Skip to content

Latest commit

 

History

History
197 lines (151 loc) · 12.1 KB

Kafka.md

File metadata and controls

197 lines (151 loc) · 12.1 KB
Raw

Kafka

Kafka sink connector

Support Those Engines

Spark
Flink
Seatunnel Zeta

Key Features

By default, we will use 2pc to guarantee the message is sent to kafka exactly once.

Description

Write Rows to a Kafka topic.

Supported DataSource Info

In order to use the Kafka connector, the following dependencies are required. They can be downloaded via install-plugin.sh or from the Maven central repository.

Datasource Supported Versions Maven
Kafka Universal Download

Sink Options

Name Type Required Default Description
topic String Yes - When the table is used as sink, the topic name is the topic to write data to.
bootstrap.servers String Yes - Comma separated list of Kafka brokers.
kafka.config Map No - In addition to the above parameters that must be specified by the Kafka producer client, the user can also specify multiple non-mandatory parameters for the producer client, covering all the producer parameters specified in the official Kafka document.
semantics String No NON Semantics that can be chosen EXACTLY_ONCE/AT_LEAST_ONCE/NON, default NON.
partition_key_fields Array No - Configure which fields are used as the key of the kafka message.
partition Int No - We can specify the partition, all messages will be sent to this partition.
assign_partitions Array No - We can decide which partition to send based on the content of the message. The function of this parameter is to distribute information.
transaction_prefix String No - If semantic is specified as EXACTLY_ONCE, the producer will write all messages in a Kafka transaction,kafka distinguishes different transactions by different transactionId. This parameter is prefix of kafka transactionId, make sure different job use different prefix.
format String No json Data format. The default format is json. Optional text format, canal-json and debezium-json.If you use json or text format. The default field separator is ", ". If you customize the delimiter, add the "field_delimiter" option.If you use canal format, please refer to canal-json for details.If you use debezium format, please refer to debezium-json for details.
field_delimiter String No , Customize the field delimiter for data format.
common-options No - Source plugin common parameters, please refer to Source Common Options for details

Parameter Interpretation

Topic Formats

Currently two formats are supported:

  1. Fill in the name of the topic.

  2. Use value of a field from upstream data as topic,the format is ${your field name}, where topic is the value of one of the columns of the upstream data.

    For example, Upstream data is the following:

    name age data
    Jack 16 data-example1
    Mary 23 data-example2

    If ${name} is set as the topic. So the first row is sent to Jack topic, and the second row is sent to Mary topic.

Semantics

In EXACTLY_ONCE, producer will write all messages in a Kafka transaction that will be committed to Kafka on a checkpoint. In AT_LEAST_ONCE, producer will wait for all outstanding messages in the Kafka buffers to be acknowledged by the Kafka producer on a checkpoint. NON does not provide any guarantees: messages may be lost in case of issues on the Kafka broker and messages may be duplicated.

Partition Key Fields

For example, if you want to use value of fields from upstream data as key, you can assign field names to this property.

Upstream data is the following:

name age data
Jack 16 data-example1
Mary 23 data-example2

If name is set as the key, then the hash value of the name column will determine which partition the message is sent to. If not set partition key fields, the null message key will be sent to. The format of the message key is json, If name is set as the key, for example '{"name":"Jack"}'. The selected field must be an existing field in the upstream.

Assign Partitions

For example, there are five partitions in total, and the assign_partitions field in config is as follows: assign_partitions = ["shoe", "clothing"] Then the message containing "shoe" will be sent to partition zero ,because "shoe" is subscribed as zero in assign_partitions, and the message containing "clothing" will be sent to partition one.For other messages, the hash algorithm will be used to divide them into the remaining partitions. This function by MessageContentPartitioner class implements org.apache.kafka.clients.producer.Partitioner interface.If we need custom partitions, we need to implement this interface as well.

Task Example

Simple:

This example defines a SeaTunnel synchronization task that automatically generates data through FakeSource and sends it to Kafka Sink. FakeSource generates a total of 16 rows of data (row.num=16), with each row having two fields, name (string type) and age (int type). The final target topic is test_topic will also be 16 rows of data in the topic. And if you have not yet installed and deployed SeaTunnel, you need to follow the instructions in Install SeaTunnel to install and deploy SeaTunnel. And then follow the instructions in Quick Start With SeaTunnel Engine to run this job.

# Defining the runtime environment
env {
  # You can set flink configuration here
  execution.parallelism = 1
  job.mode = "BATCH"
}

source {
  FakeSource {
    parallelism = 1
    result_table_name = "fake"
    row.num = 16
    schema = {
      fields {
        name = "string"
        age = "int"
      }
    }
  }
}

sink {
  kafka {
      topic = "test_topic"
      bootstrap.servers = "localhost:9092"
      partition = 3
      format = json
      kafka.request.timeout.ms = 60000
      semantics = EXACTLY_ONCE
      kafka.config = {
        acks = "all"
        request.timeout.ms = 60000
        buffer.memory = 33554432
      }
  }
}

AWS MSK SASL/SCRAM

Replace the following ${username} and ${password} with the configuration values in AWS MSK.

sink {
  kafka {
      topic = "seatunnel"
      bootstrap.servers = "localhost:9092"
      partition = 3
      format = json
      kafka.request.timeout.ms = 60000
      semantics = EXACTLY_ONCE
      kafka.config = {
         security.protocol=SASL_SSL
         sasl.mechanism=SCRAM-SHA-512
         sasl.jaas.config="org.apache.kafka.common.security.scram.ScramLoginModule required \nusername=${username}\npassword=${password};"
      }
  }
}

AWS MSK IAM

Download aws-msk-iam-auth-1.1.5.jar from https://github.com/aws/aws-msk-iam-auth/releases and put it in $SEATUNNEL_HOME/plugin/kafka/lib dir.

Please ensure the IAM policy have "kafka-cluster:Connect",. Like this:

"Effect": "Allow",
"Action": [
    "kafka-cluster:Connect",
    "kafka-cluster:AlterCluster",
    "kafka-cluster:DescribeCluster"
],

Sink Config

sink {
  kafka {
      topic = "seatunnel"
      bootstrap.servers = "localhost:9092"
      partition = 3
      format = json
      kafka.request.timeout.ms = 60000
      semantics = EXACTLY_ONCE
      kafka.config = {
         security.protocol=SASL_SSL
         sasl.mechanism=AWS_MSK_IAM
         sasl.jaas.config="software.amazon.msk.auth.iam.IAMLoginModule required;"
         sasl.client.callback.handler.class="software.amazon.msk.auth.iam.IAMClientCallbackHandler"
      }
  }
}