Introduction
In this article, we'll demonstrate creating a simple streaming data pipeline using a small micro-batching tool and CockroachDB's CDC Kafka sink.
Setup
Prerequisites:
CockroachDB Setup
Initially, create a local cluster of three nodes (or one node, not important):
cockroach start
cockroach start
cockroach start
cockroach init
Then, setup the source database tpcc and the target database tpcc_copy:
cockroach sql
cockroach sql
Finally, load the TPC-C fixture (schema and data) to the source database:
cockroach workload fixtures import tpcc
Kafka Setup
Ref: https://kafka.apache.org/quickstart
Initially, setup a local Kafka server that we'll use as CDC sink (using KRaft over ZK):
tar -xzf kafka_2.13-3.3.1.tgz
cd kafka_2.13-3.3.1
KAFKA_CLUSTER_ID="$(bin/kafka-storage.sh random-uuid)"
bin/kafka-storage.sh format -t $KAFKA_CLUSTER_ID -c config/kraft/server.properties
bin/kafka-server-start.sh config/kraft/server.properties
Optionally, start a console consumer to see the change events flashing by later. In this example for the warehouse table/topic:
bin/kafka-console-consumer.sh --topic warehouse --from-beginning --bootstrap-server localhost:9092
Pipeline Setup
Initially, clone the repo and build it locally:
git clone git@github.com:kai-niemi/roach-pipeline.git pipeline
cd pipeline
chmod +x mvnw
./mvnw clean install
The executable jar is now available under the target folder. Try it out with:
java -jar target/pipeline.jar --help
Now we are ready to create kafka2sql jobs for each TPC-C table we want to be streamed from the source to the target database.
First off, we get form templates that are going to be pre-populated with SQL statements for each table in question. We are only using a subset of the TPC-C workload tables, but the process is the same for all tables.
curl -X GET http://localhost:8090/kafka2sql/form?table=warehouse > warehouse-kafka2sql.json
curl -X GET http://localhost:8090/kafka2sql/form?table=district > district-kafka2sql.json
curl -X GET http://localhost:8090/kafka2sql/form?table=customer > customer-kafka2sql.json
Feel free to inspect the JSON files which should give an idea of how the batch jobs are configured and run. At this point, we haven't started anything yet. The JSON files typically don't need any editing if the template settings are properly set (everything defaults to using localhost).
Submit Batch Jobs
The next step is to POST the forms back which will register the jobs and start them up. The jobs need to be registered in the sorted topology order of the foreign key constraints (warehouse <- district <- customer) since we'll be creating tables on-the-fly.
curl -d "@warehouse-kafka2sql.json" -H "Content-Type:application/json" -X POST http://localhost:8090/kafka2sql
curl -d "@district-kafka2sql.json" -H "Content-Type:application/json" -X POST http://localhost:8090/kafka2sql
curl -d "@customer-kafka2sql.json" -H "Content-Type:application/json" -X POST http://localhost:8090/kafka2sql
The final step is to configure the Kafka change feeds for these three tables.
Connect to the source database and execute:
CREATE CHANGEFEED FOR TABLE warehouse INTO 'kafka://localhost:9092' WITH updated,resolved = '15s';
CREATE CHANGEFEED FOR TABLE district INTO 'kafka://localhost:9092' WITH updated,resolved = '15s';
CREATE CHANGEFEED FOR TABLE customer INTO 'kafka://localhost:9092' WITH updated,resolved = '15s';
You should see the target database starting to fill up and eventually reach the same state as the source database. If you would also run the TPC-C workload, you will see any changes reflected also in the target.
Conclusion
In this article, we looked at creating a simple streaming data pipeline at table level between two separate CockroachDB databases using the CDC Kafka sink.
