Using Apache Kafka as a target for AWS Database Migration Service - AWS Database Migration Service

Using Apache Kafka as a target for AWS Database Migration Service

You can use AWS DMS to migrate data to an Apache Kafka cluster. Apache Kafka is a distributed streaming platform. You can use Apache Kafka for ingesting and processing streaming data in real-time.

AWS also offers Amazon Managed Streaming for Apache Kafka (Amazon MSK) to use as an AWS DMS target. Amazon MSK is a fully managed Apache Kafka streaming service that simplifies the implementation and management of Apache Kafka instances. It works with open-source Apache Kafka versions, and you access Amazon MSK instances as AWS DMS targets exactly like any Apache Kafka instance. For more information, see What is Amazon MSK? in the Amazon Managed Streaming for Apache Kafka Developer Guide.

A Kafka cluster stores streams of records in categories called topics that are divided into partitions. Partitions are uniquely identified sequences of data records (messages) in a topic. Partitions can be distributed across multiple brokers in a cluster to enable parallel processing of a topic’s records. For more information on topics and partitions and their distribution in Apache Kafka, see Topics and logs and Distribution.

AWS Database Migration Service publishes records to a Kafka topic using JSON. During conversion, AWS DMS serializes each record from the source database into an attribute-value pair in JSON format.

To migrate your data from any supported data source to a target Kafka cluster, you use object mapping. With object mapping, you determine how to structure the data records in the target topic. You also define a partition key for each table, which Apache Kafka uses to group the data into its partitions.

When AWS DMS creates tables on an Apache Kafka target endpoint, it creates as many tables as in the source database endpoint. AWS DMS also sets several Apache Kafka parameter values. The cost for the table creation depends on the amount of data and the number of tables to be migrated.

Apache Kafka endpoint settings

You can specify connection details through endpoint settings in the AWS DMS console, or the --kafka-settings option in the CLI. The requirements for each setting follow:

  • Broker – Specify the broker location in the form broker-hostname:port. For example, "". This can be the location of any broker in the cluster. The cluster brokers all communicate to handle the partitioning of data records migrated to the topic.

  • Topic – (Optional) Specify the topic name with a maximum length of 255 letters and symbols. You can use period (.), underscore (_), and minus (-). Topic names with a period (.) or underscore (_) can collide in internal data structures. Use either one, but not both of these symbols in the topic name. If you don't specify a topic name, AWS DMS uses "kafka-default-topic" as the migration topic.


    To have AWS DMS create either a migration topic you specify or the default topic, set auto.create.topics.enable = true as part of your Kafka cluster configuration. For more information, see Limitations when using Apache Kafka as a target for AWS Database Migration Service

  • MessageFormat – The output format for the records created on the endpoint. The message format is JSON (default) or JSON_UNFORMATTED (a single line with no tab).

  • MessageMaxBytes – The maximum size in bytes for records created on the endpoint. The default is 1,000,000.

  • IncludeTransactionDetails – Provides detailed transaction information from the source database. This information includes a commit timestamp, a log position, and values for transaction_id, previous_transaction_id, and transaction_record_id (the record offset within a transaction). The default is false.

  • IncludePartitionValue – Shows the partition value within the Kafka message output, unless the partition type is schema-table-type. The default is false.

  • PartitionIncludeSchemaTable – Prefixes schema and table names to partition values, when the partition type is primary-key-type. Doing this increases data distribution among Kafka partitions. For example, suppose that a SysBench schema has thousands of tables and each table has only limited range for a primary key. In this case, the same primary key is sent from thousands of tables to the same partition, which causes throttling. The default is false.

  • IncludeTableAlterOperations – Includes any data definition language (DDL) operations that change the table in the control data, such as rename-table, drop-table, add-column, drop-column, and rename-column. The default is false.

  • IncludeControlDetails – Shows detailed control information for table definition, column definition, and table and column changes in the Kafka message output. The default is false.

  • IncludeNullAndEmpty – Include NULL and empty columns in the target. The default is false.

Multithreaded full load task settings

You can use settings to help increase the speed of your transfer. To do so, AWS DMS supports a multithreaded full load to an Apache Kafka target cluster. AWS DMS supports this multithreading with task settings that include the following:

  • MaxFullLoadSubTasks – Use this option to indicate the maximum number of source tables to load in parallel. AWS DMS loads each table into its corresponding Kafka target table using a dedicated subtask. The default is 8; the maximum value is 49.

  • ParallelLoadThreads – Use this option to specify the number of threads that AWS DMS uses to load each table into its Kafka target table. The maximum value for an Apache Kafka target is 32. You can ask to have this maximum limit increased.

  • ParallelLoadBufferSize – Use this option to specify the maximum number of records to store in the buffer that the parallel load threads use to load data to the Kafka target. The default value is 50. The maximum value is 1,000. Use this setting with ParallelLoadThreads. ParallelLoadBufferSize is valid only when there is more than one thread.

  • ParallelLoadQueuesPerThread – Use this option to specify the number of queues each concurrent thread accesses to take data records out of queues and generate a batch load for the target. The default is 1. However, for Kafka targets of various payload sizes, the valid range is 5–512 queues per thread.

  • Table-mapping settings for individual tables – Use table-settings rules to identify the individual tables from the source that you want to load in parallel. Also use these rules to specify how to segment the rows of each table for multithreaded loading. For more information, see Table-settings rules and operations.


    DMS assigns each segment of a table to its own thread for loading. Therefore, set ParallelLoadThreads to the maximum number of segments that you specify for a table in the source.

Multithreaded CDC load task settings

You can improve the performance of change data capture (CDC) for real-time data streaming target endpoints like Kafka using task settings to modify the behaviour of the PutRecords API call. To do this, you can specify the number of concurrent threads, queues per thread, and the number of records to store in a buffer using ParallelApply* task settings. For example, suppose you want to perform a CDC load and apply 128 threads in parallel. You also want to access 64 queues per thread, with 50 records stored per buffer.

To promote CDC performance, AWS DMS supports these task settings:

  • ParallelApplyThreads – Specifies the number of concurrent threads that AWS DMS uses during a CDC load to push data records to a Kafka target endpoint. The default value is zero (0) and the maximum value is 32.

  • ParallelApplyBufferSize – Specifies the maximum number of records to store in each buffer queue for concurrent threads to push to a Kafka target endpoint during a CDC load. The default value is 50 and the maximum value is 1,000. Use this option when ParallelApplyThreads specifies more than one thread.

  • ParallelApplyQueuesPerThread – Specifies the number of queues that each thread accesses to take data records out of queues and generate a batch load for a Kafka endpoint during CDC.

When using ParallelApply* task settings, the partition-key-type default is the primary-key of the table, not schema-name.table-name.

Using a before image to view original values of CDC rows for Apache Kafka as a target

When writing CDC updates to a data-streaming target like Kafka you can view a source database row's original values before change by an update. To make this possible, AWS DMS populates a before image of update events based on data supplied by the source database engine.

Different source database engines provide different amounts of information for a before image:

  • Oracle provides updates to columns only if they change.

  • PostgreSQL provides only data for columns that are part of the primary key (changed or not).

  • MySQL generally provides data for all columns (changed or not).

To enable before imaging to add original values from the source database to the AWS DMS output, use either the BeforeImageSettings task setting or the add-before-image-columns parameter. This parameter applies a column transformation rule.

BeforeImageSettings adds a new JSON attribute to every update operation with values collected from the source database system, as shown following.

"BeforeImageSettings": { "EnableBeforeImage": boolean, "FieldName": string, "ColumnFilter": pk-only (default) / non-lob / all (but only one) }

Apply BeforeImageSettings to full load plus CDC tasks (which migrate existing data and replicate ongoing changes), or to CDC only tasks (which replicate data changes only). Don't apply BeforeImageSettings to tasks that are full load only.

For BeforeImageSettings options, the following applies:

  • Set the EnableBeforeImage option to true to enable before imaging. The default is false.

  • Use the FieldName option to assign a name to the new JSON attribute. When EnableBeforeImage is true, FieldName is required and can't be empty.

  • The ColumnFilter option specifies a column to add by using before imaging. To add only columns that are part of the table's primary keys, use the default value, pk-only. To add only columns that are not of LOB type, use non-lob. To add any column that has a before image value, use all.

    "BeforeImageSettings": { "EnableBeforeImage": true, "FieldName": "before-image", "ColumnFilter": "pk-only" }

Using a before image transformation rule

As as an alternative to task settings, you can use the add-before-image-columns parameter, which applies a column transformation rule. With this parameter, you can enable before imaging during CDC on data streaming targets like Kafka.

By using add-before-image-columns in a transformation rule, you can apply more fine-grained control of the before image results. Transformation rules enable you to use an object locator that gives you control over tables selected for the rule. Also, you can chain transformation rules together, which allows different rules to be applied to different tables. You can then manipulate the columns produced by using other rules.


Don't use the add-before-image-columns parameter together with the BeforeImageSettings task setting within the same task. Instead, use either the parameter or the setting, but not both, for a single task.

A transformation rule type with the add-before-image-columns parameter for a column must provide a before-image-def section. The following shows an example.

{ "rule-type": "transformation", … "rule-target": "column", "rule-action": "add-before-image-columns", "before-image-def":{ "column-filter": one-of (pk-only / non-lob / all), "column-prefix": string, "column-suffix": string, } }

The value of column-prefix is prepended to a column name, and the default value of column-prefix is BI_. The value of column-suffix is appended to the column name, and the default is empty. Don't set both column-prefix and column-suffix to empty strings.

Choose one value for column-filter. To add only columns that are part of table primary keys, choose pk-only . Choose non-lob to only add columns that are not of LOB type. Or choose all to add any column that has a before-image value.

Example for a before image transformation rule

The transformation rule in the following example adds a new column called BI_emp_no in the target. So a statement like UPDATE employees SET emp_no = 3 WHERE emp_no = 1; populates the BI_emp_no field with 1. When you write CDC updates to Amazon S3 targets, the BI_emp_no column makes it possible to tell which original row was updated.

{ "rules": [ { "rule-type": "selection", "rule-id": "1", "rule-name": "1", "object-locator": { "schema-name": "%", "table-name": "%" }, "rule-action": "include" }, { "rule-type": "transformation", "rule-id": "2", "rule-name": "2", "rule-target": "column", "object-locator": { "schema-name": "%", "table-name": "employees" }, "rule-action": "add-before-image-columns", "before-image-def": { "column-prefix": "BI_", "column-suffix": "", "column-filter": "pk-only" } } ] }

For information on using the add-before-image-columns rule action, see Transformation rules and actions.

Limitations when using Apache Kafka as a target for AWS Database Migration Service

The following limitations apply when using Apache Kafka as a target:

  • AWS DMS supports a maximum message size of 1 MiB for a Kafka target.

  • Configure both your AWS DMS replication instance and your Kafka cluster in the same virtual private cloud (VPC) based on Amazon VPC and in the same security group. The Kafka cluster can either be an Amazon MSK instance or your own Kafka instance running on Amazon EC2. For more information, see Setting up a network for a replication instance.


    To specify a security group for Amazon MSK, on the Create cluster page, choose Advanced settings, select Customize settings, and select the security group or accept the default if it is the same as for your replication instance.

  • Specify a Kafka configuration file for your cluster with properties that allow AWS DMS to automatically create new topics. Include the setting, auto.create.topics.enable = true. If you are using Amazon MSK, you can specify the default configuration when you create your Kafka cluster, then change the auto.create.topics.enable setting to true. For more information about the default configuration settings, see The default Amazon MSK configuration in the Amazon Managed Streaming for Apache Kafka Developer Guide. If you need to modify an existing Kafka cluster created using Amazon MSK, run the AWS CLI command aws kafka create-configuration to update your Kafka configuration, as in the example following:

    14:38:41 $ aws kafka create-configuration --name "kafka-configuration" --kafka-versions "2.2.1" --server-properties file://~/kafka_configuration { "LatestRevision": { "Revision": 1, "CreationTime": "2019-09-06T14:39:37.708Z" }, "CreationTime": "2019-09-06T14:39:37.708Z", "Name": "kafka-configuration", "Arn": "arn:aws:kafka:us-east-1:111122223333:configuration/kafka-configuration/7e008070-6a08-445f-9fe5-36ccf630ecfd-3" }

    Here, //~/kafka_configuration is the configuration file you have created with the required property settings.

    If you are using your own Kafka instance installed on Amazon EC2, modify the Kafka cluster configuration with similar property settings, including auto.create.topics.enable = true, using the options provided with your instance.

  • AWS DMS publishes each update to a single record in the source database as one data record (message) in a given Kafka topic regardless of transactions.

  • AWS DMS supports the following two forms for partition keys:

    • SchemaName.TableName: A combination of the schema and table name.

    • ${AttributeName}: The value of one of the fields in the JSON, or the primary key of the table in the source database.

Using object mapping to migrate data to a Kafka topic

AWS DMS uses table-mapping rules to map data from the source to the target Kafka topic. To map data to a target topic, you use a type of table-mapping rule called object mapping. You use object mapping to define how data records in the source map to the data records published to a Kafka topic.

Kafka topics don't have a preset structure other than having a partition key.

To create an object-mapping rule, specify rule-type as object-mapping. This rule specifies what type of object mapping you want to use.

The structure for the rule is as follows.

{ "rules": [ { "rule-type": "object-mapping", "rule-id": "id", "rule-name": "name", "rule-action": "valid object-mapping rule action", "object-locator": { "schema-name": "case-sensitive schema name", "table-name": "" } } ] }

AWS DMS currently supports map-record-to-record and map-record-to-document as the only valid values for the rule-action parameter. The map-record-to-record and map-record-to-document values specify what AWS DMS does by default to records that aren't excluded as part of the exclude-columns attribute list. These values don't affect the attribute mappings in any way.

Use map-record-to-record when migrating from a relational database to a Kafka topic. This rule type uses the taskResourceId.schemaName.tableName value from the relational database as the partition key in the Kafka topic and creates an attribute for each column in the source database. When using map-record-to-record, for any column in the source table not listed in the exclude-columns attribute list, AWS DMS creates a corresponding attribute in the target topic. This corresponding attribute is created regardless of whether that source column is used in an attribute mapping.

One way to understand map-record-to-record is to see it in action. For this example, assume that you are starting with a relational database table row with the following structure and data.

FirstName LastName StoreId HomeAddress HomePhone WorkAddress WorkPhone DateofBirth




221B Baker Street


31 Spooner Street, Quahog



To migrate this information from a schema named Test to a Kafka topic, you create rules to map the data to the target topic. The following rule illustrates the mapping.

{ "rules": [ { "rule-type": "selection", "rule-id": "1", "rule-name": "1", "rule-action": "include", "object-locator": { "schema-name": "Test", "table-name": "%" } }, { "rule-type": "object-mapping", "rule-id": "2", "rule-name": "DefaultMapToKafka", "rule-action": "map-record-to-record", "object-locator": { "schema-name": "Test", "table-name": "Customers" } } ] }

Given a Kafka topic and a partition key (in this case, taskResourceId.schemaName.tableName), the following illustrates the resulting record format using our sample data in the Kafka target topic:

{ "FirstName": "Randy", "LastName": "Marsh", "StoreId": "5", "HomeAddress": "221B Baker Street", "HomePhone": "1234567890", "WorkAddress": "31 Spooner Street, Quahog", "WorkPhone": "9876543210", "DateOfBirth": "02/29/1988" }

Restructuring data with attribute mapping

You can restructure the data while you are migrating it to a Kafka topic using an attribute map. For example, you might want to combine several fields in the source into a single field in the target. The following attribute map illustrates how to restructure the data.

{ "rules": [ { "rule-type": "selection", "rule-id": "1", "rule-name": "1", "rule-action": "include", "object-locator": { "schema-name": "Test", "table-name": "%" } }, { "rule-type": "object-mapping", "rule-id": "2", "rule-name": "TransformToKafka", "rule-action": "map-record-to-record", "target-table-name": "CustomerData", "object-locator": { "schema-name": "Test", "table-name": "Customers" }, "mapping-parameters": { "partition-key-type": "attribute-name", "partition-key-name": "CustomerName", "exclude-columns": [ "firstname", "lastname", "homeaddress", "homephone", "workaddress", "workphone" ], "attribute-mappings": [ { "target-attribute-name": "CustomerName", "attribute-type": "scalar", "attribute-sub-type": "string", "value": "${lastname}, ${firstname}" }, { "target-attribute-name": "ContactDetails", "attribute-type": "document", "attribute-sub-type": "json", "value": { "Home": { "Address": "${homeaddress}", "Phone": "${homephone}" }, "Work": { "Address": "${workaddress}", "Phone": "${workphone}" } } } ] } } ] }

To set a constant value for partition-key, specify a partition-key value. For example, you might do this to force all the data to be stored in a single partition. The following mapping illustrates this approach.

{ "rules": [ { "rule-type": "selection", "rule-id": "1", "rule-name": "1", "object-locator": { "schema-name": "Test", "table-name": "%" }, "rule-action": "include" }, { "rule-type": "object-mapping", "rule-id": "1", "rule-name": "TransformToKafka", "rule-action": "map-record-to-document", "object-locator": { "schema-name": "Test", "table-name": "Customer" }, "mapping-parameters": { "partition-key": { "value": "ConstantPartitionKey" }, "exclude-columns": [ "FirstName", "LastName", "HomeAddress", "HomePhone", "WorkAddress", "WorkPhone" ], "attribute-mappings": [ { "attribute-name": "CustomerName", "value": "${FirstName},${LastName}" }, { "attribute-name": "ContactDetails", "value": { "Home": { "Address": "${HomeAddress}", "Phone": "${HomePhone}" }, "Work": { "Address": "${WorkAddress}", "Phone": "${WorkPhone}" } } }, { "attribute-name": "DateOfBirth", "value": "${DateOfBirth}" } ] } } ] }

The partition-key value for a control record that is for a specific table is TaskId.SchemaName.TableName. The partition-key value for a control record that is for a specific task is that record's TaskId. Specifying a partition-key value in the object mapping has no impact on the partition-key for a control record.

Message format for Apache Kafka

The JSON output is simply a list of key-value pairs.


The record type can be either data or control. Data records represent the actual rows in the source. Control records are for important events in the stream, for example a restart of the task.


For data records, the operation can be create, read, update, or delete.

For control records, the operation can be TruncateTable or DropTable.


The source schema for the record. This field can be empty for a control record.


The source table for the record. This field can be empty for a control record.


The timestamp for when the JSON message was constructed. The field is formatted with the ISO 8601 format.