Chapter 14. Cruise Control for cluster rebalancing

Cruise Control reduces the time and effort involved in running an efficient Kafka cluster, with a more evenly balanced workload across the brokers.

A typical cluster can become unevenly loaded over time. Partitions that handle large amounts of message traffic might be unevenly distributed across the available brokers. To rebalance the cluster, administrators must monitor the load on brokers and manually reassign busy partitions to brokers with spare capacity.

Cruise Control automates this cluster rebalancing process. It constructs a workload model of resource utilization, based on CPU, disk, and network load. Using a set of configurable optimization goals, you can instruct Cruise Control to generate dry run optimization proposals for more balanced partition assignments.

After you have reviewed a dry run optimization proposal, you can instruct Cruise Control to initiate a cluster rebalance based on that proposal, or generate a new proposal.

When a cluster rebalancing operation is complete, the brokers are used more effectively and the load on the Kafka cluster is more evenly balanced.

A zipped distribution of Cruise Control for AMQ Streams on Red Hat Enterprise Linux is available for download from the Red Hat Customer Portal.

Before starting Cruise Control, you must configure the Kafka brokers to use the provided Cruise Control Metrics Reporter.

When loaded at runtime, the Metrics Reporter sends metrics to the __CruiseControlMetrics topic, one of three auto-created topics. Cruise Control uses these metrics to create and update the workload model and to calculate optimization proposals.

Configure the properties used by Cruise Control and then start the Cruise Control server using the cruise-control-start.sh script. The server is hosted on a single machine for the whole Kafka cluster.

Three topics are auto-created when Cruise Control starts. For more information, see Auto-created topics.

Auto-created topics

The following table shows the three topics that are automatically created when Cruise Control starts. These topics are required for Cruise Control to work properly and must not be deleted or changed.

To ensure that log compaction is disabled in the auto-created topics, make sure that you configure the Cruise Control Metrics Reporter as described in Section 14.3, “Deploying the Cruise Control Metrics Reporter”. Log compaction can remove records that are needed by Cruise Control and prevent it from working properly.

To rebalance a Kafka cluster, Cruise Control uses optimization goals to generate optimization proposals. Optimization goals are constraints on workload redistribution and resource utilization across a Kafka cluster.

AMQ Streams on Red Hat Enterprise Linux supports all the optimization goals developed in the Cruise Control project. The supported goals, in the default descending order of priority, are as follows:

For more information on each optimization goal, see Goals in the Cruise Control Wiki.

Goals configuration in the Cruise Control properties file

You configure optimization goals in the cruisecontrol.properties file in the cruise-control/config/ directory. There are configurations for hard optimization goals that must be satisfied, as well as master and default optimization goals.

Optional, user-provided optimization goals are set at runtime as parameters in requests to the /rebalance endpoint.

Optimization goals are subject to any capacity limits on broker resources.

The following sections describe each goal configuration in more detail.

Master optimization goals

The master optimization goals are available to all users. Goals that are not listed in the master optimization goals are not available for use in Cruise Control operations.

The following master optimization goals are preset in the cruisecontrol.properties file, in the goals property, in descending priority order:

RackAwareGoal; MinTopicLeadersPerBrokerGoal; ReplicaCapacityGoal; DiskCapacityGoal; NetworkInboundCapacityGoal; NetworkOutboundCapacityGoal; ReplicaDistributionGoal; PotentialNwOutGoal; DiskUsageDistributionGoal; NetworkInboundUsageDistributionGoal; NetworkOutboundUsageDistributionGoal; CpuUsageDistributionGoal; TopicReplicaDistributionGoal; LeaderReplicaDistributionGoal; LeaderBytesInDistributionGoal; PreferredLeaderElectionGoal

For simplicity, we recommend that you do not change the preset master optimization goals, unless you need to completely exclude one or more goals from being used to generate optimization proposals. The priority order of the master optimization goals can be modified, if desired, in the configuration for default optimization goals.

If you need to modify the preset master optimization goals, specify a list of goals, in descending priority order, in the goals property. Use fully-qualified domain names as shown in the cruisecontrol.properties file.

You must specify at least one master goal, or Cruise Control will crash.

Hard goals and soft goals

Hard goals are goals that must be satisfied in optimization proposals. Goals that are not configured as hard goals are known as soft goals. You can think of soft goals as best effort goals: they do not need to be satisfied in optimization proposals, but are included in optimization calculations.

Cruise Control will calculate optimization proposals that satisfy all the hard goals and as many soft goals as possible (in their priority order). An optimization proposal that does not satisfy all the hard goals is rejected by the Analyzer and is not sent to the user.

The following master optimization goals are preset as hard goals in the cruisecontrol.properties file, in the hard.goals property:

RackAwareGoal; MinTopicLeadersPerBrokerGoal; ReplicaCapacityGoal; DiskCapacityGoal; NetworkInboundCapacityGoal; NetworkOutboundCapacityGoal; CpuCapacityGoal

To change the hard goals, edit the hard.goals property and specify the desired goals, using their fully-qualified domain names.

Increasing the number of hard goals reduces the likelihood that Cruise Control will calculate and generate valid optimization proposals.

Default optimization goals

Cruise Control uses the default optimization goals list to generate the cached optimization proposal. For more information, see Section 14.6, “Optimization proposals overview”.

You can override the default optimization goals at runtime by setting user-provided optimization goals.

The following default optimization goals are preset in the cruisecontrol.properties file, in the default.goals property, in descending priority order:

RackAwareGoal; MinTopicLeadersPerBrokerGoal; ReplicaCapacityGoal; DiskCapacityGoal; NetworkInboundCapacityGoal; NetworkOutboundCapacityGoal; CpuCapacityGoal; ReplicaDistributionGoal; PotentialNwOutGoal; DiskUsageDistributionGoal; NetworkInboundUsageDistributionGoal; NetworkOutboundUsageDistributionGoal; CpuUsageDistributionGoal; TopicReplicaDistributionGoal; LeaderReplicaDistributionGoal; LeaderBytesInDistributionGoal

You must specify at least one default goal, or Cruise Control will crash.

To modify the default optimization goals, specify a list of goals, in descending priority order, in the default.goals property. Default goals must be a subset of the master optimization goals; use fully-qualified domain names.

User-provided optimization goals

User-provided optimization goals narrow down the configured default goals for a particular optimization proposal. You can set them, as required, as parameters in HTTP requests to the /rebalance endpoint. For more information, see Section 14.9, “Generating optimization proposals”.

User-provided optimization goals can generate optimization proposals for different scenarios. For example, you might want to optimize leader replica distribution across the Kafka cluster without considering disk capacity or disk utilization. So, you send a request to the /rebalance endpoint containing a single goal for leader replica distribution.

User-provided optimization goals must:

To ignore the configured hard goals in an optimization proposal, add the skip_hard_goals_check=true parameter to the request.

An optimization proposal is a summary of proposed changes that, if applied, will produce a more balanced Kafka cluster, with partition workloads distributed more evenly among the brokers. Each optimization proposal is based on the set of optimization goals that was used to generate it, subject to any configured capacity limits on broker resources.

When you make a POST request to the /rebalance endpoint, an optimization proposal is returned in response. Use the information in the proposal to decide whether to initiate a cluster rebalance based on the proposal. Alternatively, you can change the optimization goals and then generate another proposal.

By default, optimization proposals are generated as dry runs that must be initiated separately. There is no limit to the number of optimization proposals that can be generated.

Cached optimization proposal

Cruise Control maintains a cached optimization proposal based on the configured default optimization goals. Generated from the workload model, the cached optimization proposal is updated every 15 minutes to reflect the current state of the Kafka cluster.

The most recent cached optimization proposal is returned when the following goal configurations are used:

To change the cached optimization proposal refresh interval, edit the proposal.expiration.ms setting in the cruisecontrol.properties file. Consider a shorter interval for fast changing clusters, although this increases the load on the Cruise Control server.

Contents of optimization proposals

The following table describes the properties contained in an optimization proposal.

You can adjust several performance tuning options for cluster rebalances. These options control how partition replica and leadership movements in a rebalance are executed, as well as the bandwidth that is allocated to a rebalance operation.

Partition reassignment commands

Optimization proposals are composed of separate partition reassignment commands. When you initiate a proposal, the Cruise Control server applies these commands to the Kafka cluster.

A partition reassignment command consists of either of the following types of operations:

Cruise Control issues partition reassignment commands to the Kafka cluster in batches. The performance of the cluster during the rebalance is affected by the number of each type of movement contained in each batch.

To configure partition reassignment commands, see Rebalance tuning options.

Replica movement strategies

Cluster rebalance performance is also influenced by the replica movement strategy that is applied to the batches of partition reassignment commands. By default, Cruise Control uses the BaseReplicaMovementStrategy, which applies the commands in the order in which they were generated. However, if there are some very large partition reassignments early in the proposal, this strategy can slow down the application of the other reassignments.

Cruise Control provides three alternative replica movement strategies that can be applied to optimization proposals:

These strategies can be configured as a sequence. The first strategy attempts to compare two partition reassignments using its internal logic. If the reassignments are equivalent, then it passes them to the next strategy in the sequence to decide the order, and so on.

To configure replica movement strategies, see Rebalance tuning options.

Rebalance tuning options

Cruise Control provides several configuration options for tuning rebalance parameters. These options are set in the following ways:

The relevant configurations for both methods are summarized in the following table.

Changing the default settings affects the length of time that the rebalance takes to complete, as well as the load placed on the Kafka cluster during the rebalance. Using lower values reduces the load but increases the amount of time taken, and vice versa.

The config/cruisecontrol.properties file contains the configuration for Cruise Control. The file consists of properties in one of the following types:

You can specify and configure all the properties listed in the Configurations section of the Cruise Control Wiki.

Capacity configuration

Cruise Control uses capacity limits to determine if certain resource-based optimization goals are being broken. An attempted optimization fails if one or more of these resource-based goals is set as a hard goal and then broken. This prevents the optimization from being used to generate an optimization proposal.

You specify capacity limits for Kafka broker resources in one of the following three .json files in cruise-control/config:

Set the file in the capacity.config.file property in cruisecontrol.properties. The selected file will be used for broker capacity resolution. For example:

capacity.config.file=config/capacityJBOD.json

Capacity limits can be set for the following broker resources in the described units:

To apply the same capacity limits to every broker monitored by Cruise Control, set capacity limits for broker ID -1. To set different capacity limits for individual brokers, specify each broker ID and its capacity configuration.

{
  "brokerCapacities":[
    {
      "brokerId": "-1",
      "capacity": {
        "DISK": "100000",
        "CPU": "100",
        "NW_IN": "10000",
        "NW_OUT": "10000"
      },
      "doc": "This is the default capacity. Capacity unit used for disk is in MB, cpu is in percentage, network throughput is in KB."
    },
    {
      "brokerId": "0",
      "capacity": {
        "DISK": "500000",
        "CPU": "100",
        "NW_IN": "50000",
        "NW_OUT": "50000"
      },
      "doc": "This overrides the capacity for broker 0."
    }
  ]
}

For more information, see Populating the Capacity Configuration File in the Cruise Control Wiki.

Log cleanup policy for Cruise Control Metrics topic

It is important that the auto-created __CruiseControlMetrics topic (see auto-created topics) has a log cleanup policy of DELETE rather than COMPACT. Otherwise, records that are needed by Cruise Control might be removed.

As described in Section 14.3, “Deploying the Cruise Control Metrics Reporter”, setting the following options in the Kafka configuration file ensures that the COMPACT log cleanup policy is correctly set:

If topic auto-creation is disabled in the Cruise Control Metrics Reporter (cruise.control.metrics.topic.auto.create=false), but enabled in the Kafka cluster, then the __CruiseControlMetrics topic is still automatically created by the broker. In this case, you must change the log cleanup policy of the __CruiseControlMetrics topic to DELETE using the kafka-configs.sh tool.

If topic auto-creation is disabled in both the Cruise Control Metrics Reporter and the Kafka cluster, you must create the __CruiseControlMetrics topic manually and then configure it to use the DELETE log cleanup policy using the kafka-configs.sh tool.

For more information, see Section 5.9, “Modifying a topic configuration”.

Logging configuration

Cruise Control uses log4j1 for all server logging. To change the default configuration, edit the log4j.properties file in /opt/cruise-control/config/log4j.properties.

You must restart the Cruise Control server before the changes take effect.

When you make a POST request to the /rebalance endpoint, Cruise Control generates an optimization proposal to rebalance the Kafka cluster, based on the provided optimization goals.

The optimization proposal is generated as a dry run, unless the dryrun parameter is supplied and set to false. In "dry run mode", Cruise Control generates the optimization proposal and the estimated result, but doesn’t initiate the proposal by rebalancing the cluster.

You can analyze the information returned in the optimization proposal and decide whether to initiate it.

Following are the key parameters for requests to the /rebalance endpoint. For information about all the available parameters, see REST APIs in the Cruise Control Wiki.

Informs Cruise Control whether you want to generate an optimization proposal only (true), or generate an optimization proposal and perform a cluster rebalance (false).

When dryrun=true (the default), you can also pass the verbose parameter to return more detailed information about the state of the Kafka cluster. This includes metrics for the load on each Kafka broker before and after the optimization proposal is applied, and the differences between the before and after values.

A regular expression that matches the topics to exclude from the calculation of the optimization proposal.

List of user-provided optimization goals to use to prepare the optimization proposal. If goals are not supplied, the configured default.goals list in the cruisecontrol.properties file is used.

By default, Cruise Control checks that the user-provided optimization goals (in the goals parameter) contain all the configured hard goals (in hard.goals). A request fails if you supply goals that are not a subset of the configured hard.goals.

Set skip_hard_goals_check to true if you want to generate an optimization proposal with user-provided optimization goals that do not include all the configured hard.goals.

Controls the type of response returned by the Cruise Control server. If not supplied, or set to false, then Cruise Control returns text formatted for display on the command line. If you want to extract elements of the returned information programmatically, set json=true. This will return JSON formatted text that can be piped to tools such as jq, or parsed in scripts and programs.

Controls the level of detail in responses that are returned by the Cruise Control server. Can be used with dryrun=true.

Asynchronous responses

The Cruise Control REST API endpoints timeout after 10 seconds by default, although proposal generation continues on the server. A timeout might occur if the most recent cached optimization proposal is not ready, or if user-provided optimization goals were specified with ignore_proposal_cache=true.

To allow you to retrieve the optimization proposal at a later time, take note of the request’s unique identifier, which is given in the header of responses from the /rebalance endpoint.

To obtain the response using curl, specify the verbose (-v) option:

curl -v -X POST 'cruise-control-server:9090/kafkacruisecontrol/rebalance'

Here is an example header:

* Connected to cruise-control-server (::1) port 9090 (#0)
> POST /kafkacruisecontrol/rebalance HTTP/1.1
> Host: cc-host:9090
> User-Agent: curl/7.70.0
> Accept: /
>
* Mark bundle as not supporting multiuse
< HTTP/1.1 200 OK
< Date: Mon, 01 Jun 2020 15:19:26 GMT
< Set-Cookie: JSESSIONID=node01wk6vjzjj12go13m81o7no5p7h9.node0; Path=/
< Expires: Thu, 01 Jan 1970 00:00:00 GMT
< User-Task-ID: 274b8095-d739-4840-85b9-f4cfaaf5c201
< Content-Type: text/plain;charset=utf-8
< Cruise-Control-Version: 2.0.103.redhat-00002
< Cruise-Control-Commit_Id: 58975c9d5d0a78dd33cd67d4bcb497c9fd42ae7c
< Content-Length: 12368
< Server: Jetty(9.4.26.v20200117-redhat-00001)

If an optimization proposal is not ready within the timeout, you can re-submit the POST request, this time including the User-Task-ID of the original request in the header:

curl -v -X POST -H 'User-Task-ID: 274b8095-d739-4840-85b9-f4cfaaf5c201' 'cruise-control-server:9090/kafkacruisecontrol/rebalance'

If you are satisfied with an optimization proposal, you can instruct Cruise Control to initiate the cluster rebalance and begin reassigning partitions, as summarized in the proposal.

Leave as little time as possible between generating an optimization proposal and initiating the cluster rebalance. If some time has passed since you generated the original optimization proposal, the cluster state might have changed. Therefore, the cluster rebalance that is initiated might be different to the one you reviewed. If in doubt, first generate a new optimization proposal.

Only one cluster rebalance, with a status of "Active", can be in progress at a time.

You can stop the cluster rebalance that is currently in progress.

This instructs Cruise Control to finish the current batch of partition reassignments and then stop the rebalance. When the rebalance has stopped, completed partition reassignments have already been applied; therefore, the state of the Kafka cluster is different when compared to before the start of the rebalance operation. If further rebalancing is required, you should generate a new optimization proposal.