JDBC Connector

The JDBC connector allows you to import data from any relational database with a JDBC driver into Kafka topics. By using JDBC, this connector can support a wide variety of databases without requiring custom code for each one.

Data is loaded by periodically executing a SQL query and creating an output record for each row in the result set. By default, all tables in a database are copied, each to its own output topic. The database is monitored for new or deleted tables and adapts automatically. When copying data from a table, the connector can load only new or modified rows by specifying which columns should be used to detect new or modified data.

Quickstart

To see the basic functionality of the connector, we’ll copy a single table from a local SQLite database. In this simple example, we’ll assume each entry in the table is assigned a unique ID and is not modified after creation.

Note

You can use your favorite database instead of SQLite. Follow the same steps, but adjust the connection.url setting for your database. Confluent Platform includes JDBC drivers for SQLite and PostgreSQL, but if you’re using a different database you’ll also need to make sure the JDBC driver is available on the Kafka Connect process’s CLASSPATH.

Start by creating a database (you’ll need to install SQLite if you haven’t already):

$ sqlite3 test.db
SQLite version 3.8.10.2 2015-05-20 18:17:19
Enter ".help" for usage hints.
sqlite>

Next in the SQLite command prompt, create a table and seed it with some data:

sqlite> CREATE TABLE accounts(id INTEGER PRIMARY KEY AUTOINCREMENT NOT NULL, name VARCHAR(255));
sqlite> INSERT INTO accounts(name) VALUES('alice');
sqlite> INSERT INTO accounts(name) VALUES('bob');

Now we create a configuration file that will load data from this database. This file is included with the connector in ./etc/kafka-connect-jdbc/quickstart-sqlite.properties and contains the following settings:

name=test-sqlite-jdbc-autoincrement
connector.class=io.confluent.connect.jdbc.JdbcSourceConnector
tasks.max=1
connection.url=jdbc:sqlite:test.db
mode=incrementing
incrementing.column.name=id
topic.prefix=test-sqlite-jdbc-

The first few settings are common settings you’ll specify for all connectors. The connection.url specifies the database to connect to, in this case a local SQLite database file. The mode indicates how we want to query the data. In this case, we have an auto-incrementing unique ID, so we select incrementing mode and set incrementing.column.name. In this mode, each query for new data will only return rows with IDs larger than the maximum ID seen in previous queries. Finally, we can control the names of the topics that each table’s output is sent to with the topic.prefix setting. Since we only have one table, the only output topic in this example will be test-sqlite-jdbc-accounts.

Now, run the connector in a standalone Kafka Connect worker in another terminal (this assumes Avro settings and that Kafka and the Schema Registry are running locally on the default ports):

$ ./bin/connect-standalone ./etc/schema-registry/connect-avro-standalone.properties ./etc/kafka-connect-jdbc/quickstart-sqlite.properties

You should see the process start up and log some messages, and then it will begin executing queries and sending the results to Kafka. In order to check that it has copied the data that was present when we started Kafka Connect, start a console consumer, reading from the beginning of the topic:

$ ./bin/kafka-avro-console-consumer --new-consumer --bootstrap-server localhost:9092 --topic test-sqlite-jdbc-accounts --from-beginning
{"id":1,"name":{"string":"alice"}}
{"id":2,"name":{"string":"bob"}}

The output shows the two records as expected, one per line, in the JSON encoding of the Avro records. Each row is represented as an Avro record and each column is a field in the record. We can see both columns in the table, id and name. The IDs were auto-generated and the column is of type INTEGER NOT NULL, which can be encoded directly as an integer. The name column has type STRING and can be NULL. The JSON encoding of Avro encodes the strings in the format {"type": value}, so we can see that both rows have string values with the names specified when we inserted the data.

Now, keeping the consumer process running, add another record via the SQLite command prompt:

sqlite> INSERT INTO accounts(name) VALUES('cathy');

If you switch back to the console consumer, you should see the new record added (and, importantly, the old entries are not repeated):

{"id":3,"name":{"string":"cathy"}}

Note that the default polling interval is 5 seconds, so it may take a few seconds to show up. Depending on your expected rate up updates or desired latency, a smaller poll interval could be used to deliver updates more quickly.

Of course, all the features of Kafka Connect, including offset management and fault tolerance, work with the JDBC connector. You can restart and kill the processes and they will pick up where they left off, copying only new data (as defined by the mode setting).

Features

The JDBC connector supports copying tables with a variety of JDBC data types, adding and removing tables from the database dynamically, whitelists and blacklists, varying polling intervals, and other settings. However, the most important features for most users are the settings controlling how data is incrementally copied from the database.

Kafka Connect tracks the latest record it retrieved from each table, so it can start in the correct location on the next iteration (or in case of a crash). The JDBC connector uses this functionality to only get updated rows from a table (or from the output of a custom query) on each iteration. Several modes are supported, each of which differs in how modified rows are detected.

Incremental Query Modes

Each incremental query mode tracks a set of columns for each row, which it uses to keep track of which rows have been processed and which rows are new or have been updated. The mode setting controls this behavior and supports the following options:

• Incrementing Column: A single column containing a unique ID for each row, where newer rows are guaranteed to have larger IDs, i.e. an AUTOINCREMENT column. Note that this mode can only detect new rows. Updates to existing rows cannot be detected, so this mode should only be used for immutable data. One example where you might use this mode is when streaming fact tables in a data warehouse, since those are typically insert-only.
• Timestamp Column: In this mode, a single column containing a modification timestamp is used to track the last time data was processed and to query only for rows that have been modified since that time. Note that because timestamps are no necessarily unique, this mode cannot guarantee all updated data will be delivered: if 2 rows share the same timestamp and are returned by an incremental query, but only one has been processed before a crash, the second update will be missed when the system recovers.
• Timestamp and Incrementing Columns: This is the most robust and accurate mode, combining an incrementing column with a timestamp column. By combining the two, as long as the timestamp is sufficiently granular, each (id, timestamp) tuple will uniquely identify an update to a row. Even if an update fails after partially completing, unprocessed updates will are still correctly detected and delivered when the system recovers.
• Custom Query: The JDBC connector supports using custom queries instead of copying whole tables. With a custom query, one of the other update automatic update modes can be used as long as the necessary WHERE clause can be correctly appended to the query. Alternatively, the specified query may handle filtering to new updates itself; however, note that no offset tracking will be performed (unlike the automatic modes where incrementing and/or timestamp column values are recorded for each record), so the query must track offsets itself.
• Bulk: This mode is unfiltered and therefore not incremental at all. It will load all rows from a table on each iteration. This can be useful if you want to periodically dump an entire table where entries are eventually deleted and the downstream system can safely handle duplicates.

Note that all incremental query modes that use certain columns to detect changes will require indexes on those columns to efficiently perform the queries.

For incremental query modes that use timestamps, the JDBC connector uses a configuration timestamp.delay.interval.ms to control the waiting period after a row with certain timestamp appears before we include it in the result. The additional wait allows transactions with earlier timestamps to complete and the related changes to be included in the result.

Configuration

The JDBC connector gives you quite a bit of flexibility in the databases you can import data from and how that data is imported. This section first describes how to access databases whose drivers are not included with Confluent Platform, then gives a few example configuration files that cover common scenarios, then provides an exhaustive description of the available configuration options.

JDBC Drivers

The JDBC connector implements the data copying functionality on the generic JDBC APIs, but relies on JDBC drivers to handle the database-specific implementation of those APIs. Confluent Platform ships with a few JDBC drivers, but if the driver for your database is not included you will need to make it available via the CLASSPATH.

One option is to install the JDBC driver jar alongside the connector. The packaged connector is installed in the share/java/kafka-connect-jdbc directory, relative to the installation directory. If you have installed from Debian or RPM packages, the connector will be installed in /usr/share/java/kafka-connect-jdbc. If you installed from zip or tar files, the connector will be installed in the path given above under the directory where you unzipped the Confluent Platform archive.

Alternatively, you can set the CLASSPATH variable before running connect-standalone or connect-distributed. For example:

$ CLASSPATH=/usr/local/firebird/* ./bin/connect-distributed ./config/connect-distributed.properties

would add the JDBC driver for the Firebird database, located in /usr/local/firebird, and allow you to use JDBC connection URLs like jdbc:firebirdsql:localhost/3050:/var/lib/firebird/example.db.

Examples

The full set of configuration options are listed in the next section, but here we provide a few template configurations that cover some common usage scenarios.

Use a whitelist to limit changes to a subset of tables in a MySQL database, using id and modified columns that are standard on all whitelisted tables to detect rows that have been modified. This mode is the most robust because it can combine the unique, immutable row IDs with modification timestamps to guarantee modifications are not missed even if the process dies in the middle of an incremental update query.

name=mysql-whitelist-timestamp-source
connector.class=io.confluent.connect.jdbc.JdbcSourceConnector
tasks.max=10

connection.url=jdbc:mysql://mysql.example.com:3306/my_database?user=alice&password=secret
table.whitelist=users,products,transactions

mode=timestamp+incrementing
timestamp.column.name=modified
incrementing.column.name=id

topic.prefix=mysql-

Use a custom query instead of loading tables, allowing you to join data from multiple tables. As long as the query does not include its own filtering, you can still use the built-in modes for incremental queries (in this case, using a timestamp column). Note that this limits you to a single output per connector and because there is no table name, the topic “prefix” is actually the full topic name in this case.

name=mysql-whitelist-timestamp-source
connector.class=io.confluent.connect.jdbc.JdbcSourceConnector
tasks.max=10

connection.url=jdbc:postgresql://postgres.example.com/test_db?user=bob&password=secret&ssl=true
query=SELECT users.id, users.name, transactions.timestamp, transactions.user_id, transactions.payment FROM users JOIN transactions ON (users.id = transactions.user_id)
mode=timestamp
timestamp.column.name=timestamp

topic.prefix=mysql-joined-data

Schema Evolution

The JDBC connector supports schema evolution when the Avro converter is used. When there is a change in a database table schema, the JDBC connector can detect the change, create a new Kafka Connect schema and try to register a new Avro schema in the Schema Registry. Whether we can successfully register the schema or not depends on the compatibility level of the Schema Registry, which is backward by default.

For example, if we remove a column from a table, the change is backward compatible and the corresponding Avro schema can be successfully registered in the Schema Registry. If we modify the database table schema to change a column type or add a column, when the Avro schema is registered to the Schema Registry, it will be rejected as the changes are not backward compatible.

You can change the compatibility level of Schema Registry to allow incompatible schemas or other compatibility levels. There are two ways to do this:

• Set the compatibility level for subjects which are used by the connector using PUT /config/(string: subject). The subjects have format of topic-key and topic-value where the topic is determined by topic.prefix config and table name.
• Configure the Schema Registry to use other schema compatibility level by setting avro.compatibility.level in Schema Registry. Note that this is a global setting that applies to all schemas in the Schema Registry.

However, due to the limitation of the JDBC API, some compatible schema changes may be treated as incompatible change. For example, adding a column with default value is a backward compatible change. However, limitations of the JDBC API make it difficult to map this to default values of the correct type in a Kafka Connect schema, so the default values are currently omitted. The implications is that even some changes of the database table schema is backward compatible, the schema registered in the Schema Registry is not backward compatible as it doesn’t contain a default value.

If the JDBC connector is used together with the HDFS connector, there are some restrictions to schema compatibility as well. When Hive integration is enabled, schema compatibility is required to be backward, forward and full to ensure that the Hive schema is able to query the whole data under a topic. As some compatible schema change will be treated as incompatible schema change, those changes will not work as the resulting Hive schema will not be able to query the whole data for a topic.