Sink to Kafka

This topic describes how to sink data from RisingWave to a Kafka broker and how to specify security (encryption and authentication) settings.

A sink is an external target that you can send data to. To stream data out of RisingWave, you need to create a sink. Use the CREATE SINK statement to create a sink. You can create a sink with data from a materialized view or a table. RisingWave only supports writing messages in non-transactional mode.

Delivery semantics

The Kafka sink provides at-least-once delivery guarantees. RisingWave will retry failed sends, which may result in duplicate messages being written to Kafka topics during failures or retries.

• For upsert sinks, duplicates are naturally handled by Kafka consumers through primary key deduplication.
• For append-only sinks, downstream consumers should handle potential duplicates.

To enable idempotent producer behavior at the Kafka level (preventing duplicates due to broker acknowledgment issues), you can set properties.enable.idempotence = true. Note that this provides idempotence for individual sends but does not guarantee end-to-end exactly-once when combined with RisingWave’s checkpoint mechanism.

Syntax

CREATE SINK [ IF NOT EXISTS ] sink_name
[FROM sink_from | AS select_query]
WITH (
   connector='kafka',
   connector_parameter = 'value', ...
)
FORMAT data_format ENCODE data_encode [ (
    key = 'value'
) ]
[KEY ENCODE key_encode [(...)]]
;

Names and unquoted identifiers are case-insensitive. Therefore, you must double-quote any of these fields for them to be case-sensitive. See also Identifiers.

Basic parameters

All WITH options are required unless explicitly mentioned as optional.

Parameter or clause	Description
sink_name	Name of the sink to be created.
sink_from	A clause that specifies the direct source from which data will be output. sink_from can be a materialized view or a table. Either this clause or a SELECT query must be specified.
AS select_query	A SELECT query that specifies the data to be output to the sink. Either this query or a FROM clause must be specified. See SELECT for the syntax and examples of the SELECT command.
connector	Sink connector type must be kafka for Kafka sink.
properties.bootstrap.server	Address of the Kafka broker. Format: ip:port. If there are multiple brokers, separate them with commas.
topic	Address of the Kafka topic. One sink can only correspond to one topic.
primary_key	Conditional. The primary keys of the sink. Use , to delimit the primary key columns. This field is optional if creating a PLAIN sink, but required if creating a DEBEZIUM or UPSERT sink.

Additional Kafka parameters

When creating a Kafka sink in RisingWave, you can specify the following Kafka-specific parameters. To set the parameter, add the RisingWave equivalent of the Kafka parameter as a WITH option. For additional details on these parameters, see the Configuration properties.

Kafka parameter name	RisingWave parameter name	Type
allow.auto.create.topics	properties.allow.auto.create.topics	bool
batch.num.messages	properties.batch.num.messages	int
batch.size	properties.batch.size	int
client.id	properties.client.id	string
enable.idempotence	properties.enable.idempotence	bool
enable.ssl.certificate.verification	properties.enable.ssl.certificate.verification	bool
max.in.flight.requests.per.connection	properties.max.in.flight.requests.per.connection	int
message.max.bytes	properties.message.max.bytes	int
message.send.max.retries	properties.message.send.max.retries	int
message.timeout.ms	properties.message.timeout.ms	int
queue.buffering.max.kbytes	properties.queue.buffering.max.kbytes	int
queue.buffering.max.messages	properties.queue.buffering.max.messages	int
queue.buffering.max.ms	properties.queue.buffering.max.ms	float
request.required.acks	properties.request.required.acks	int
retry.backoff.ms	properties.retry.backoff.ms	int
receive.message.max.bytes	properties.receive.message.max.bytes	int
ssl.endpoint.identification.algorithm	properties.ssl.endpoint.identification.algorithm	str
statistics.interval.ms	properties.statistics.interval.ms	int

Set properties.ssl.endpoint.identification.algorithm to none to bypass the verification of CA certificates and resolve SSL handshake failure. This parameter can be set to either https or none. By default, it is https.

To monitor Kafka metrics in Grafana, set properties.statistics.interval.ms to a non-zero value. The granularity is 1000ms.

Starting with version 2.0, the default value for properties.message.timeout.ms has changed from 5 seconds to 5 minutes, aligning with the default setting in the official Kafka library.

FORMAT and ENCODE options

These options should be set in FORMAT data_format ENCODE data_encode (key = 'value'), instead of the WITH clause.

Field	Notes
data_format	Data format. Allowed formats: PLAIN: Output data with insert operations. DEBEZIUM: Output change data capture (CDC) log in Debezium format. UPSERT: Output data as a changelog stream. primary_key must be specified in this case. To learn about when to define the primary key if creating an UPSERT sink, see the Overview.
data_encode	Data encode. Allowed encodes: AVRO: Supports UPSERT AVRO and PLAIN AVRO sinks. PROTOBUF: Supports PLAIN PROTOBUF and UPSERT PROTOBUF sinks. JSON: Supports PLAIN JSON, UPSERT JSON and DEBEZIUM JSON sinks. BYTES: Supports PLAIN BYTES sinks. The table must have exactly one column of type bytea. This encoding outputs the raw binary data directly to Kafka. For UPSERT PROTOBUF sinks, you must specify key encode text, while it remains optional for other format/encode combinations.
force_append_only	If true, forces the sink to be PLAIN (also known as append-only), even if it cannot be.
timestamptz.handling.mode	Controls the timestamptz output format. This parameter specifically applies to append-only or upsert sinks using JSON encoding. If omitted, the output format of timestamptz is 2023-11-11T18:30:09.453000Z which includes the UTC suffix Z. When utc_without_suffix is specified, the format is changed to 2023-11-11 18:30:09.453000.
schemas.enable	Only configurable for upsert JSON sinks. By default, this value is false for upsert JSON sinks and true for debezium JSON sinks. If true, RisingWave will sink the data with the schema to the Kafka sink. This is not referring to a schema registry containing a JSON schema, but rather schema formats defined using Kafka Connect.
key_encode	Optional. When specified, the key encode can only be TEXT or BYTES. If set to TEXT, the primary key should be one and only one of the following types: varchar, bool, smallint, int, and bigint; If set to BYTES, the primary key should be one and only one of type bytea; When absent, both key and value will use the same setting of ENCODE data_encode ( ... ).

Avro

When creating an Avro sink, the following options can be used following FORMAT UPSERT ENCODE AVRO or FORMAT PLAIN ENCODE AVRO.

Field	Notes
schema.registry	Required. The address of the schema registry.
schema.registry.username	Optional. The user name used to access the schema registry.
schema.registry.password	Optional. The password associated with the user name.
schema.registry.name.strategy	Optional. Accepted options include topic_name_strategy (default), record_name_strategy, and topic_record_name_strategy.
key.message	Required if schema.registry.name.strategy is set to record_name_strategy or topic_record_name_strategy.
message	Required if schema.registry.name.strategy is set to record_name_strategy or topic_record_name_strategy.

Syntax

FORMAT [ UPSERT | PLAIN ]
ENCODE AVRO (
   schema.registry = 'schema_registry_url',
   [schema.registry.username = 'username'],
   [schema.registry.password = 'password'],
   [schema.registry.name.strategy = 'topic_name_strategy'],
   [key.message = 'test_key'],
   [message = 'main_message',]
)

For data type mapping, the serial type is supported. We map the serial type to the 64-bit signed integer.

Use AWS Glue Schema Registry

PREMIUM FEATURE

This is a premium feature. For a comprehensive overview of all premium features and their usage, please see RisingWave premium features.

AWS Glue Schema Registry is a serverless feature of AWS Glue that allows you to centrally manage and enforce schemas for data streams, enabling data validation and compatibility checks. It helps in improving the quality of data streams by providing a central repository for managing and enforcing schemas across various AWS services and custom applications.

You can specify the following configurations in the FORMAT PLAIN ENCODE AVRO (...) or FORMAT UPSERT ENCODE AVRO (...) clause. This allows RisingWave to load schemas from and encode metadata for AWS Glue Schema Registry, in addition to Confluent Schema Registry.

Auth-related configurations:

Parameter	Description
aws.region	The region of the AWS Glue Schema Registry. For example, us-west-2.
aws.credentials.access_key_id	Your AWS access key ID.
aws.credentials.secret_access_key	Your AWS secret access key.
aws.credentials.role.arn	The Amazon Resource Name (ARN) of the role to assume. For example, arn:aws:iam::123456123456:role/MyGlueRole. This IAM role shall be granted permissions for the action glue:GetSchemaVersion.

In RisingWave Cloud, AWS credentials (aws.credentials.access_key_id, aws.credentials.secret_access_key) are required and cannot be omitted. In self-hosted deployments, you may omit them to rely on the AWS SDK default credential chain (for example, EC2 instance profile or environment variables).

ARN to the schema:

Parameter	Description
aws.glue.schema_arn	The ARN of the schema in AWS Glue Schema Registry. For example, 'arn:aws:glue:ap-southeast-1:123456123456:schema/default-registry/MyEvent'.

Example

ENCODE AVRO (
  aws.glue.schema_arn = 'arn:aws:glue:ap-southeast-1:123456123456:schema/default-registry/MyEvent',
  aws.region = 'US-WEST-2',
  aws.credentials.access_key_id = 'your_access_key_id',
  aws.credentials.secret_access_key = 'your_secret_access_key',
  aws.credentials.role.arn = 'arn:aws:iam::123456123456:role/MyGlueRole'
  ...
) [KEY ENCODE [TEXT | BYTES]] -- when specifying `primary_key` as Kafka key

Protobuf

When creating an append-only Protobuf sink, the following options can be used following FORMAT PLAIN ENCODE PROTOBUF or FORMAT UPSERT ENCODE PROTOBUF.

Field	Notes
message	Required. Package qualified message name of the main Message in the schema definition.
schema.location	Required if schema.registry is not specified. Only one of schema.location or schema.registry can be defined. The schema location. This can be in either file://, http://, https:// format.
schema.registry	Required if schema.location is not specified. Only one of schema.location or schema.registry can be defined. The address of the schema registry.
schema.registry.username	Optional. The user name used to access the schema registry.
schema.registry.password	Optional. The password associated with the user name.
schema.registry.name.strategy	Optional. Accepted options include topic_name_strategy (default), record_name_strategy, and topic_record_name_strategy.

The file:// format is not recommended for production use. If it is used, it needs to be available for both Meta and Compute Nodes.

When using Protobuf with Confluent Schema Registry, RisingWave bundles only protobuf well-known types, such as google/protobuf/timestamp.proto, duration.proto, any.proto, empty.proto, struct.proto, and wrappers. Other imports, including Google common types such as google/type/date.proto, google/type/latlng.proto, google/rpc/*, google/api/*, and confluent/meta.proto, must also be registered as schema references in Schema Registry so that RisingWave can encode them properly.

FORMAT as PLAIN

FORMAT PLAIN
ENCODE PROTOBUF (
   message = 'com.example.MyMessage',
   schema.location = 'location'
)

FORMAT as UPSERT

FORMAT UPSERT
ENCODE PROTOBUF (
   message = 'com.example.MyMessage',
   schema.location = 'location'
) KEY ENCODE TEXT

For data type mapping, the serial type is supported. We map the serial type to the 64-bit signed integer.

JSON

The jsonb.handling.mode determines how jsonb data types are encoded. This parameter has two possible values:

• string: Encodes the jsonb type to a string. For example, if you set this parameter, {"k": 2} will be converted to "{\"k\": 2}".
• dynamic: Dynamically encodes a jsonb type value to a JSON type value. For example, if you set this parameter, {"k": 2} will be converted to {"k": 2}. Here the jsonb value is encoded to a JSON object type value.

You can set this parameter in the WITH clause of ENCODE JSON.

For data mapping, the serial type is supported. However, note that it is mapped into a JSON string like "0x05fb93d677c4e000" instead of a JSON number 431100738685689856.

This string form avoids JSON number precision issues with large int64 values, and you can still order by the fixed-length hexadecimal string to obtain the same order as the serial number (whereas variable-length string "12" sorts before "7").

BYTES

The BYTES encoding allows you to sink raw binary data directly to Kafka topics. This is useful for scenarios where you need to output unencoded binary payloads.

Requirements:

• Only supported with FORMAT PLAIN
• The source table or materialized view must have exactly one column of type bytea
• force_append_only='true' must be specified

Example

CREATE TABLE t_binary_data (
    data bytea
);

CREATE SINK sink_bytes FROM t_binary_data WITH (
    connector = 'kafka',
    properties.bootstrap.server = 'localhost:9092',
    topic = 'binary-topic'
) FORMAT PLAIN ENCODE BYTES (
    force_append_only = 'true'
);

Examples

Create a sink by selecting an entire materialized view.

CREATE SINK sink1 FROM mv1
WITH (
   connector='kafka',
   properties.bootstrap.server='localhost:9092',
   topic='test'
)
FORMAT PLAIN ENCODE JSON;

Create a sink with the Kafka configuration message.max.bytes set at 2000 by setting properties.message.max.bytes to 2000.

CREATE SINK sink1 FROM mv1
WITH (
   connector='kafka',
   properties.bootstrap.server='localhost:9092',
   topic='test',
   properties.message.max.bytes = 2000
)
FORMAT PLAIN ENCODE JSON;

Create a sink by selecting the average distance and duration from taxi_trips.

The schema of taxi_trips is like this:

{
  "id": VARCHAR,
  "distance": DOUBLE PRECISION,
  "duration": DOUBLE PRECISION,
  "fare": DOUBLE PRECISION
}

The table may look like this:

 id | distance | duration |   city
----+----------+----------+----------
  1 |       16 |       23 | Dallas
  2 |       23 |        9 | New York
  3 |        6 |       15 | Chicago
  4 |        9 |       35 | New York

CREATE SINK sink2 AS
SELECT
   avg(distance) as avg_distance,
   avg(duration) as avg_duration
FROM taxi_trips
WITH (
   connector='kafka',
   properties.bootstrap.server='localhost:9092',
   topic='test'
)
FORMAT PLAIN ENCODE JSON;

Create sink with PrivateLink connection

If your Kafka sink service is located in a different VPC from RisingWave, use AWS PrivateLink or GCP Private Service Connect to establish a secure and direct connection. For details on how to set up an AWS PrivateLink connection, see Create an AWS PrivateLink connection.

To create a Kafka sink with a PrivateLink connection, in the WITH section of your CREATE SINK statement, specify the following parameters.

Parameter	Notes
privatelink.targets	The PrivateLink targets that correspond to the Kafka brokers. The targets should be in JSON format. Note that each target listed corresponds to each broker specified in the properties.bootstrap.server field. If the order is incorrect, there will be connectivity issues.
privatelink.endpoint	The DNS name of the VPC endpoint. If you’re using RisingWave Cloud, you can find the auto-generated endpoint after you created a connection. See details in Create a VPC connection.

Here is an example of creating a Kafka sink using a PrivateLink connection. Notice that {"port": 8001} corresponds to the broker ip1:9092, and {"port": 8002} corresponds to the broker ip2:9092.

CREATE SINK sink2 FROM mv2
WITH (
   connector='kafka',
   properties.bootstrap.server='b-1.xxx.amazonaws.com:9092,b-2.test.xxx.amazonaws.com:9092',
   topic='msk_topic',
   privatelink.endpoint='10.148.0.4',
   privatelink.targets = '[{"port": 8001}, {"port": 8002}]'
)
FORMAT PLAIN ENCODE JSON (
   force_append_only='true'
);

TLS/SSL encryption and SASL authentication

RisingWave can sink data to Kafka that is encrypted with Transport Layer Security (TLS) and/or authenticated with SASL.

Secure Sockets Layer (SSL) was the predecessor of Transport Layer Security (TLS), and has been deprecated since June 2015. For historical reasons, SSL is used in configuration and code instead of TLS.

Simple Authentication and Security Layer (SASL) is a framework for authentication and data security in Internet protocols.

RisingWave supports these SASL authentication mechanisms:

• SASL/PLAIN
• SASL/SCRAM
• SASL/GSSAPI
• SASL/OAUTHBEARER

SSL encryption can be used concurrently with SASL authentication mechanisms.

To learn about how to enable SSL encryption and SASL authentication in Kafka, including how to generate the keys and certificates, see the Security Tutorial from Confluent.

You need to specify encryption and authentication parameters in the WITH section of a CREATE SINK statement.

SSL without SASL

To sink data encrypted with SSL without SASL authentication, specify these parameters in the WITH section of your CREATE SINK statement.

Parameter	Notes
properties.security.protocol	Set to SSL.
properties.ssl.ca.location
properties.ssl.certificate.location
properties.ssl.key.location
properties.ssl.key.password

For the definitions of the parameters, see the librdkafka properties list. Note that the parameters in the list assumes all parameters start with properties. and therefore do not include this prefix.

Here is an example of creating a sink encrypted with SSL without using SASL authentication.

CREATE SINK sink1 FROM mv1
WITH (
   connector='kafka',
   topic='quickstart-events',
   properties.bootstrap.server='localhost:9093',
   properties.security.protocol='SSL',
   properties.ssl.ca.location='/home/ubuntu/kafka/secrets/ca-cert',
   properties.ssl.certificate.location='/home/ubuntu/kafka/secrets/client_risingwave_client.pem',
   properties.ssl.key.location='/home/ubuntu/kafka/secrets/client_risingwave_client.key',
   properties.ssl.key.password='abcdefgh'
)
FORMAT PLAIN ENCODE JSON;

SASL/PLAIN

Parameter	Notes
properties.security.protocol	For SASL/PLAIN without SSL, set to SASL_PLAINTEXT. For SASL/PLAIN with SSL, set to SASL_SSL.
properties.sasl.mechanism	Set to PLAIN.
properties.sasl.username
properties.sasl.password

For the definitions of the parameters, see the librdkafka properties list. Note that the parameters in the list assumes all parameters start with properties. and therefore do not include this prefix.

For SASL/PLAIN with SSL, you need to include these SSL parameters:

• properties.ssl.ca.location
• properties.ssl.certificate.location
• properties.ssl.key.location
• properties.ssl.key.password

Here is an example of creating a sink authenticated with SASL/PLAIN without SSL encryption.

CREATE SINK sink1 FROM mv1
WITH (
   connector='kafka',
   topic='quickstart-events',
   properties.bootstrap.server='localhost:9093',
   properties.sasl.mechanism='PLAIN',
   properties.security.protocol='SASL_PLAINTEXT',
   properties.sasl.username='admin',
   properties.sasl.password='admin-secret'
)
FORMAT PLAIN ENCODE JSON;

This is an example of creating a sink authenticated with SASL/PLAIN with SSL encryption.

CREATE SINK sink1 FROM mv1
WITH (
   connector='kafka',
   topic='quickstart-events',
   properties.bootstrap.server='localhost:9093',
   properties.sasl.mechanism='PLAIN',
   properties.security.protocol='SASL_SSL',
   properties.sasl.username='admin',
   properties.sasl.password='admin-secret',
   properties.ssl.ca.location='/home/ubuntu/kafka/secrets/ca-cert',
   properties.ssl.certificate.location='/home/ubuntu/kafka/secrets/client_risingwave_client.pem',
   properties.ssl.key.location='/home/ubuntu/kafka/secrets/client_risingwave_client.key',
   properties.ssl.key.password='abcdefgh'
)
FORMAT PLAIN ENCODE JSON;

SASL/SCRAM

Parameter	Notes
properties.security.protocol	For SASL/SCRAM without SSL, set to SASL_PLAINTEXT. For SASL/SCRAM with SSL, set to SASL_SSL.
properties.sasl.mechanism	Set to SCRAM-SHA-256 or SCRAM-SHA-512 depending on the encryption method used.
properties.sasl.username
properties.sasl.password

For the definitions of the parameters, see the librdkafka properties list. Note that the parameters in the list assumes all parameters start with properties. and therefore do not include this prefix.

For SASL/SCRAM with SSL, you also need to include these SSL parameters:

• properties.ssl.ca.location
• properties.ssl.certificate.location
• properties.ssl.key.location
• properties.ssl.key.password

Here is an example of creating a sink authenticated with SASL/SCRAM without SSL encryption.

CREATE SINK sink1 FROM mv1
WITH (
   connector='kafka',
   topic='quickstart-events',
   properties.bootstrap.server='localhost:9093',
   properties.sasl.mechanism='SCRAM-SHA-256',
   properties.security.protocol='SASL_PLAINTEXT',
   properties.sasl.username='admin',
   properties.sasl.password='admin-secret'
)
FORMAT PLAIN ENCODE JSON;

SASL/GSSAPI

Parameter	Notes
properties.security.protocol	Set to SASL_PLAINTEXT, as RisingWave does not support using SASL/GSSAPI with SSL.
properties.sasl.mechanism	Set to GSSAPI.
properties.sasl.kerberos.service.name
properties.sasl.kerberos.keytab
properties.sasl.kerberos.principal
properties.sasl.kerberos.kinit.cmd
properties.sasl.kerberos.min.time.before.relogin

For the definitions of the parameters, see the librdkafka properties list. Note that the parameters in the list assumes all parameters start with properties. and therefore do not include this prefix.

Here is an example of creating a sink authenticated with SASL/GSSAPI without SSL encryption.

CREATE SINK sink1 FROM mv1
WITH (
   connector='kafka',
   type = 'append-only',
   topic='quickstart-events',
   properties.bootstrap.server='localhost:9093',
   properties.sasl.mechanism='GSSAPI',
   properties.security.protocol='SASL_PLAINTEXT',
   properties.sasl.kerberos.service.name='kafka',
   properties.sasl.kerberos.keytab='/etc/krb5kdc/kafka.client.keytab',
   properties.sasl.kerberos.principal='[e-mail]',
   properties.sasl.kerberos.kinit.cmd='sudo kinit -R -kt "%{sasl.kerberos.keytab}" %{sasl.kerberos.principal} || sudo kinit -kt "%{sasl.kerberos.keytab}" %{sasl.kerberos.principal}',
   properties.sasl.kerberos.min.time.before.relogin='10000'
) FORMAT PLAIN ENCODE JSON;

SASL/OAUTHBEARER

The implementation of SASL/OAUTHBEARER in RisingWave validates only unsecured client side tokens, and does not support OpenID Connect (OIDC) authentication. Therefore, it should not be used in production environments.

Parameter	Notes
properties.security.protocol	For SASL/OAUTHBEARER without SSL, set to SASL_PLAINTEXT. For SASL/OAUTHBEARER with SSL, set to SASL_SSL.
properties.sasl.mechanism	Set to OAUTHBEARER.
properties.sasl.oauthbearer.config

For the definitions of the parameters, see the librdkafka properties list. Note that the parameters in the list assumes all parameters start with properties. and therefore do not include this prefix. Also, due to the limitation of the SASL/OAUTHBEARER implementation, you only need to specify one OAUTHBEARER parameter: properties.sasl.oauthbearer.config. Other OAUTHBEARER parameters are not applicable.

For SASL/OAUTHBEARER with SSL, you also need to include these SSL parameters:

• properties.ssl.ca.location
• properties.ssl.certificate.location
• properties.ssl.key.location
• properties.ssl.key.password

This is an example of creating a sink authenticated with SASL/OAUTHBEARER without SSL encryption.

CREATE SINK sink1 FROM mv1
WITH (
   connector='kafka',
   type = 'append-only',
   topic='quickstart-events',
   properties.bootstrap.server='localhost:9093',
   properties.sasl.mechanism='OAUTHBEARER',
   properties.security.protocol='SASL_PLAINTEXT',
   properties.sasl.oauthbearer.config='principal=bob'
) FORMAT PLAIN ENCODE JSON;

Data type mapping - RisingWave and Debezium JSON

RisingWave Data Type	Schema Type in JSON	Schema Name in JSON
boolean	boolean	n/a
smallint	int16	n/a
integer	int32	n/a
bigint	int64	n/a
real	float	n/a
double precision	double	n/a
character varying	string	n/a
bytea	bytes	n/a
numeric	string	n/a
date	int32	org.apache.kafka.connect.data.Date
time without time zone	int64	org.apache.kafka.connect.data.Time
timestamp	int64	org.apache.kafka.connect.data.Timestamp
timestamptz	string	io.debezium.time.ZonedTimestamp
interval	string	io.debezium.time.Interval
JSONB	string	io.debezium.data.Json
struct	string	n/a
array	string	n/a