Catalogs

Catalogs #

Catalogs provide metadata, such as databases, tables, partitions, views, and functions and information needed to access data stored in a database or other external systems.

One of the most crucial aspects of data processing is managing metadata. It may be transient metadata like temporary tables, or UDFs registered against the table environment. Or permanent metadata, like that in a Hive Metastore. Catalogs provide a unified API for managing metadata and making it accessible from the Table API and SQL Queries.

Catalog enables users to reference existing metadata in their data systems, and automatically maps them to Flink’s corresponding metadata. For example, Flink can map JDBC tables to Flink table automatically, and users don’t have to manually re-writing DDLs in Flink. Catalog greatly simplifies steps required to get started with Flink with users' existing system, and greatly enhanced user experiences.

Catalog Types #

GenericInMemoryCatalog #

The GenericInMemoryCatalog is an in-memory implementation of a catalog. All objects will be available only for the lifetime of the session.

JdbcCatalog #

The JdbcCatalog enables users to connect Flink to relational databases over JDBC protocol. PostgresCatalog is the only implementation of JDBC Catalog at the moment. See JdbcCatalog documentation for more details on setting up the catalog.

HiveCatalog #

The HiveCatalog serves two purposes; as persistent storage for pure Flink metadata, and as an interface for reading and writing existing Hive metadata. Flink’s Hive documentation provides full details on setting up the catalog and interfacing with an existing Hive installation.

The Hive Metastore stores all meta-object names in lower case. This is unlike GenericInMemoryCatalog which is case-sensitive

User-Defined Catalog #

Catalogs are pluggable and users can develop custom catalogs by implementing the Catalog interface.

In order to use custom catalogs with Flink SQL, users should implement a corresponding catalog factory by implementing the CatalogFactory interface. The factory is discovered using Java’s Service Provider Interfaces (SPI). Classes that implement this interface can be added to META_INF/services/org.apache.flink.table.factories.Factory in JAR files. The provided factory identifier will be used for matching against the required type property in a SQL CREATE CATALOG DDL statement.

Using SQL DDL #

Users can use SQL DDL to create tables in catalogs in both Table API and SQL.

TableEnvironment tableEnv = ...

// Create a HiveCatalog 
Catalog catalog = new HiveCatalog("myhive", null, "<path_of_hive_conf>");

// Register the catalog
tableEnv.registerCatalog("myhive", catalog);

// Create a catalog database
tableEnv.executeSql("CREATE DATABASE mydb WITH (...)");

// Create a catalog table
tableEnv.executeSql("CREATE TABLE mytable (name STRING, age INT) WITH (...)");

tableEnv.listTables(); // should return the tables in current catalog and database.

val tableEnv = ...

// Create a HiveCatalog 
val catalog = new HiveCatalog("myhive", null, "<path_of_hive_conf>")

// Register the catalog
tableEnv.registerCatalog("myhive", catalog)

// Create a catalog database
tableEnv.executeSql("CREATE DATABASE mydb WITH (...)")

// Create a catalog table
tableEnv.executeSql("CREATE TABLE mytable (name STRING, age INT) WITH (...)")

tableEnv.listTables() // should return the tables in current catalog and database.

from pyflink.table.catalog import HiveCatalog

# Create a HiveCatalog
catalog = HiveCatalog("myhive", None, "<path_of_hive_conf>")

# Register the catalog
t_env.register_catalog("myhive", catalog)

# Create a catalog database
t_env.execute_sql("CREATE DATABASE mydb WITH (...)")

# Create a catalog table
t_env.execute_sql("CREATE TABLE mytable (name STRING, age INT) WITH (...)")

# should return the tables in current catalog and database.
t_env.list_tables()

// the catalog should have been registered via yaml file
Flink SQL> CREATE DATABASE mydb WITH (...);

Flink SQL> CREATE TABLE mytable (name STRING, age INT) WITH (...);

Flink SQL> SHOW TABLES;
mytable

For detailed information, please check out Flink SQL CREATE DDL.

Using Java, Scala or Python #

Users can use Java, Scala or Python to create catalog tables programmatically.

import org.apache.flink.table.api.*;
import org.apache.flink.table.catalog.*;
import org.apache.flink.table.catalog.hive.HiveCatalog;

TableEnvironment tableEnv = TableEnvironment.create(EnvironmentSettings.inStreamingMode());

// Create a HiveCatalog 
Catalog catalog = new HiveCatalog("myhive", null, "<path_of_hive_conf>");

// Register the catalog
tableEnv.registerCatalog("myhive", catalog);

// Create a catalog database 
catalog.createDatabase("mydb", new CatalogDatabaseImpl(...));

// Create a catalog table
final Schema schema = Schema.newBuilder()
    .column("name", DataTypes.STRING())
    .column("age", DataTypes.INT())
    .build();

tableEnv.createTable("myhive.mydb.mytable", TableDescriptor.forConnector("kafka")
    .schema(schema)
    // …
    .build());

List<String> tables = catalog.listTables("mydb"); // tables should contain "mytable"
import org.apache.flink.table.api._
import org.apache.flink.table.catalog._
import org.apache.flink.table.catalog.hive.HiveCatalog

val tableEnv = TableEnvironment.create(EnvironmentSettings.inStreamingMode())

// Create a HiveCatalog 
val catalog = new HiveCatalog("myhive", null, "<path_of_hive_conf>")

// Register the catalog
tableEnv.registerCatalog("myhive", catalog)

// Create a catalog database 
catalog.createDatabase("mydb", new CatalogDatabaseImpl(...))

// Create a catalog table
val schema = Schema.newBuilder()
    .column("name", DataTypes.STRING())
    .column("age", DataTypes.INT())
    .build()

tableEnv.createTable("myhive.mydb.mytable", TableDescriptor.forConnector("kafka")
    .schema(schema)
    // …
    .build())
    
val tables = catalog.listTables("mydb") // tables should contain "mytable"
from pyflink.table import *
from pyflink.table.catalog import HiveCatalog, CatalogDatabase, ObjectPath, CatalogBaseTable

settings = EnvironmentSettings.in_batch_mode()
t_env = TableEnvironment.create(settings)

# Create a HiveCatalog
catalog = HiveCatalog("myhive", None, "<path_of_hive_conf>")

# Register the catalog
t_env.register_catalog("myhive", catalog)

# Create a catalog database
database = CatalogDatabase.create_instance({"k1": "v1"}, None)
catalog.create_database("mydb", database)

# Create a catalog table
schema = Schema.new_builder() \
    .column("name", DataTypes.STRING()) \
    .column("age", DataTypes.INT()) \
    .build()
    
catalog_table = t_env.create_table("myhive.mydb.mytable", TableDescriptor.for_connector("kafka")
    .schema(schema)
    // 
    .build())

# tables should contain "mytable"
tables = catalog.list_tables("mydb")

Catalog API #

Note: only catalog program APIs are listed here. Users can achieve many of the same functionalities with SQL DDL. For detailed DDL information, please refer to SQL CREATE DDL.

Database operations #

// create database
catalog.createDatabase("mydb", new CatalogDatabaseImpl(...), false);

// drop database
catalog.dropDatabase("mydb", false);

// alter database
catalog.alterDatabase("mydb", new CatalogDatabaseImpl(...), false);

// get database
catalog.getDatabase("mydb");

// check if a database exist
catalog.databaseExists("mydb");

// list databases in a catalog
catalog.listDatabases();
from pyflink.table.catalog import CatalogDatabase

# create database
catalog_database = CatalogDatabase.create_instance({"k1": "v1"}, None)
catalog.create_database("mydb", catalog_database, False)

# drop database
catalog.drop_database("mydb", False)

# alter database
catalog.alter_database("mydb", catalog_database, False)

# get database
catalog.get_database("mydb")

# check if a database exist
catalog.database_exists("mydb")

# list databases in a catalog
catalog.list_databases()

Table operations #

// create table
catalog.createTable(new ObjectPath("mydb", "mytable"), new CatalogTableImpl(...), false);

// drop table
catalog.dropTable(new ObjectPath("mydb", "mytable"), false);

// alter table
catalog.alterTable(new ObjectPath("mydb", "mytable"), new CatalogTableImpl(...), false);

// rename table
catalog.renameTable(new ObjectPath("mydb", "mytable"), "my_new_table");

// get table
catalog.getTable("mytable");

// check if a table exist or not
catalog.tableExists("mytable");

// list tables in a database
catalog.listTables("mydb");
from pyflink.table import *
from pyflink.table.catalog import CatalogBaseTable, ObjectPath
from pyflink.table.descriptors import Kafka

table_schema = TableSchema.builder() \
    .field("name", DataTypes.STRING()) \
    .field("age", DataTypes.INT()) \
    .build()

table_properties = Kafka() \
    .version("0.11") \
    .start_from_earlist() \
    .to_properties()

catalog_table = CatalogBaseTable.create_table(schema=table_schema, properties=table_properties, comment="my comment")

# create table
catalog.create_table(ObjectPath("mydb", "mytable"), catalog_table, False)

# drop table
catalog.drop_table(ObjectPath("mydb", "mytable"), False)

# alter table
catalog.alter_table(ObjectPath("mydb", "mytable"), catalog_table, False)

# rename table
catalog.rename_table(ObjectPath("mydb", "mytable"), "my_new_table")

# get table
catalog.get_table("mytable")

# check if a table exist or not
catalog.table_exists("mytable")

# list tables in a database
catalog.list_tables("mydb")

View operations #

// create view
catalog.createTable(new ObjectPath("mydb", "myview"), new CatalogViewImpl(...), false);

// drop view
catalog.dropTable(new ObjectPath("mydb", "myview"), false);

// alter view
catalog.alterTable(new ObjectPath("mydb", "mytable"), new CatalogViewImpl(...), false);

// rename view
catalog.renameTable(new ObjectPath("mydb", "myview"), "my_new_view", false);

// get view
catalog.getTable("myview");

// check if a view exist or not
catalog.tableExists("mytable");

// list views in a database
catalog.listViews("mydb");
from pyflink.table import *
from pyflink.table.catalog import CatalogBaseTable, ObjectPath

table_schema = TableSchema.builder() \
    .field("name", DataTypes.STRING()) \
    .field("age", DataTypes.INT()) \
    .build()

catalog_table = CatalogBaseTable.create_view(
    original_query="select * from t1",
    expanded_query="select * from test-catalog.db1.t1",
    schema=table_schema,
    properties={},
    comment="This is a view"
)

catalog.create_table(ObjectPath("mydb", "myview"), catalog_table, False)

# drop view
catalog.drop_table(ObjectPath("mydb", "myview"), False)

# alter view
catalog.alter_table(ObjectPath("mydb", "mytable"), catalog_table, False)

# rename view
catalog.rename_table(ObjectPath("mydb", "myview"), "my_new_view", False)

# get view
catalog.get_table("myview")

# check if a view exist or not
catalog.table_exists("mytable")

# list views in a database
catalog.list_views("mydb")

Partition operations #

// create view
catalog.createPartition(
    new ObjectPath("mydb", "mytable"),
    new CatalogPartitionSpec(...),
    new CatalogPartitionImpl(...),
    false);

// drop partition
catalog.dropPartition(new ObjectPath("mydb", "mytable"), new CatalogPartitionSpec(...), false);

// alter partition
catalog.alterPartition(
    new ObjectPath("mydb", "mytable"),
    new CatalogPartitionSpec(...),
    new CatalogPartitionImpl(...),
    false);

// get partition
catalog.getPartition(new ObjectPath("mydb", "mytable"), new CatalogPartitionSpec(...));

// check if a partition exist or not
catalog.partitionExists(new ObjectPath("mydb", "mytable"), new CatalogPartitionSpec(...));

// list partitions of a table
catalog.listPartitions(new ObjectPath("mydb", "mytable"));

// list partitions of a table under a give partition spec
catalog.listPartitions(new ObjectPath("mydb", "mytable"), new CatalogPartitionSpec(...));

// list partitions of a table by expression filter
catalog.listPartitionsByFilter(new ObjectPath("mydb", "mytable"), Arrays.asList(epr1, ...));
from pyflink.table.catalog import ObjectPath, CatalogPartitionSpec, CatalogPartition

catalog_partition = CatalogPartition.create_instance({}, "my partition")

catalog_partition_spec = CatalogPartitionSpec({"third": "2010", "second": "bob"})
catalog.create_partition(
    ObjectPath("mydb", "mytable"),
    catalog_partition_spec,
    catalog_partition,
    False)

# drop partition
catalog.drop_partition(ObjectPath("mydb", "mytable"), catalog_partition_spec, False)

# alter partition
catalog.alter_partition(
    ObjectPath("mydb", "mytable"),
    CatalogPartitionSpec(...),
    catalog_partition,
    False)

# get partition
catalog.get_partition(ObjectPath("mydb", "mytable"), catalog_partition_spec)

# check if a partition exist or not
catalog.partition_exists(ObjectPath("mydb", "mytable"), catalog_partition_spec)

# list partitions of a table
catalog.list_partitions(ObjectPath("mydb", "mytable"))

# list partitions of a table under a give partition spec
catalog.list_partitions(ObjectPath("mydb", "mytable"), catalog_partition_spec)

Function operations #

// create function
catalog.createFunction(new ObjectPath("mydb", "myfunc"), new CatalogFunctionImpl(...), false);

// drop function
catalog.dropFunction(new ObjectPath("mydb", "myfunc"), false);

// alter function
catalog.alterFunction(new ObjectPath("mydb", "myfunc"), new CatalogFunctionImpl(...), false);

// get function
catalog.getFunction("myfunc");

// check if a function exist or not
catalog.functionExists("myfunc");

// list functions in a database
catalog.listFunctions("mydb");
from pyflink.table.catalog import ObjectPath, CatalogFunction

catalog_function = CatalogFunction.create_instance(class_name="my.python.udf")

# create function
catalog.create_function(ObjectPath("mydb", "myfunc"), catalog_function, False)

# drop function
catalog.drop_function(ObjectPath("mydb", "myfunc"), False)

# alter function
catalog.alter_function(ObjectPath("mydb", "myfunc"), catalog_function, False)

# get function
catalog.get_function("myfunc")

# check if a function exist or not
catalog.function_exists("myfunc")

# list functions in a database
catalog.list_functions("mydb")

Table API and SQL for Catalog #

Registering a Catalog #

Users have access to a default in-memory catalog named default_catalog, that is always created by default. This catalog by default has a single database called default_database. Users can also register additional catalogs into an existing Flink session.

tableEnv.registerCatalog(new CustomCatalog("myCatalog"));
t_env.register_catalog(catalog)

All catalogs defined using YAML must provide a type property that specifies the type of catalog. The following types are supported out of the box.

Catalog Type Value
GenericInMemory generic_in_memory
Hive hive
catalogs:
   - name: myCatalog
     type: custom_catalog
     hive-conf-dir: ...

Changing the Current Catalog And Database #

Flink will always search for tables, views, and UDF’s in the current catalog and database.

tableEnv.useCatalog("myCatalog");
tableEnv.useDatabase("myDb");
t_env.use_catalog("myCatalog")
t_env.use_database("myDb")
Flink SQL> USE CATALOG myCatalog;
Flink SQL> USE myDB;

Metadata from catalogs that are not the current catalog are accessible by providing fully qualified names in the form catalog.database.object.

tableEnv.from("not_the_current_catalog.not_the_current_db.my_table");
t_env.from_path("not_the_current_catalog.not_the_current_db.my_table")
Flink SQL> SELECT * FROM not_the_current_catalog.not_the_current_db.my_table;

List Available Catalogs #

tableEnv.listCatalogs();
t_env.list_catalogs()
Flink SQL> show catalogs;

List Available Databases #

tableEnv.listDatabases();
t_env.list_databases()
Flink SQL> show databases;

List Available Tables #

tableEnv.listTables();
t_env.list_tables()
Flink SQL> show tables;