Important

You are viewing documentation for an older version of Confluent Platform. For the latest, click here.

JDBC Source Connector for Confluent Platform¶

The Kafka Connect JDBC source connector allows you to import data from any relational database with a JDBC driver into Apache 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.

You can configure Java streams applications to deserialize and ingest data in multiple ways, including Kafka console producers, JDBC source connectors, and Java client producers. For full code examples, see Pipelining with Kafka Connect and Kafka Streams.

Configuration Properties¶

For a complete list of configuration properties for this connector, see JDBC Connector Source Connector Configuration Properties.

Note

For an example of how to get Kafka Connect connected to Confluent Cloud, see Distributed Cluster in Connect Kafka Connect to Confluent Cloud.

Quick Start¶

Prerequisites

Confluent Platform
Confluent CLI (requires separate installation)

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

Prerequisites¶

Confluent Platform is installed and services are running by using the Confluent CLI. This quick start assumes that you are using the Confluent CLI, but standalone installations are also supported. By default ZooKeeper, Kafka, Schema Registry, Kafka Connect REST API, and Kafka Connect are started with the confluent local start command. For more information, see On-Premises Deployments.
SQLite is installed. You can also use another database. If you are using another database, be sure to adjust the connection.url setting. Confluent Platform includes JDBC drivers for SQLite and PostgreSQL, but if you’re using a different database you must also verify that the JDBC driver is available on the Kafka Connect process’s CLASSPATH.
Kafka and Schema Registry are running locally on the default ports.

Create SQLite Database and Load Data¶

Create a SQLite database with this command:

sqlite3 test.db

Your output should resemble:

SQLite version 3.19.3 2017-06-27 16:48:08
Enter ".help" for usage hints.
sqlite>

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');

Tip

You can run SELECT * from accounts; to verify your table has been created.

Load the JDBC Source Connector¶

Load the predefined JDBC source connector.

Optional: View the available predefined connectors with the confluent local list command.

Tip

The command syntax for the Confluent CLI development commands changed in 5.3.0. These commands have been moved to confluent local. For example, the syntax for confluent start is now confluent local start. For more information, see confluent local.
```
confluent local list connectors
```
Your output should resemble:
```
Bundled Predefined Connectors (edit configuration under etc/):
  elasticsearch-sink
  file-source
  file-sink
  jdbc-source
  jdbc-sink
  hdfs-sink
  s3-sink
```

Load the jdbc-source connector. The test.db file must be in the same directory where Connect is started.

confluent local load jdbc-source

Your output should resemble:

{
  "name": "jdbc-source",
  "config": {
    "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-",
    "name": "jdbc-source"
  },
  "tasks": [],
  "type": null
}

Tip

For non-CLI users, you can load the JDBC sink connector with this command:

<path-to-confluent>/bin/connect-standalone \
<path-to-confluent>/etc/schema-registry/connect-avro-standalone.properties \
<path-to-confluent>/etc/kafka-connect-jdbc/sink-quickstart-sqlite.properties

To check that it has copied the data that was present when you started Kafka Connect, start a console consumer, reading from the beginning of the topic:

  ./bin/kafka-avro-console-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. You 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 you can see that both rows have string values with the names specified when you inserted the data.

Add a Record to the Consumer¶

Add another record via the SQLite command prompt:

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

You can switch back to the console consumer and see the new record is added and, importantly, the old entries are not repeated:

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

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

All the features of Kafka Connect, including offset management and fault tolerance, work with the source 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 source 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 source 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. Incrementing columns must be integral types.
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 still be correctly detected and delivered when the system recovers.
Custom Query: The source 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 source connector uses a configuration timestamp.delay.interval.ms to control the waiting period after a row with certain timestamp appears before you 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. For more information, see JDBC Connector Source Connector Configuration Properties.

Message Keys¶

Kafka messages are key/value pairs. For a JDBC connector, the value (payload) is the contents of the table row being ingested. However, the JBDC connector does not generate the key by default.

Message keys are useful in setting up partitioning strategies. Keys can direct messages to a specific partition and can support downstream processing where joins are used. If no message key is used, messages are sent to partitions using round-robin distribution.

To set a message key for the JBDC connector, you use two Single Message Transformations (SMTs): the ValueToKey SMT and the ExtractField SMT. You add these two SMTs to the JBDC connector configuration. For example, the following shows a snippet added to a configuration that takes the id column of the accounts table to use as the message key.

curl -X POST http://localhost:8083/connectors -H "Content-Type: application/json" -d '{
      "name": "jdbc_source_mysql_01",
      "config": {
              "connector.class": "io.confluent.connect.jdbc.JdbcSourceConnector",
              "connection.url": "jdbc:mysql://mysql:3306/test",
              "connection.user": "connect_user",
              "connection.password": "connect_password",
              "topic.prefix": "mysql-01-",
              "poll.interval.ms" : 3600000,
              "table.whitelist" : "test.accounts",
              "mode":"bulk",
              "transforms":"createKey,extractInt",
              "transforms.createKey.type":"org.apache.kafka.connect.transforms.ValueToKey",
              "transforms.createKey.fields":"id",
              "transforms.extractInt.type":"org.apache.kafka.connect.transforms.ExtractField$Key",
              "transforms.extractInt.field":"id"
              }
      }'

Mapping Column Types¶

The source connector has a few options for controlling how column types are mapped into Kafka Connect field types. By default, the connector maps SQL/JDBC types to the most accurate representation in Java, which is straightforward for many SQL types but may be a bit unexpected for some types, as described in the following section.

Numeric mapping property¶

SQL’s NUMERIC and DECIMAL types have exact semantics controlled by precision and scale. The most accurate representation for these types is Connect’s Decimal logical type which uses Java’s BigDecimal representation. Avro serializes Decimal types as bytes that may be difficult to consume and that may require additional conversion to an appropriate data type. The source connector’s numeric.mapping configuration property does this by casting numeric values to the most appropriate primitive type using the numeric.mapping=best_fit value. The following values are available for the numeric.mapping configuration property:

none: Use this value if all NUMERIC columns are to be represented by the Kafka Connect Decimal logical type. This is the default value for this property. Decimal types are mapped to their binary representation.
best_fit: Use this value if all NUMERIC columns should be cast to Connect INT8, INT16, INT32, INT64, or FLOAT64 based upon the column’s precision and scale. This is the property value you should likely use if you have NUMERIC/NUMBER source data. It attempts to map NUMERIC columns to the Connect INT8, INT16, INT32, INT64, and FLOAT64 primitive type, based upon the column’s precision and scale values, as shown below:

Precision Scale Connect primitive type

1 to 2 -84 to 0 INT8

3 to 4 -84 to 0 INT16

5 to 9 -84 to 0 INT32

10 to 18 -84 to 0 INT64

1 to 18 positive FLOAT64
precision_only: Use this to map NUMERIC columns based only on the column’s precision (assuming that column’s scale is 0). This option attempts to map NUMERIC columns to Connect INT8, INT16, INT32, and INT64 types based only upon the column’s precision, and where the scale is always 0.

Precision Scale Connect primitive type

1 to 2 0 INT8

3 to 4 0 INT16

5 to 9 0 INT32

10 to 18 0 INT64

Tip

For a deeper dive into this topic, see the Confluent blog article Bytes, Decimals, Numerics and oh my.

Note

The numeric.precision.mapping property is older and is now deprecated. When enabled, it is equivalent to numeric.mapping=precision_only. When not enabled, it is equivalent to numeric.mapping=none.

Configuration¶

The source 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.

The full set of configuration options are listed in JDBC Connector Source Connector Configuration Properties, but here are 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 Connect schema and try to register a new Avro schema in Schema Registry. Whether you can successfully register the schema or not depends on the compatibility level of Schema Registry, which is backward by default.

For example, if you remove a column from a table, the change is backward compatible and the corresponding Avro schema can be successfully registered in Schema Registry. If you modify the database table schema to change a column type or add a column, when the Avro schema is registered to 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 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 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 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.

Pre-execution SQL logging¶

If the connector does not behave as expected, you can enable the connector to log the actual queries and statements before the connector sends them to the database for execution. This allows you to view the complete SQL statements and queries in the log for troubleshooting.

Complete the steps below to troubleshoot the JDBC source connector using pre-execution SQL logging:

Temporarily change the default Connect log4j.logger.io.confluent.connect.jdbc.source property from INFO to TRACE. You can do this in the connect-log4j.properties file or by entering the following curl command:
```
curl -Ss -X PUT '{"level": "TRACE"}' http://localhost:8083/admin/loggers/io.confluent.connect.jdbc.source
```
Important
- You should use TRACE logging temporarily. After troubleshooting is completed, change this property back to INFO.
- This change affects all JDBC source connectors running in the Connect cluster.
Review the log. When using the Confluent CLI to run Confluent Platform locally for development, you can display JDBC source connector log messages using the following CLI command:
```
confluent local log connect | grep JdbcSourceConnector
```
Search for messages in the output that resemble the example below:
```
“Statement to execute: <query-or-statement>“
```

After troubleshooting, return the level to INFO using the following curl command:

curl -Ss -X PUT '{"level": "INFO"}' http://localhost:8083/admin/loggers/io.confluent.connect.jdbc.source

Additional documentation¶

JDBC Connector Source Connector Configuration Properties

Precision	Scale	Connect primitive type
1 to 2	-84 to 0	INT8
3 to 4	-84 to 0	INT16
5 to 9	-84 to 0	INT32
10 to 18	-84 to 0	INT64
1 to 18	positive	FLOAT64

Precision	Scale	Connect primitive type
1 to 2	0	INT8
3 to 4	0	INT16
5 to 9	0	INT32
10 to 18	0	INT64