Spark Configuration

Catalogs configuration

Using Mixed-Format in a standalone catalog

Starting from version 3.x, Spark supports configuring an independent Catalog. If you want to use a Mixed-Format table in a standalone Catalog, you can configure it as follows:

spark.sql.catalog.arctic_catalog=com.netease.arctic.spark.ArcticSparkCatalog
spark.sql.catalog.arctic_catalog.url=thrift://${AMS_HOST}:${AMS_PORT}/${AMS_CATALOG_NAME_HIVE}

Then, execute the following SQL in the Spark SQL Client to switch to the corresponding catalog.

use arctic_catalog;

Of course, you can also access Mixed-Format tables by directly using the triplet arctic_catalog.{db_name}.{table_name}.

You can also set Spark’s default Catalog to your configured Catalog using the following properties. In this way, you don’t need to use the use {catalog} command to switch the default catalog.

spark.sql.defaultCatalog=arctic_catalog
spark.sql.catalog.arctic_catalog=com.netease.arctic.spark.ArcticSparkCatalog
spark.sql.catalog.arctic_catalog.url=thrift://${AMS_HOST}:${AMS_PORT}/${AMS_CATALOG_NAME_HIVE}

In a standalone AmoroSparkCatalog scenario, only Mixed-Format tables can be created and accessed in the corresponding catalog

Using Mixed-Format in session catalog

If you want to access both existing Hive tables or Spark datasource tables and Mixed-Format tables in Spark, you can use the AmoroSparkSessionCatalog as the implementation of the Spark default session catalog. The configuration method is as follows.

spark.sql.catalog.spark_catalog=com.netease.arctic.spark.ArcticSparkSessionCatalog
spark.sql.catalog.spark_catalog.url=thrift://${AMS_HOST}:${AMS_PORT}/${AMS_CATALOG_NAME_HIVE}

When using the ArcticSparkSessionCatalog as the implementation of the spark_catalog, it behaves as follows

  • Load Table: When resolving a db_name.table_name identifier, it will load the table metadata by Spark’s built-in session catalog implementation, and then checking the MixedFormat flag defined in table properties. If the table has the MixedFormat flag, it will be loaded by ArcticSparkCatalog again.

  • Create Table: The behavior of CREATE TABLE is determined by the using {provider} clause in the DDL statement. If the clause contains using arctic, a Mixed-Format table will be created. Otherwise, the default Spark implementation will be used to create the table.

When using the ArcticSparkSessionCatalog, there are several points to keep in mind:

  • ArcticSparkSessionCatalog can only be configured under the spark_catalog
  • The spark.sql.catalogImplementation must be configured as HIVE
  • Catalogs registered on AMS must use a Metastore of the Hive type.

The high availability configuration

If AMS is configured with high availability, you can configure the spark.sql.catalog.{catalog_name}.url property in the following way to achieve higher availability.

spark.sql.catalog.arctic_catalog=com.netease.arctic.spark.ArcticSparkCatalog
spark.sql.catalog.arctic_catalog.url=zookeeper://{zookeeper-endpoint-list}/{cluster-name}/{catalog-name}

Among above:

  • zookeeper-endpoint-list: a list of host:port pairs separated by commas. A valid value could be 192.168.1.1:2181,192.168.1.2:2181,192.168.1.3:2181
  • cluster-name: is the value of ams.cluster.name configured in the configuration file config.yml of AMS, which is used to identify the user space on ZooKeeper.
  • catalog-name: the name of the Catalog on AMS.