Tutorial: Add and Remove Brokers with Auto Data Balancer in Confluent Platform¶
This tutorial runs Confluent Auto Data Balancer (ADB) on Confluent Server, which allows you to shift data to create an even workload across your cluster. By the end of this tutorial, you will have successfully run the Auto Data Balancer CLI tool to rebalance data after adding and removing brokers.
Starting in Confluent Platform 6.0.0, Manage Self-Balancing Kafka Clusters in Confluent Platform is the preferred alternative to Auto Data Balancer. For a detailed feature comparison, see Self-Balancing vs. Auto Data Balancer. For the Self-Balancing Docker Quick Start, see Quick Start for Self-Balancing Kafka Clusters in Confluent Platform.
Installing and Running Docker¶
For this tutorial, you run Docker using the Docker client. If you are interested in information on using Docker Compose to run the images, skip to the bottom of this guide.
To get started, install Docker and get it running. The Confluent Platform Docker Images require Docker version 1.11 or greater.
- Prerequisites
- Docker version 1.11 or later installed and running.
- If you’re running on macOS, you must use Docker for Mac.
- If you’re running on Windows, you must use Docker for Windows.
- You must allocate at least 8 GB of RAM (2 GB is the default).
- Git
- Docker version 1.11 or later installed and running.
Note
In this tutorial, Kafka and ZooKeeper are configured to store data locally in the Docker containers. For production deployments, you should use mounted volumes for persisting data in the event that a container stops running or is restarted. Kafka relies heavily on the filesystem for storing and caching messages, so this is important when running Kafka on Docker. For an example of how to add mounted volumes to the host machine, see Mount Docker External Volumes in Confluent Platform.
Docker Client: Setting Up a Three Node Kafka Cluster¶
Note
In the following steps, each Docker container runs in detached mode. You are shown how to access to the
logs for a running container. You can also run the containers in the foreground by replacing the -d flags
with -it.
Clone the Git repository and navigate to the examples directory:
git clone git@github.com:confluentinc/kafka-images.git
cd kafka-images/examples/enterprise-kafka
Start the services using the example Docker Compose file. The Docker Compose file has configuration properties for one ZooKeeper and six Kafka brokers. These brokers are configured to be on two racks. One rack with 3 brokers is started, a topic with sample data is created, and then the Auto Data Balancer CLI tool runs to balance the cluster. After this step, you will add another rack of brokers and run the Auto Data Balancer CLI tool to rebalance the data across the newly added brokers.
Start ZooKeeper and first rack of brokers using the Docker Compose commands.
docker-compose createYou should see the following
Creating enterprise-kafka_zookeeper_1 ... done Creating enterprise-kafka_kafka-1_1 ... done Creating enterprise-kafka_kafka-2_1 ... done Creating enterprise-kafka_kafka-3_1 ... done Creating enterprise-kafka_kafka-4_1 ... done Creating enterprise-kafka_kafka-5_1 ... done Creating enterprise-kafka_kafka-6_1 ... done
Start the services.
docker-compose start zookeeper kafka-1 kafka-2 kafka-3
You should see the following
Starting zookeeper ... done Starting kafka-1 ... done Starting kafka-2 ... done Starting kafka-3 ... done
These are optional verification steps.
Verify the services are up and running with this command:
docker-compose psYou should see the following:
Name Command State Ports ------------------------------------------------------------------------ enterprisekafka_kafka-1_1 /etc/confluent/docker/run Up enterprisekafka_kafka-2_1 /etc/confluent/docker/run Up enterprisekafka_kafka-3_1 /etc/confluent/docker/run Up enterprisekafka_kafka-4_1 /etc/confluent/docker/run Exit 0 enterprisekafka_kafka-5_1 /etc/confluent/docker/run Exit 0 enterprisekafka_kafka-6_1 /etc/confluent/docker/run Exit 0 enterprisekafka_zookeeper_1 /etc/confluent/docker/run Up
Verify that ZooKeeper is healthy by checking the ZooKeeper logs.
docker-compose logs zookeeper | grep -i binding
You should see the following in your terminal window:
zookeeper_1 | [2016-10-21 22:15:22,494] INFO binding to port 0.0.0.0/0.0.0.0:22181 (org.apache.zookeeper.server.NIOServerCnxnFactory)
Check the Kafka logs for the destination cluster to verify that broker is healthy.
docker-compose logs kafka-1 | grep -i started
You should see message a message that looks like the following:
kafka-1_1 | [2016-10-21 22:19:50,964] INFO [Socket Server on Broker 1], Started 1 acceptor threads (kafka.network.SocketServer) kafka-1_1 | [2016-10-21 22:19:51,300] INFO [Kafka Server 1], started (kafka.server.KafkaServer) ....
Now that the brokers are up, creates a test topic called
adb-test.docker run \ --net=host \ --rm confluentinc/cp-kafka:7.7.2 \ kafka-topics --create --topic adb-test --partitions 20 --replication-factor 3 --if-not-exists --bootstrap-server localhost:19092
You should see the following output in your terminal window:
Created topic "adb-test".
Optional: Verify that the topic was created successfully:
docker run \ --net=host \ --rm confluentinc/cp-kafka:7.7.2 \ kafka-topics --describe --topic adb-test --bootstrap-server localhost:19092
You should see the following output in your terminal window:
Topic:adb-test PartitionCount:20 ReplicationFactor:3 Configs: Topic: adb-test Partition: 0 Leader: 2 Replicas: 2,1,3 Isr: 2,1,3 Topic: adb-test Partition: 1 Leader: 3 Replicas: 3,2,1 Isr: 3,2,1 Topic: adb-test Partition: 2 Leader: 1 Replicas: 1,3,2 Isr: 1,3,2 Topic: adb-test Partition: 3 Leader: 2 Replicas: 2,3,1 Isr: 2,3,1 Topic: adb-test Partition: 4 Leader: 3 Replicas: 3,1,2 Isr: 3,1,2 Topic: adb-test Partition: 5 Leader: 1 Replicas: 1,2,3 Isr: 1,2,3 Topic: adb-test Partition: 6 Leader: 2 Replicas: 2,1,3 Isr: 2,1,3 Topic: adb-test Partition: 7 Leader: 3 Replicas: 3,2,1 Isr: 3,2,1 Topic: adb-test Partition: 8 Leader: 1 Replicas: 1,3,2 Isr: 1,3,2 Topic: adb-test Partition: 9 Leader: 2 Replicas: 2,3,1 Isr: 2,3,1 Topic: adb-test Partition: 10 Leader: 3 Replicas: 3,1,2 Isr: 3,1,2 Topic: adb-test Partition: 11 Leader: 1 Replicas: 1,2,3 Isr: 1,2,3 Topic: adb-test Partition: 12 Leader: 2 Replicas: 2,1,3 Isr: 2,1,3 Topic: adb-test Partition: 13 Leader: 3 Replicas: 3,2,1 Isr: 3,2,1 Topic: adb-test Partition: 14 Leader: 1 Replicas: 1,3,2 Isr: 1,3,2 Topic: adb-test Partition: 15 Leader: 2 Replicas: 2,3,1 Isr: 2,3,1 Topic: adb-test Partition: 16 Leader: 3 Replicas: 3,1,2 Isr: 3,1,2 Topic: adb-test Partition: 17 Leader: 1 Replicas: 1,2,3 Isr: 1,2,3 Topic: adb-test Partition: 18 Leader: 2 Replicas: 2,1,3 Isr: 2,1,3 Topic: adb-test Partition: 19 Leader: 3 Replicas: 3,2,1 Isr: 3,2,1
Next, we’ll try generating some data to our new topic:
docker run \ --net=host \ --rm \ confluentinc/cp-kafka:7.7.2 \ bash -c 'kafka-producer-perf-test --topic adb-test --num-records 2000000 --record-size 1000 --throughput 100000 --producer-props bootstrap.servers=localhost:19092'
This command will use the built-in Kafka Performance Producer to produce 2 GB of sample data to the topic. Upon running it, you should see the following:
209047 records sent, 41784.3 records/sec (39.85 MB/sec), 91.1 ms avg latency, 520.0 max latency. 325504 records sent, 65100.8 records/sec (62.08 MB/sec), 35.6 ms avg latency, 474.0 max latency. 258023 records sent, 51573.7 records/sec (49.18 MB/sec), 359.6 ms avg latency, 1264.0 max latency. 287934 records sent, 57586.8 records/sec (54.92 MB/sec), 455.1 ms avg latency, 1429.0 max latency. 413091 records sent, 81978.8 records/sec (78.18 MB/sec), 200.6 ms avg latency, 757.0 max latency. 282214 records sent, 56128.5 records/sec (53.53 MB/sec), 495.6 ms avg latency, 1738.0 max latency. 85071 records sent, 16815.8 records/sec (16.04 MB/sec), 468.0 ms avg latency, 3861.0 max latency. 115 records sent, 8.8 records/sec (0.01 MB/sec), 8307.4 ms avg latency, 13127.0 max latency. 13358 records sent, 2671.6 records/sec (2.55 MB/sec), 15408.9 ms avg latency, 23005.0 max latency. 74948 records sent, 14284.0 records/sec (13.62 MB/sec), 6555.0 ms avg latency, 22782.0 max latency. 5052 records sent, 1010.4 records/sec (0.96 MB/sec), 3228.3 ms avg latency, 8508.0 max latency. 2000000 records sent, 30452.988199 records/sec (29.04 MB/sec), 786.61 ms avg latency, 23005.00 ms max latency, 82 ms 50th, 1535 ms 95th, 22539 ms 99th, 22929 ms 99.9th.
Run
confluent-rebalancerto balance the data in the cluster.docker run \ --net=host \ --rm \ confluentinc/cp-enterprise-kafka:7.7.2 \ bash -c "confluent-rebalancer execute --bootstrap-server localhost:19092 --metrics-bootstrap-server localhost:19092 --throttle 100000000 --force --verbose"
You should see the rebalancing start and should see the following:
You are about to move 6 replica(s) for 6 partitions to 1 broker(s) with total size 0.9 MB. The preferred leader for 6 partition(s) will be changed. In total, the assignment for 7 partitions will be changed. The following brokers will require more disk space during the rebalance and, in some cases, after the rebalance: Broker Current (MB) During Rebalance (MB) After Rebalance (MB) 2 2,212.8 2,213.8 2,213.8 Min/max stats for brokers (before -> after): Type Leader Count Replica Count Size (MB) Min 8 (id: 2) -> 10 (id: 1) 21 (id: 2) -> 27 (id: 1) 2,069.6 (id: 1) -> 2,069.1 (id: 1) Max 12 (id: 3) -> 11 (id: 2) 30 (id: 1) -> 27 (id: 1) 2,212.8 (id: 2) -> 2,213.8 (id: 2) Rack stats (before -> after): Rack Leader Count Replica Count Size (MB) rack-a 31 -> 31 81 -> 81 6,352 -> 6,352 Broker stats (before -> after): Broker Leader Count Replica Count Size (MB) 1 11 -> 10 30 -> 27 2,069.6 -> 2,069.1 2 8 -> 11 21 -> 27 2,212.8 -> 2,213.8 3 12 -> 10 30 -> 27 2,069.6 -> 2,069.1 The rebalance has been started, run ``status`` to check progress. Warning: You must run the ``status`` or ``finish`` command periodically, until the rebalance completes, to ensure the throttle is removed. You can also alter the throttle by re-running the execute command passing a new value.
You can check the status of the rebalance operation by running the following command:
docker run \ --net=host \ --rm \ confluentinc/cp-enterprise-kafka:7.7.2 \ bash -c "confluent-rebalancer status --bootstrap-server localhost:19092"
If you see the a message like
7 partitions are being rebalanced, wait for 15-20 seconds and rerun the above command until you seeNo rebalance is currently in progress. This means that the rebalance action has completed successfully.You can finish the rebalance action by running the following command (this command ensures that the replication throttle is removed):
docker run \ --net=host \ --rm \ confluentinc/cp-enterprise-kafka:7.7.2 \ bash -c "confluent-rebalancer finish --bootstrap-server localhost:19092"
You should see the following in the logs:
The rebalance has completed and throttling has been disabled
Auto Data Balancer makes it easy to add new brokers to the cluster. You can now add an entire new rack to your cluster and run the rebalance operation again to balance the data across the cluster.
Start the new rack by running the following command:
docker-compose start kafka-4 kafka-5 kafka-6
You should follow the instructions in step 4 to verify the Kafka brokers are healthy.
Now start the rebalance operation by following step #. After the rebalance operation has finished, data should be balanced across the cluster. We will verify that by describing the topic metadata as follows.
docker run \ --net=host \ --rm confluentinc/cp-kafka:7.7.2 \ kafka-topics --describe --topic adb-test --bootstrap-server localhost:19092
You should see that partitions are spread across all of the brokers (i.e you should see some replicas and leaders assigned to brokers 4, 5, or 6).
Topic:adb-test PartitionCount:20 ReplicationFactor:3 Configs: Topic: adb-test Partition: 0 Leader: 1 Replicas: 1,5,6 Isr: 5,1,6 Topic: adb-test Partition: 1 Leader: 3 Replicas: 3,5,4 Isr: 5,3,4 Topic: adb-test Partition: 2 Leader: 6 Replicas: 6,4,1 Isr: 1,6,4 Topic: adb-test Partition: 3 Leader: 6 Replicas: 6,5,3 Isr: 5,6,3 Topic: adb-test Partition: 4 Leader: 1 Replicas: 1,4,5 Isr: 5,1,4 Topic: adb-test Partition: 5 Leader: 3 Replicas: 6,4,3 Isr: 6,3,4 Topic: adb-test Partition: 6 Leader: 1 Replicas: 5,1,6 Isr: 5,1,6 Topic: adb-test Partition: 7 Leader: 3 Replicas: 3,5,4 Isr: 5,3,4 Topic: adb-test Partition: 8 Leader: 4 Replicas: 4,6,1 Isr: 1,6,4 Topic: adb-test Partition: 9 Leader: 5 Replicas: 5,6,3 Isr: 5,6,3 Topic: adb-test Partition: 10 Leader: 2 Replicas: 2,4,5 Isr: 5,2,4 Topic: adb-test Partition: 11 Leader: 4 Replicas: 4,2,6 Isr: 6,2,4 Topic: adb-test Partition: 12 Leader: 5 Replicas: 5,2,6 Isr: 5,6,2 Topic: adb-test Partition: 13 Leader: 2 Replicas: 2,5,4 Isr: 5,2,4 Topic: adb-test Partition: 14 Leader: 4 Replicas: 4,6,2 Isr: 6,2,4 Topic: adb-test Partition: 15 Leader: 1 Replicas: 1,3,2 Isr: 1,2,3 Topic: adb-test Partition: 16 Leader: 2 Replicas: 3,2,1 Isr: 2,1,3 Topic: adb-test Partition: 17 Leader: 3 Replicas: 3,2,1 Isr: 3,2,1 Topic: adb-test Partition: 18 Leader: 1 Replicas: 1,2,3 Isr: 1,2,3 Topic: adb-test Partition: 19 Leader: 2 Replicas: 2,3,1 Isr: 2,3,1
Now you can try removing a broker and running the rebalance operation again.
Hint: You must notify the rebalancer to exclude broker from the rebalance plan. For example, to remove broker 1 you must run the following command:
docker run \ --net=host \ --rm \ confluentinc/cp-enterprise-kafka:7.7.2 \ bash -c "confluent-rebalancer execute --bootstrap-server localhost:19092 --metrics-bootstrap-server localhost:19092 --throttle 100000000 --force --verbose --remove-broker-ids 1"
Feel free to experiment with the
confluent-rebalancecommand. When you are done, use the following commands to shut down all the components.docker-compose stopIf you want to remove all the containers, run:
docker-compose rm