Spark Doris Connector

Spark Doris Connector

Spark Doris Connector can support reading data stored in Doris and writing data to Doris through Spark.

Github: https://github.com/apache/doris-spark-connector

• Support reading data from Doris.
• Support Spark DataFrame batch/stream writing data to Doris
• You can map the Doris table to DataFrame or RDD, it is recommended to use DataFrame.
• Support the completion of data filtering on the Doris side to reduce the amount of data transmission.

Version Compatibility

Connector	Spark	Doris	Java	Scala
1.3.1	3.4 ~ 3.1, 2.4, 2.3	1.0 +	8	2.12, 2.11
1.2.0	3.2, 3.1, 2.3	1.0 +	8	2.12, 2.11
1.1.0	3.2, 3.1, 2.3	1.0 +	8	2.12, 2.11
1.0.1	3.1, 2.3	0.12 - 0.15	8	2.12, 2.11

Build and Install

Ready to work

1. Modify the custom_env.sh.tpl file and rename it to custom_env.sh

2. Execute following command in source dir: sh build.sh Follow the prompts to enter the Scala and Spark versions you need to start compiling.

After the compilation is successful, the target jar package will be generated in the dist directory, such as: spark-doris-connector-3.1_2.12-1.2.0-SNAPSHOT.jar. Copy this file to ClassPath in Spark to use Spark-Doris-Connector. For example, Spark running in Local mode, put this file in the jars/ folder. Spark running in Yarn cluster mode, put this file in the pre-deployment package.

For example upload spark-doris-connector-3.1_2.12-1.2.0-SNAPSHOT.jar to hdfs and add hdfs file path in spark.yarn.jars.

1. Upload spark-doris-connector-3.1_2.12-1.2.0-SNAPSHOT.jar Jar to hdfs.

hdfs dfs -mkdir /spark-jars/
hdfs dfs -put /your_local_path/spark-doris-connector-3.1_2.12-1.2.0-SNAPSHOT.jar /spark-jars/

2. Add spark-doris-connector-3.1_2.12-1.2.0-SNAPSHOT.jar dependency in Cluster.

spark.yarn.jars=hdfs:///spark-jars/spark-doris-connector-3.1_2.12-1.2.0-SNAPSHOT.jar

Using Maven

<dependency>
    <groupId>org.apache.doris</groupId>
    <artifactId>spark-doris-connector-3.4_2.12</artifactId>
    <version>1.3.0</version>
</dependency>

Notes

Please replace the Connector version according to the different Spark and Scala versions.

Example

Read

SQL

CREATE
TEMPORARY VIEW spark_doris
USING doris
OPTIONS(
  "table.identifier"="$YOUR_DORIS_DATABASE_NAME.$YOUR_DORIS_TABLE_NAME",
  "fenodes"="$YOUR_DORIS_FE_HOSTNAME:$YOUR_DORIS_FE_RESFUL_PORT",
  "user"="$YOUR_DORIS_USERNAME",
  "password"="$YOUR_DORIS_PASSWORD"
);

SELECT *
FROM spark_doris;

DataFrame

val dorisSparkDF = spark.read.format("doris")
  .option("doris.table.identifier", "$YOUR_DORIS_DATABASE_NAME.$YOUR_DORIS_TABLE_NAME")
  .option("doris.fenodes", "$YOUR_DORIS_FE_HOSTNAME:$YOUR_DORIS_FE_RESFUL_PORT")
  .option("user", "$YOUR_DORIS_USERNAME")
  .option("password", "$YOUR_DORIS_PASSWORD")
  .load()

dorisSparkDF.show(5)

RDD

import org.apache.doris.spark._

val dorisSparkRDD = sc.dorisRDD(
  tableIdentifier = Some("$YOUR_DORIS_DATABASE_NAME.$YOUR_DORIS_TABLE_NAME"),
  cfg = Some(Map(
    "doris.fenodes" -> "$YOUR_DORIS_FE_HOSTNAME:$YOUR_DORIS_FE_RESFUL_PORT",
    "doris.request.auth.user" -> "$YOUR_DORIS_USERNAME",
    "doris.request.auth.password" -> "$YOUR_DORIS_PASSWORD"
  ))
)

dorisSparkRDD.collect()

pySpark

dorisSparkDF = spark.read.format("doris")
  .option("doris.table.identifier", "$YOUR_DORIS_DATABASE_NAME.$YOUR_DORIS_TABLE_NAME")
  .option("doris.fenodes", "$YOUR_DORIS_FE_HOSTNAME:$YOUR_DORIS_FE_RESFUL_PORT")
  .option("user", "$YOUR_DORIS_USERNAME")
  .option("password", "$YOUR_DORIS_PASSWORD")
  .load()
// show 5 lines data 
dorisSparkDF.show(5)

Write

SQL

CREATE
TEMPORARY VIEW spark_doris
USING doris
OPTIONS(
  "table.identifier"="$YOUR_DORIS_DATABASE_NAME.$YOUR_DORIS_TABLE_NAME",
  "fenodes"="$YOUR_DORIS_FE_HOSTNAME:$YOUR_DORIS_FE_RESFUL_PORT",
  "user"="$YOUR_DORIS_USERNAME",
  "password"="$YOUR_DORIS_PASSWORD"
);

INSERT INTO spark_doris
VALUES ("VALUE1", "VALUE2", ...);
# or
INSERT INTO spark_doris
SELECT *
FROM YOUR_TABLE
# or
INSERT OVERWRITE 
SELECT *
FROM YOUR_TABLE 

DataFrame(batch/stream)

## batch sink
val mockDataDF = List(
  (3, "440403001005", "21.cn"),
  (1, "4404030013005", "22.cn"),
  (33, null, "23.cn")
).toDF("id", "mi_code", "mi_name")
mockDataDF.show(5)

mockDataDF.write.format("doris")
  .option("doris.table.identifier", "$YOUR_DORIS_DATABASE_NAME.$YOUR_DORIS_TABLE_NAME")
  .option("doris.fenodes", "$YOUR_DORIS_FE_HOSTNAME:$YOUR_DORIS_FE_RESFUL_PORT")
  .option("user", "$YOUR_DORIS_USERNAME")
  .option("password", "$YOUR_DORIS_PASSWORD")
  //other options
  //specify the fields to write
  .option("doris.write.fields", "$YOUR_FIELDS_TO_WRITE")
  // Support setting Overwrite mode to overwrite data
  // .option("save_mode", SaveMode.Overwrite)
  .save()

## stream sink(StructuredStreaming)

### Result DataFrame with structured data, the configuration method is the same as the batch mode.
val sourceDf = spark.readStream.
       .format("your_own_stream_source")
       .load()

val resultDf = sourceDf.<transformations>

resultDf.writeStream
      .format("doris")
      .option("checkpointLocation", "$YOUR_CHECKPOINT_LOCATION")
      .option("doris.table.identifier", "$YOUR_DORIS_DATABASE_NAME.$YOUR_DORIS_TABLE_NAME")
      .option("doris.fenodes", "$YOUR_DORIS_FE_HOSTNAME:$YOUR_DORIS_FE_RESFUL_PORT")
      .option("user", "$YOUR_DORIS_USERNAME")
      .option("password", "$YOUR_DORIS_PASSWORD")
      .start()
      .awaitTermination()

### There is a column value in the Result DataFrame that can be written directly, such as the value in the kafka message that conforms to the import format

val kafkaSource = spark.readStream
  .format("kafka")
  .option("kafka.bootstrap.servers", "$YOUR_KAFKA_SERVERS")
  .option("startingOffsets", "latest")
  .option("subscribe", "$YOUR_KAFKA_TOPICS")
  .load()
kafkaSource.selectExpr("CAST(key AS STRING)", "CAST(value as STRING)")
  .writeStream
  .format("doris")
  .option("checkpointLocation", "$YOUR_CHECKPOINT_LOCATION")
  .option("doris.table.identifier", "$YOUR_DORIS_DATABASE_NAME.$YOUR_DORIS_TABLE_NAME")
  .option("doris.fenodes", "$YOUR_DORIS_FE_HOSTNAME:$YOUR_DORIS_FE_RESFUL_PORT")
  .option("user", "$YOUR_DORIS_USERNAME")
  .option("password", "$YOUR_DORIS_PASSWORD")
  // Set this option to true, and the value column in the Kafka message will be written directly without processing.
  .option("doris.sink.streaming.passthrough", "true")
  .option("doris.sink.properties.format", "json")
  //other options
  .start()
  .awaitTermination()

Configuration

General

Key	Default Value	Comment
doris.fenodes	--	Doris FE http address, support multiple addresses, separated by commas
doris.table.identifier	--	Doris table identifier, eg, db1.tbl1
doris.request.retries	3	Number of retries to send requests to Doris
doris.request.connect.timeout.ms	30000	Connection timeout for sending requests to Doris
doris.request.read.timeout.ms	30000	Read timeout for sending request to Doris
doris.request.query.timeout.s	3600	Query the timeout time of doris, the default is 1 hour, -1 means no timeout limit
doris.request.tablet.size	Integer.MAX_VALUE	The number of Doris Tablets corresponding to an RDD Partition. The smaller this value is set, the more partitions will be generated. This will increase the parallelism on the Spark side, but at the same time will cause greater pressure on Doris.
doris.read.field	--	List of column names in the Doris table, separated by commas
doris.batch.size	1024	The maximum number of rows to read data from BE at one time. Increasing this value can reduce the number of connections between Spark and Doris. Thereby reducing the extra time overhead caused by network delay.
doris.exec.mem.limit	2147483648	Memory limit for a single query. The default is 2GB, in bytes.
doris.deserialize.arrow.async	false	Whether to support asynchronous conversion of Arrow format to RowBatch required for spark-doris-connector iteration
doris.deserialize.queue.size	64	Asynchronous conversion of the internal processing queue in Arrow format takes effect when doris.deserialize.arrow.async is true
doris.write.fields	--	Specifies the fields (or the order of the fields) to write to the Doris table, fileds separated by commas. By default, all fields are written in the order of Doris table fields.
doris.sink.batch.size	100000	Maximum number of lines in a single write BE
doris.sink.max-retries	0	Number of retries after writing BE failed
doris.sink.properties.format	--	Data format of the stream load. Supported formats: csv, json, arrow(since version 1.4.0) More Multi-parameter details
doris.sink.properties.*	--	Import parameters for Stream Load. For example: Specify column separator: 'doris.sink.properties.column_separator' = ','. More parameter details
doris.sink.task.partition.size	--	The number of partitions corresponding to the Writing task. After filtering and other operations, the number of partitions written in Spark RDD may be large, but the number of records corresponding to each Partition is relatively small, resulting in increased writing frequency and waste of computing resources. The smaller this value is set, the less Doris write frequency and less Doris merge pressure. It is generally used with doris.sink.task.use.repartition.
doris.sink.task.use.repartition	false	Whether to use repartition mode to control the number of partitions written by Doris. The default value is false, and coalesce is used (note: if there is no Spark action before the write, the whole computation will be less parallel). If it is set to true, then repartition is used (note: you can set the final number of partitions at the cost of shuffle).
doris.sink.batch.interval.ms	50	The interval time of each batch sink, unit ms.
doris.sink.enable-2pc	false	Whether to enable two-stage commit. When enabled, transactions will be committed at the end of the job, and all pre-commit transactions will be rolled back when some tasks fail.
doris.sink.auto-redirect	true	Whether to redirect StreamLoad requests. After being turned on, StreamLoad will write through FE and no longer obtain BE information explicitly.

SQL & Dataframe Configuration

Key	Default Value	Comment
user	--	Doris username
password	--	Doris password
doris.filter.query.in.max.count	100	In the predicate pushdown, the maximum number of elements in the in expression value list. If this number is exceeded, the in-expression conditional filtering is processed on the Spark side.
doris.ignore-type	--	In a temporary view, specify the field types to ignore when reading the schema. eg: when 'doris.ignore-type'='bitmap,hll'

Structured Streaming Configuration

Key	Default Value	Comment
doris.sink.streaming.passthrough	false	Write the value of the first column directly without processing.

RDD Configuration

Key	Default Value	Comment
doris.request.auth.user	--	Doris username
doris.request.auth.password	--	Doris password
doris.filter.query	--	Filter expression of the query, which is transparently transmitted to Doris. Doris uses this expression to complete source-side data filtering.

tip

1. In Spark SQL, when writing data through insert into, if the target table of doris contains BITMAP or HLL type data, you need to set the parameter doris.ignore-type to the corresponding type, and set doris.write.fields maps the corresponding columns, the usage is as follows:

   BITMAP

   CREATE TEMPORARY VIEW spark_doris
   USING doris
   OPTIONS(
   "table.identifier"="$YOUR_DORIS_DATABASE_NAME.$YOUR_DORIS_TABLE_NAME",
   "fenodes"="$YOUR_DORIS_FE_HOSTNAME:$YOUR_DORIS_FE_RESFUL_PORT",
   "user"="$YOUR_DORIS_USERNAME",
   "password"="$YOUR_DORIS_PASSWORD"
   "doris.ignore-type"="bitmap",
   "doris.write.fields"="col1,col2,col3,bitmap_col2=to_bitmap(col2),bitmap_col3=bitmap_hash(col3)"
   );

   HLL

   CREATE TEMPORARY VIEW spark_doris
   USING doris
   OPTIONS(
   "table.identifier"="$YOUR_DORIS_DATABASE_NAME.$YOUR_DORIS_TABLE_NAME",
   "fenodes"="$YOUR_DORIS_FE_HOSTNAME:$YOUR_DORIS_FE_RESFUL_PORT",
   "user"="$YOUR_DORIS_USERNAME",
   "password"="$YOUR_DORIS_PASSWORD"
   "doris.ignore-type"="hll",
   "doris.write.fields"="col1,hll_col1=hll_hash(col1)"
   );

2. Since version 1.3.0, the default value of doris.sink.max-retries configuration is 0, which means no retries are performed by default. When this parameter is set greater than 0, batch-level failure retries will be performed, and data of the configured size of doris.sink.batch.size will be cached in the Spark Executor memory. The memory allocation may need to be appropriately increased.

3. Since version 1.3.0, overwrite mode insertion is supported (only full table-level overwrite insertion is supported). The specific usage is as follows

   DataFrame

   resultDf.format("doris")
     .option("doris.fenodes","$YOUR_DORIS_FE_HOSTNAME:$YOUR_DORIS_FE_RESFUL_PORT")
     // your own options
     .option("save_mode", SaveMode.Overwrite)
     .save()

   SQL

   INSERT OVERWRITE your_target_table
   SELECT * FROM your_source_table

Doris & Spark Column Type Mapping

Doris Type	Spark Type
NULL_TYPE	DataTypes.NullType
BOOLEAN	DataTypes.BooleanType
TINYINT	DataTypes.ByteType
SMALLINT	DataTypes.ShortType
INT	DataTypes.IntegerType
BIGINT	DataTypes.LongType
FLOAT	DataTypes.FloatType
DOUBLE	DataTypes.DoubleType
DATE	DataTypes.DateType
DATETIME	DataTypes.StringType1
DECIMAL	DecimalType
CHAR	DataTypes.StringType
LARGEINT	DecimalType
VARCHAR	DataTypes.StringType
TIME	DataTypes.DoubleType
HLL	Unsupported datatype
Bitmap	Unsupported datatype

• Note: In Connector, DATETIME is mapped to String. Due to the processing logic of the Doris underlying storage engine, when the time type is used directly, the time range covered cannot meet the demand. So use String type to directly return the corresponding time readable text.