Change Data Capture

With a Dgraph enterprise license, you can use change data capture (CDC) to track data changes over time; including mutations and drops in your database. Dgraph’s CDC implementation lets you use Kafka or a local file as a sink to store CDC updates streamed by Dgraph Alpha leader nodes.

When CDC is enabled, Dgraph streams events for all set and delete mutations, except those that affect password fields; along with any drop events. Live Loader events are recorded by CDC, but Bulk Loader events aren’t.

CDC events are based on changes to Raft logs. So, if the sink is not reachable by the Alpha leader node, then Raft logs expand as events are collected on that node until the sink is available again. You should enable CDC on all Dgraph Alpha nodes to avoid interruptions in the stream of CDC events.

Enable CDC with Kafka sink

Kafka records CDC events under the dgraph-cdc topic. To enable CDC and sink events to Kafka, start Dgraph Alpha with the --cdc command and the sub-options shown below, as follows:

dgraph alpha --cdc "kafka=kafka-hostname:port; sasl-user=tstark; sasl-password=m3Ta11ic"

If you use Kafka on the localhost without SASL authentication, you can just specify the hostname and port used by Kafka, as follows:

dgraph alpha --cdc "localhost:9092"

Enable CDC with file sink

To enable CDC and sink results to a local unencrypted file, start Dgraph Alpha with the --cdc command and the sub-option shown below, as follows:

dgraph alpha --cdc "file=local-file-path"

CDC command reference

The --cdc option includes several sub-options that you can use to configure CDC when running the dgraph alpha command:

CDC data format

CDC events are in JSON format. Most CDC events look like the following example:

{ "key": "0", "value": {"meta":{"commit_ts":5},"type":"mutation","event":{"operation":"set","uid":2,"attr":"counter.val","value":1,"value_type":"int"}}}

The Meta.Commit_Ts value (shown above as "meta":{"commit_ts":5}) will increase with each CDC event, so you can use this value to find duplicate events if those occur due to Raft leadership changes in your Dgraph Alpha group.

Mutation event examples

A set mutation event updating counter.val to 10 would look like the following:

{"meta":{"commit_ts":29},"type":"mutation","event":{"operation":"set","uid":3,"attr":"counter.val","value":10,"value_type":"int"}}

Similarly, a delete mutation event that removes all values for the Author.name field for a specified node would look like the following:

{"meta":{"commit_ts":44},"type":"mutation","event":{"operation":"del","uid":7,"attr":"Author.name","value":"_STAR_ALL","value_type":"default"}}

Drop event examples

CDC drop events look like the following example event for “drop all”:

{"meta":{"commit_ts":13},"type":"drop","event":{"operation":"all"}}

The operation field specifies which drop operation (attribute, type, specified data, or all data) is tracked by the CDC event.

CDC and multi-tenancy

When you enable CDC in a multi-tenant environment, CDC events streamed to Kafka are distributed by the Kafka client. It distributes events between the available Kafka partitions based on their multi-tenancy namespace.

Known limitations

CDC has the following known limitations:

• CDC events do not track old values that are updated or removed by mutation or drop operations; only new values are tracked
• CDC is not currently supported when Dgraph is in Ludicrous mode
• CDC does not currently track schema updates
• You can only configure or enable CDC when starting Alpha nodes using the dgraph alpha command
• If a node crashes or the leadership of a Raft group changes, CDC might have duplicate events, but no data loss