Spark Getting Started

Iceberg Format

The Iceberg Format can be accessed using the Connector provided by Iceberg. Refer to the documentation at Iceberg Spark Connector for more information.

Paimon Format

The Paimon Format can be accessed using the Connector provided by Paimon. Refer to the documentation at Paimon Spark Connector for more information.

Mixed Format

To use Amoro in a Spark shell, use the –packages option:

spark-shell --packages com.netease.amoro:amoro-spark-3.3-runtime:0.5.0

If you want to include the connector in your Spark installation, add the amoro-spark-3.3-runtime Jar to Spark’s jars folder.

Adding catalogs

${SPARK_HOME}/bin/spark-sql \
    --conf spark.sql.extensions=com.netease.arctic.spark.ArcticSparkExtensions \
    --conf spark.sql.catalog.local_catalog=com.netease.arctic.spark.ArcticSparkCatalog \
    --conf spark.sql.catalog.local_catalog.url=thrift://${AMS_HOST}:${AMS_PORT}/${AMS_CATALOG_NAME}

Amoro manages the Catalog through AMS, and Spark catalog needs to be mapped to Amoro Catalog via URL, in the following format: thrift://${AMS_HOST}:${AMS_PORT}/${AMS_CATALOG_NAME}, The arctic-spark-connector will automatically download the Hadoop site configuration file through the thrift protocol for accessing the HDFS cluster

The AMS_PORT is the port number of the AMS service’s thrift API interface, with a default value of 1260 The AMS_CATALOG_NAME is the name of the Catalog you want to access on AMS.

Regarding detailed configurations for Spark, please refer to Spark Configurations

Creating a table

In Spark SQL command line, you can execute a create table command using the CREATE TABLE statement.

Before executing a create table operation, please make sure to create the database first.

-- switch to arctic catalog defined in spark conf
use local_catalog;

-- create databsae first 
create database if not exists test_db;

Then switch to the newly created database and perform the create table operation.

use test_db;

-- create a table with 3 columns
create table test1 (id int, data string, ts timestamp) using arctic;

-- create a table with hidden partition
create table test2 (id int, data string, ts timestamp) using arctic partitioned by (days(ts));

-- create a table with hidden partition and primary key
create table test3 (id int, data string, ts timestamp, primary key(id)) using arctic partitioned by (days(ts));

For more information on Spark DDL related to tables, please refer to Spark DDL

Writing to the table

If you are using Spark SQL, you can use the INSERT OVERWRITE or INSERT SQL statement to write data to an Amoro table.

-- insert values into unkeyed table
insert into test2 values 
( 1, "aaa", timestamp('2022-1-1 00:00:00')),
( 2, "bbb", timestamp('2022-1-2 00:00:00')),
( 3, "bbb", timestamp('2022-1-3 00:00:00'));

-- dynamic overwrite table 
insert overwrite test3 values 
( 1, "aaa", timestamp('2022-1-1 00:00:00')),
( 2, "bbb", timestamp('2022-1-2 00:00:00')),
( 3, "bbb", timestamp('2022-1-3 00:00:00'));

If you are using Static Overwrite, you cannot define transforms on partition fields.

Alternatively, you can use the DataFrame API to write data to an Amoro table within a JAR job.

val df = spark.read().load("/path-to-table")
df.writeTo('test_db.table1').overwritePartitions()

For more information on writing to tables, please refer to Spark Writes

Reading from the table

To query the table using SELECT SQL statements

select count(1) as count, data 
from test2 
group by data;

For table with primary keys defined, you can query on ChangeStore by .change

select count(1) as count, data
from test_db.test3.change group by data;

For more information on reading from tables, please refer to Spark Queries