Get started EARLY ACCESS

Get set up for using CDC in YugabyteDB

To stream data change events from YugabyteDB databases, you need to use Debezium YugabyteDB connector. To deploy a Debezium YugabyteDB connector, you install the Debezium YugabyteDB connector archive, configure the connector, and start the connector by adding its configuration to Kafka Connect. You can download the connector from GitHub releases. The connector supports Kafka Connect version 2.x and later, and for YugabyteDB, it supports version 2.14 and later. For more connector configuration details and complete steps, refer to Debezium connector.

Ordering guarantees

Ordering guarantee Description
Per-tablet ordered delivery guarantee All changes for a row (or rows in the same tablet) are received in the order in which they happened. However, due to the distributed nature of the problem, there is no guarantee of the order across tablets.
At least once delivery Updates for rows are streamed at least once. This can happen in the case of Kafka Connect Node failure. If the Kafka Connect Node pushes the records to Kafka and crashes before committing the offset, on restart, it will again get the same set of records.
No gaps in change stream Note that after you have received a change for a row for some timestamp t, you won't receive a previously unseen change for that row at a lower timestamp. Receiving any change implies that you have received all older changes for that row.

Set up YugabyteDB for CDC

The following steps are necessary to set up YugabyteDB for use with the Debezium YugabyteDB connector:

  • Create a DB stream ID.

    Before you use the YugabyteDB connector to retrieve data change events from YugabyteDB, create a stream ID using the yb-admin CLI command. Refer to the yb-admin CDC command reference documentation for more details.

  • Make sure the YB-Master and YB-TServer ports are open.

    The connector connects to the YB-Master and YB-TServer processes running on the YugabyteDB server. Make sure the ports on which these processes are running are open. The default ports on which the processes run are 7100 and 9100 respectively.

  • Monitor available disk space.

    The change records for CDC are read from the WAL. YugabyteDB CDC maintains checkpoints internally for each DB stream ID and garbage collects the WAL entries if those have been streamed to the CDC clients.

    In case CDC is lagging or away for some time, the disk usage may grow and cause YugabyteDB cluster instability. To avoid this scenario, if a stream is inactive for a configured amount of time, the WAL is garbage collected. This is configurable using a YB-TServer flag.

Serialization

The YugabyteDB source connector also supports AVRO serialization with schema registry. To use AVRO serialization, add the following configuration to your connector:

{
  ...
  "key.converter":"io.confluent.connect.avro.AvroConverter",
  "key.converter.schema.registry.url":"http://host-url-for-schema-registry:8081",
  "value.converter":"io.confluent.connect.avro.AvroConverter",
  "value.converter.schema.registry.url":"http://host-url-for-schema-registry:8081"
  ...
}

For JSON schema serialization, you can use the Kafka JSON Serializer and equivalent deserializer. After downloading and including the required JAR file in the Kafka-Connect environment, you can directly configure the CDC source and sink connectors to use this converter.

For source connectors:

{
  ...
  "value.serializer":"io.confluent.kafka.serializers.KafkaJsonSerializer",
  ...
}

For sink connectors:

{
  ...
  "value.deserializer":"io.confluent.kafka.serializers.KafkaJsonDeserializer",
  ...
}

To use the protobuf format for the serialization/de-serialization of the Kafka messages, you can use the Protobuf Converter. After downloading and including the required JAR files in the Kafka-Connect environment, you can directly configure the CDC source and sink connectors to use this converter.

{
  ...,
  config: {
    ...,
     "key.converter": "io.confluent.connect.protobuf.ProtobufConverter",
     "value.converter": "io.confluent.connect.protobuf.ProtobufConverter"
  }
}

Before image

Before image refers to the state of the row before the change event occurred. The YugabyteDB connector sends the before image of the row when it will be configured using a stream ID enabled with before image. It is populated for UPDATE and DELETE events. For INSERT events, before image doesn't make sense as the change record itself is in the context of new row insertion.

Yugabyte uses multi-version concurrency control (MVCC) mechanism, and compacts data at regular intervals. The compaction or the history retention is controlled by the history retention interval flag. However, when before image is enabled for a database, YugabyteDB adjusts the history retention for that database based on the most lagging active CDC stream so that the previous row state is retained, and available. Consequently, in the case of a lagging CDC stream, the amount of space required for the database grows as more data is retained. On the other hand, older rows that are not needed for any of the active CDC streams are identified and garbage collected.

Schema version that is currently being used by a CDC stream will be used to frame before and current row images. The before image functionality is disabled by default unless it is specifically turned on during the CDC stream creation. The yb-admin create_change_data_stream command can be used to create a CDC stream with before image enabled.

Use transformers

Add a transformer in the source connector while using with before image; you can add the following property directly to your configuration:

...
"transforms":"unwrap,extract",
"transforms.unwrap.type":"io.debezium.connector.yugabytedb.transforms.PGCompatible",
"transforms.unwrap.drop.tombstones":"false",
"transforms.extract.type":"io.debezium.transforms.ExtractNewRecordState",
"transforms.extract.drop.tombstones":"false",
...

After you've enabled before image and are using the suggested transformers, the effect of an update statement with the record structure is as follows:

UPDATE customers SET email = 'service@example.com' WHERE id = 1;
{
  "schema": {...},
  "payload": {
    "before": { --> 1
      "id": 1,
      "name": "Vaibhav Kushwaha",
      "email": "vaibhav@example.com"
    }
    "after": { --> 2
      "id": 1,
      "name": "Vaibhav Kushwaha",
      "email": "service@example.com"
    },
    "source": { --> 3
      "version": "1.9.5.y.11",
      "connector": "yugabytedb",
      "name": "dbserver1",
      "ts_ms": -8881476960074,
      "snapshot": "false",
      "db": "yugabyte",
      "sequence": "[null,\"1:5::0:0\"]",
      "schema": "public",
      "table": "customers",
      "txId": "",
      "lsn": "1:5::0:0",
      "xmin": null
    },
    "op": "u", --> 4
    "ts_ms": 1646149134341,
    "transaction": null
  }
}

The highlighted fields in the update event are:

Item Field name Description
1 before The value of the row before the update operation.
2 after Specifies the state of the row after the change event occurred. In this example, the value of email has changed to service@example.com.
3 source Mandatory field that describes the source metadata for the event. This has the same fields as a create event, but some values are different. The source metadata includes:
  • Debezium version
  • Connector type and name
  • Database and table that contains the new row
  • Schema name
  • If the event was part of a snapshot (always false for update events)
  • ID of the transaction in which the operation was performed
  • Offset of the operation in the database log
  • Timestamp for when the change was made in the database
4 op In an update event, this field's value is u, signifying that this row changed because of an update.

Here is one more example, consider the following employee table into which a row is inserted, subsquently updated, and deleted:

create table employee(employee_id int primary key, employee_name varchar);

insert into employee values(1001, 'Alice');

update employee set employee_name='Bob' where employee_id=1001;

delete from employee where employee_id=1001;

CDC records for update and delete statements without enabling before image would be as follows:

With before image enabled, the update and delete records look like the following:

CDC record for UPDATE: CDC record for DELETE:
{
  "before": {
    "public.employee.Value":{
      "employee_id": {
        "value": 1001
      },
      "employee_name": {
        "employee_name": {
          "value": {
            "string": "Alice"
          }
        }
      }
    }
  },
  "after": {
    "public.employee.Value":{
      "employee_id": {
        "value": 1001
      },
      "employee_name": {
        "employee_name": {
          "value": {
            "string": "Bob"
          }
        }
      }
    }
  },
  "op": "u"
}
{
  "before": {
    "public.employee.Value":{
      "employee_id": {
        "value": 1001
      },
      "employee_name": {
        "employee_name": {
          "value": {
            "string": "Bob"
          }
        }
      }
    }
  },
  "after": null,
  "op": "d"
}

Schema evolution

Table schema is needed for decoding and processing the changes and populating CDC records. Thus, older schemas are retained if CDC streams are lagging. Also, older schemas that are not needed for any of the existing active CDC streams are garbage collected. In addition, if before image is enabled, the schema needed for populating before image is also retained. The YugabyteDB source connector caches schema at the tablet level. This means that for every tablet the connector has a copy of the current schema for the tablet it is polling the changes for. As soon as a DDL command is executed on the source table, the CDC service emits a record with the new schema for all the tablets. The YugabyteDB source connector then reads those records and modifies its cached schema gracefully.

No backfill support

If you alter the schema of the source table to add a default value for an existing column, the connector will NOT emit any event for the schema change. The default value will only be published in the records created after schema change is made. In such cases, it is recommended to alter the schema in your sinks to add the default value there as well.

Consider the following employee table (with schema version 0 at the time of table creation) into which a row is inserted, followed by a DDL resulting in schema version 1 and an update of the row inserted, and subsequently another DDL incrementing the schema version to 2. If a CDC stream created for the employee table lags and is in the process of streaming the update, corresponding schema version 1 is used for populating the update record.

create table employee(employee_id int primary key, employee_name varchar); // schema version 0

insert into employee values(1001, 'Alice');

alter table employee add dept_id int; // schema version 1

update employee set dept_id=9 where employee_id=1001; // currently streaming record corresponding to this update

alter table employee add dept_name varchar; // schema version 2

Update CDC record would be as follows:

CDC record for UPDATE (using schema version 1):
{
  "before": {
    "public.employee.Value":{
      "employee_id": {
        "value": 1001
      },
      "employee_name": {
        "employee_name": {
          "value": {
            "string": "Alice"
          }
        }
      },
      "dept_id": null
    }
  },

  "after": {   "public.employee.Value":{
        "employee_id": {
          "value": 1001
        },
        "employee_name": {
          "employee_name": {
            "value": {
              "string": "Alice"
            }
          }
        },
        "dept_id": {
          "dept_id": {
            "value": {
              "int": 9
            }
          }
        }
      }
    },
    "op": "u"
}

Important configuration settings

You can use several flags to fine-tune YugabyteDB's CDC behavior. These flags are documented in the Change data capture flags section of the YB-TServer reference and Change data capture flags section of the YB-Master reference. The following flags are particularly important for configuring CDC:

  • cdc_intent_retention_ms - Controls retention of intents, in ms. If a request for change records is not received for this interval, un-streamed intents are garbage collected and the CDC stream is considered expired. This expiry is not reversible, and the only course of action would be to create a new CDC stream. The default value of this flag is 4 hours (4 x 3600 x 1000 ms).

  • cdc_wal_retention_time_secs - Controls how long WAL is retained, in seconds. This is irrespective of whether a request for change records is received or not. The default value of this flag is 4 hours (14400 seconds).

  • cdc_snapshot_batch_size - This flag's default value is 250 records included per batch in response to an internal call to get the snapshot. If the table contains a very large amount of data, you may need to increase this value to reduce the amount of time it takes to stream the complete snapshot. You can also choose not to take a snapshot by modifying the Debezium configuration.

  • cdc_max_stream_intent_records - Controls how many intent records can be streamed in a single GetChanges call. Essentially, intents of large transactions are broken down into batches of size equal to this flag, hence this controls how many batches of GetChanges calls are needed to stream the entire large transaction. The default value of this flag is 1680, and transactions with intents less than this value are streamed in a single batch. The value of this flag can be increased, if the workload has larger transactions and CDC throughput needs to be increased. Note that high values of this flag can increase the latency of each GetChanges call.

Retaining data for longer durations

To increase retention of data for CDC, change the two flags, cdc_intent_retention_ms and cdc_wal_retention_time_secs as required.

Important

Longer values of cdc_intent_retention_ms, coupled with longer CDC lags (periods of downtime where the client is not requesting changes) can result in increased memory footprint in the YB-TServer and affect read performance.

Content-based routing

By default, the Yugabyte Debezium connector streams all of the change events that it reads from a table to a single static topic. However, you may want to re-route the events into different Kafka topics based on the event's content. You can do this using the Debezium ContentBasedRouter. But first, two additional dependencies need to be placed in the Kafka-Connect environment. These are not included in the official yugabyte-debezium-connector for security reasons. These dependencies are:

  • Debezium routing SMT (Single Message Transform)
  • Groovy JSR223 implementation (or other scripting languages that integrate with JSR 223)

To get started, you can rebuild the yugabyte-debezium-connector image including these dependencies. Here's what the Dockerfile would look like:

FROM quay.io/yugabyte/debezium-connector:latest
# Add the required jar files for content based routing
RUN cd $KAFKA_CONNECT_YB_DIR && curl -so debezium-scripting-2.1.2.Final.jar https://repo1.maven.org/maven2/io/debezium/debezium-scripting/2.1.2.Final/debezium-scripting-2.1.2.Final.jar
RUN cd $KAFKA_CONNECT_YB_DIR && curl -so groovy-4.0.9.jar  https://repo1.maven.org/maven2/org/apache/groovy/groovy/4.0.9/groovy-4.0.9.jar
RUN cd $KAFKA_CONNECT_YB_DIR && curl -so groovy-jsr223-4.0.9.jar  https://repo1.maven.org/maven2/org/apache/groovy/groovy-jsr223/4.0.9/groovy-jsr223-4.0.9.jar

To configure a content-based router, you need to add the following lines to your connector configuration:

{
  ...,
  config: {
    ...,
    "transforms": "router",
    "transforms.router.type": "io.debezium.transforms.ContentBasedRouter",
    "transforms.router.language": "jsr223.groovy",
    "transforms.router.topic.expression": "<routing-expression>",
  }
}

The <routing-expression> contains the logic for routing of the events. For example, if you want to re-route the events based on the country column in user's table, you may use a expression similar to the following:

value.after != null ? (value.after?.country?.value == '\''UK'\'' ? '\''uk_users'\'' : null) : (value.before?.country?.value == '\''UK'\'' ? '\''uk_users'\'' : null)"

This expression checks if the value of the row after the operation has the country set to "UK". If yes then the expression returns "uk_users." If no, it returns null, and in case the row after the operation is null (for example, in a "delete" operation), the expression also checks for the same condition on row values before the operation. The value that is returned determines which new Kafka Topic will receive the re-routed event. If it returns null, the event is sent to the default topic.

For more advanced routing configuration, refer to the Debezium documentation on content-based routing.