOptions in the single hudiOptions variable 
val hudiOptions = Map[String,String](
  HoodieWriteConfig.TBL_NAME.key -> "tableName",
  DataSourceWriteOptions.TABLE_TYPE.key -> "COPY_ON_WRITE", 
  DataSourceWriteOptions.RECORDKEY_FIELD_OPT_KEY -> "id",
  DataSourceWriteOptions.PARTITIONPATH_FIELD_OPT_KEY -> "creation_date",
  DataSourceWriteOptions.PRECOMBINE_FIELD_OPT_KEY -> "last_update_time",
  DataSourceWriteOptions.HIVE_SYNC_ENABLED_OPT_KEY -> "true",
  DataSourceWriteOptions.HIVE_TABLE_OPT_KEY -> "tableName",
  DataSourceWriteOptions.HIVE_PARTITION_FIELDS_OPT_KEY -> "creation_date",
  HoodieSyncConfig.META_SYNC_PARTITION_EXTRACTOR_CLASS.key -> "org.apache.hudi.hive.MultiPartKeysValueExtractor",
  HoodieSyncConfig.META_SYNC_ENABLED.key -> "true",
  HiveSyncConfig.HIVE_SYNC_MODE.key -> "hms",
  HoodieSyncConfig.META_SYNC_TABLE_NAME.key -> "tableName",
  HoodieSyncConfig.META_SYNC_PARTITION_FIELDS.key -> "creation_date"
)

// Write the DataFrame as a Hudi dataset
(inputDF.write
    .format("hudi")
    .options(hudiOptions)
    .option(DataSourceWriteOptions.OPERATION_OPT_KEY,"insert")
    .mode(SaveMode.Overwrite)
    .save("s3://DOC-EXAMPLE-BUCKET/myhudidataset/"))

// Create a DataFrame
val inputDF = Seq(
 ("100", "2015-01-01", "2015-01-01T13:51:39.340396Z"),
 ("101", "2015-01-01", "2015-01-01T12:14:58.597216Z"),
 ("102", "2015-01-01", "2015-01-01T13:51:40.417052Z"),
 ("103", "2015-01-01", "2015-01-01T13:51:40.519832Z"),
 ("104", "2015-01-02", "2015-01-01T12:15:00.512679Z"),
 ("105", "2015-01-02", "2015-01-01T13:51:42.248818Z")
 ).toDF("id", "creation_date", "last_update_time")

//Specify common DataSourceWriteOptions in the single hudiOptions variable 
val hudiOptions = Map[String,String](
  HoodieWriteConfig.TABLE_NAME -> "tableName",
  DataSourceWriteOptions.TABLE_TYPE_OPT_KEY -> "COPY_ON_WRITE", 
  DataSourceWriteOptions.RECORDKEY_FIELD_OPT_KEY -> "id",
  DataSourceWriteOptions.PARTITIONPATH_FIELD_OPT_KEY -> "creation_date",
  DataSourceWriteOptions.PRECOMBINE_FIELD_OPT_KEY -> "last_update_time",
  DataSourceWriteOptions.HIVE_SYNC_ENABLED_OPT_KEY -> "true",
  DataSourceWriteOptions.HIVE_TABLE_OPT_KEY -> "tableName",
  DataSourceWriteOptions.HIVE_PARTITION_FIELDS_OPT_KEY -> "creation_date",
  DataSourceWriteOptions.HIVE_PARTITION_EXTRACTOR_CLASS_OPT_KEY -> classOf[MultiPartKeysValueExtractor].getName
)

// Write the DataFrame as a Hudi dataset
(inputDF.write
    .format("org.apache.hudi")
    .option(DataSourceWriteOptions.OPERATION_OPT_KEY, DataSourceWriteOptions.INSERT_OPERATION_OPT_VAL)
    .options(hudiOptions)
    .mode(SaveMode.Overwrite)
    .save("s3://DOC-EXAMPLE-BUCKET/myhudidataset/"))

# Create a DataFrame
inputDF = spark.createDataFrame(
    [
        ("100", "2015-01-01", "2015-01-01T13:51:39.340396Z"),
        ("101", "2015-01-01", "2015-01-01T12:14:58.597216Z"),
        ("102", "2015-01-01", "2015-01-01T13:51:40.417052Z"),
        ("103", "2015-01-01", "2015-01-01T13:51:40.519832Z"),
        ("104", "2015-01-02", "2015-01-01T12:15:00.512679Z"),
        ("105", "2015-01-02", "2015-01-01T13:51:42.248818Z"),
    ],
    ["id", "creation_date", "last_update_time"]
)

# Specify common DataSourceWriteOptions in the single hudiOptions variable
hudiOptions = {
'hoodie.table.name': 'tableName',
'hoodie.datasource.write.recordkey.field': 'id',
'hoodie.datasource.write.partitionpath.field': 'creation_date',
'hoodie.datasource.write.precombine.field': 'last_update_time',
'hoodie.datasource.hive_sync.enable': 'true',
'hoodie.datasource.hive_sync.table': 'tableName',
'hoodie.datasource.hive_sync.partition_fields': 'creation_date',
'hoodie.datasource.hive_sync.partition_extractor_class': 'org.apache.hudi.hive.MultiPartKeysValueExtractor'
}

# Write a DataFrame as a Hudi dataset
inputDF.write \
.format('org.apache.hudi') \
.option('hoodie.datasource.write.operation', 'insert') \
.options(**hudiOptions) \
.mode('overwrite') \
.save('s3://DOC-EXAMPLE-BUCKET/myhudidataset/')
Note

You might see "hoodie" instead of Hudi in code examples and notifications. The Hudi codebase widely uses the old "hoodie" spelling.

DataSourceWriteOptions reference for Hudi
Option Description

TABLE_NAME

The table name under which to register the dataset.

TABLE_TYPE_OPT_KEY

Optional. Specifies whether the dataset is created as "COPY_ON_WRITE" or "MERGE_ON_READ". The default is "COPY_ON_WRITE".

RECORDKEY_FIELD_OPT_KEY

The record key field whose value will be used as the recordKey component of HoodieKey. Actual value will be obtained by invoking .toString() on the field value. Nested fields can be specified using the dot notation, for example, a.b.c.

PARTITIONPATH_FIELD_OPT_KEY

The partition path field whose value will be used as the partitionPath component of HoodieKey. The actual value will be obtained by invoking .toString() on the field value.

PRECOMBINE_FIELD_OPT_KEY

The field used in pre-combining before actual write. When two records have the same key value, Hudi picks the one with the largest value for the precombine field as determined by Object.compareTo(..).

The following options are required only to register the Hudi dataset table in your metastore. If you do not register your Hudi dataset as a table in the Hive metastore, these options are not required.

DataSourceWriteOptions reference for Hive
Option Description

HIVE_DATABASE_OPT_KEY

The Hive database to sync to. The default is "default".

HIVE_PARTITION_EXTRACTOR_CLASS_OPT_KEY

The class used to extract partition field values into Hive partition columns.

HIVE_PARTITION_FIELDS_OPT_KEY

The field in the dataset to use for determining Hive partition columns.

HIVE_SYNC_ENABLED_OPT_KEY

When set to "true", registers the dataset with the Apache Hive metastore. The default is "false".

HIVE_TABLE_OPT_KEY

Required. The name of the table in Hive to sync to. For example, "my_hudi_table_cow".

HIVE_USER_OPT_KEY

Optional. The Hive user name to use when syncing. For example, "hadoop".

HIVE_PASS_OPT_KEY

Optional. The Hive password for the user specified by HIVE_USER_OPT_KEY.

HIVE_URL_OPT_KEY

The Hive metastore URL.

Upsert data

The following example demonstrates how to upsert data by writing a DataFrame. Unlike the previous insert example, the OPERATION_OPT_KEY value is set to UPSERT_OPERATION_OPT_VAL. In addition, .mode(SaveMode.Append) is specified to indicate that the record should be appended.

Note

Amazon EMR 6.7.0 uses Apache Hudi 0.11.0-amzn-0, which contains significant improvements over previous Hudi versions. For more information, see the Apache Hudi 0.11.0 Migration Guide. The examples on this tab reflect these changes.

// Create a new DataFrame from the first row of inputDF with a different creation_date value
val updateDF = inputDF.limit(1).withColumn("creation_date", lit("new_value"))

(updateDF.write
    .format("hudi")
    .options(hudiOptions)
    .option(DataSourceWriteOptions.OPERATION_OPT_KEY, "upsert")
    .mode(SaveMode.Append)
    .save("s3://DOC-EXAMPLE-BUCKET/myhudidataset/"))
    
// Create a new DataFrame from the first row of inputDF with a different creation_date value
val updateDF = inputDF.limit(1).withColumn("creation_date", lit("new_value"))

(updateDF.write
    .format("org.apache.hudi")
    .option(DataSourceWriteOptions.OPERATION_OPT_KEY, DataSourceWriteOptions.UPSERT_OPERATION_OPT_VAL)
    .options(hudiOptions)
    .mode(SaveMode.Append)
    .save("s3://DOC-EXAMPLE-BUCKET/myhudidataset/"))
from pyspark.sql.functions import lit

# Create a new DataFrame from the first row of inputDF with a different creation_date value
updateDF = inputDF.limit(1).withColumn('creation_date', lit('new_value'))

updateDF.write \
    .format('org.apache.hudi') \
    .option('hoodie.datasource.write.operation', 'upsert') \
    .options(**hudiOptions) \
    .mode('append') \
    .save('s3://DOC-EXAMPLE-BUCKET/myhudidataset/')

Delete a record

To hard delete a record, you can upsert an empty payload. In this case, the PAYLOAD_CLASS_OPT_KEY option specifies the EmptyHoodieRecordPayload class. The example uses the same DataFrame, updateDF, used in the upsert example to specify the same record.

Note

Amazon EMR 6.7.0 uses Apache Hudi 0.11.0-amzn-0, which contains significant improvements over previous Hudi versions. For more information, see the Apache Hudi 0.11.0 Migration Guide. The examples on this tab reflect these changes.

(updateDF.write
    .format("hudi")
    .options(hudiOptions)
    .option(DataSourceWriteOptions.OPERATION_OPT_KEY, "delete")
    .mode(SaveMode.Append)
    .save("s3://DOC-EXAMPLE-BUCKET/myhudidataset/"))
(updateDF.write
    .format("org.apache.hudi")
    .option(DataSourceWriteOptions.OPERATION_OPT_KEY, DataSourceWriteOptions.UPSERT_OPERATION_OPT_VAL)
    .option(DataSourceWriteOptions.PAYLOAD_CLASS_OPT_KEY, "org.apache.hudi.common.model.EmptyHoodieRecordPayload")
    .mode(SaveMode.Append)
    .save("s3://DOC-EXAMPLE-BUCKET/myhudidataset/"))
updateDF.write \
    .format('org.apache.hudi') \
    .option('hoodie.datasource.write.operation', 'upsert') \
    .option('hoodie.datasource.write.payload.class', 'org.apache.hudi.common.model.EmptyHoodieRecordPayload') \
    .options(**hudiOptions) \
    .mode('append') \
    .save('s3://DOC-EXAMPLE-BUCKET/myhudidataset/')

You can also hard delete data by setting OPERATION_OPT_KEY to DELETE_OPERATION_OPT_VAL to remove all records in the dataset you submit. For instructions on performing soft deletes, and for more information about deleting data stored in Hudi tables, see Deletes in the Apache Hudi documentation.

Read from a Hudi dataset

To retrieve data at the present point in time, Hudi performs snapshot queries by default. Following is an example of querying the dataset written to S3 in Write to a Hudi dataset. Replace s3://DOC-EXAMPLE-BUCKET/myhudidataset with your table path, and add wildcard asterisks for each partition level, plus one additional asterisk. In this example, there is one partition level, so we've added two wildcard symbols.

Note

Amazon EMR 6.7.0 uses Apache Hudi 0.11.0-amzn-0, which contains significant improvements over previous Hudi versions. For more information, see the Apache Hudi 0.11.0 Migration Guide. The examples on this tab reflect these changes.

val snapshotQueryDF = spark.read
    .format("hudi")
    .load("s3://DOC-EXAMPLE-BUCKET/myhudidataset") 
    .show()
(val snapshotQueryDF = spark.read
    .format("org.apache.hudi")
    .load("s3://DOC-EXAMPLE-BUCKET/myhudidataset" + "/*/*"))

snapshotQueryDF.show()
snapshotQueryDF = spark.read \
    .format('org.apache.hudi') \
    .load('s3://DOC-EXAMPLE-BUCKET/myhudidataset' + '/*/*')
    
snapshotQueryDF.show()

Incremental queries

You can also perform incremental queries with Hudi to get a stream of records that have changed since a given commit timestamp. To do so, set the QUERY_TYPE_OPT_KEY field to QUERY_TYPE_INCREMENTAL_OPT_VAL. Then, add a value for BEGIN_INSTANTTIME_OPT_KEY to obtain all records written since the specified time. Incremental queries are typically ten times more efficient than their batch counterparts since they only process changed records.

When you perform incremental queries, use the root (base) table path without the wildcard asterisks used for Snapshot queries.

Note

Presto does not support incremental queries.

(val incQueryDF = spark.read
    .format("org.apache.hudi")
    .option(DataSourceReadOptions.QUERY_TYPE_OPT_KEY, DataSourceReadOptions.QUERY_TYPE_INCREMENTAL_OPT_VAL)
    .option(DataSourceReadOptions.BEGIN_INSTANTTIME_OPT_KEY, <beginInstantTime>)
    .load("s3://DOC-EXAMPLE-BUCKET/myhudidataset" ))
     
incQueryDF.show()
readOptions = {
  'hoodie.datasource.query.type': 'incremental',
  'hoodie.datasource.read.begin.instanttime': <beginInstantTime>,
}

incQueryDF = spark.read \
    .format('org.apache.hudi') \
    .options(**readOptions) \
    .load('s3://DOC-EXAMPLE-BUCKET/myhudidataset')
    
incQueryDF.show()

For more information about reading from Hudi datasets, see Querying Hudi tables in the Apache Hudi documentation.