Status

Current state: Accepted

Discussion thread: here 

JIRA: KAFKA-15045

Please keep the discussion on the mailing list rather than commenting on the wiki (wiki discussions get unwieldy fast).

Motivation

Back in KIP-441, we introduced a new task assignor which prioritized stateful task availability over stickiness, but kept the old assignment logic which prioritized stickiness called the StickyTaskAssignor. As a safeguard we added a backdoor internal config for the task assignor, the idea being that we could recommend this to users in case of critical bugs in the new assignor. However over time it has become clear that there are valid use cases for wanting the old StickyTaskAssignor over the new HighAvailabilityTaskAssignor, primarily relating to the possibility of extreme task shuffling induced by the HAAssignor in less stable environments.

There are also reasonable scenarios where users may want to plug in their own task assignor. For example, a user may want to configure their own static assignment to work around an issue in the available assignors. Or they may want to implement their own assignment logic that considers metrics they've collected, or task migration cost related to their specific processors.

Finally, there are good reasons for a user to want to extend or modify the behaviour of the Kafka Streams partition assignor beyond just changing the task assignment. For example, a user may want to implement their own initialization logic that initializes resources (much the same way the Streams Partition Assignor initializes internal topics).

With these motivations in mind, we are proposing to add a new group of configurable interfaces for plugging custom behaviour into the Streams Partition Assignor. This configuration will supplant the existing internal task assignor config. In this KIP, we will limit the scope of these configs to supplying a custom task assignor. However, this gives us a pattern based on which to, in future KIPs, add further configs which a user can set to plug in custom behavior.

Public Interfaces

We will introduce a new config that supplies an instance of TaskAssignor  (discussed below):

StreamsConfig
/** {@code partition.assignor.task.assignor.class} */
public static final String TASK_ASSIGNOR_CLASS_CONFIG = "task.assignor.class";
private static final String TASK_ASSIGNOR_CLASS_DOC = "A task assignor class or class name implementing the <@link TaskAssignor> interface". Defaults to the <@link HighAvailabilityTaskAssignor> class.";

We will also remove the old internal config (which we can do without deprecation as this was an internal config and thus by definition not part of the public API, also discussed in more detail in "Rejected Alternatives")

StreamsConfig
public static class InternalConfig {
    // This will be removed
    public static final String INTERNAL_TASK_ASSIGNOR_CLASS = "internal.task.assignor.class";
}

Note that the thread-level assignment will remain an un-configurable internal implementation detail of the partition assignor. See Consumer Assignments in the Proposed Changes section below for more details.

To enable users to actually plug something in by implementing TaskAssignor , we will need to move the TaskAssignor interface from o.a.k.streams.processor.internals.assignment to a public package, along with some of the supporting classes such as the assignment configs container class and ClientState which both appear in the TaskAssignor#assign method (although those will be heavily refactored, discussed below). All these new public APIs will be placed in a new public package that mirrors their old internal package, specifically org.apache.kafka.streams.processor.assignment.

Both the input parameter and return value will be encapsulated in wrapper classes for the sake of forwards compatibility. This will let us avoid the cycle of adding, deprecating, and removing new #assign overloads if/when we want to evolve the assignor in the future, for example to pass in additional metadata or enable the assignor to output new kinds of information or instructions to the StreamsPartitionAssignor. The analogous ConsumerPartitionAssignor works similarly, returning a single GroupAssignment object that wraps the collection of individual consumer assignments for the same reason.

This KIP introduces a number of classes and interfaces. Some of these are intended for the user to implement, and some of them are "read-only" APIs for the user to access but not implement. We'll start with the former category of user-implemented APIs and then go into the read-only "support class" APIs

User APIs

The following classes/interfaces will need to be implemented by the user in order to plug in a custom task assignor

TaskAssignor

The first and most important interface for users to implement in the TaskAssignor itself:

TaskAssignor
package org.apache.kafka.streams.processor.assignment;

public interface TaskAssignor extends Configurable {    

   /**
     * NONE: no error detected
     * ACTIVE_TASK_ASSIGNED_MULTIPLE_TIMES: multiple KafkaStreams clients assigned with the same active task
     * ACTIVE_AND_STANDBY_TASK_ASSIGNED_TO_SAME_KAFKASTREAMS: active task and standby task assigned to the same KafkaStreams client
     * INVALID_STANDBY_TASK: stateless task assigned as a standby task
     * UNKNOWN_PROCESS_ID: unrecognized ProcessId not matching any of the participating consumers
     * UNKNOWN_TASK_ID: unrecognized TaskId not matching any of the tasks to be assigned
     */   
    enum AssignmentError {     
	    NONE,
	    ACTIVE_TASK_ASSIGNED_MULTIPLE_TIMES,     
	    ACTIVE_AND_STANDBY_TASK_ASSIGNED_TO_SAME_KAFKASTREAMS,
	    INVALID_STANDBY_TASK,
        UNKNOWN_PROCESS_ID,
	    UNKNOWN_TASK_ID
  }

  /**
   * @param applicationState the metadata for this Kafka Streams application
   *
   * @return the assignment of active and standby tasks to KafkaStreams clients 
   *
   * @throws TaskAssignmentException If an error occurs during assignment and you wish for the rebalance to be retried,
   *                                 you can throw this exception to keep the assignment unchanged and automatically
   *                                 schedule an immediate followup rebalance. 
   */
  TaskAssignment assign(ApplicationState applicationState);

  /**
   * This callback can be used to observe the final assignment returned to the brokers and check for any errors that 
   * were detected while processing the returned assignment. If any errors were found, the corresponding 
   * will be returned and a StreamsException will be thrown after this callback returns. The StreamsException will
   * be thrown up to kill the StreamThread and can be handled as any other uncaught exception would if the application
   * has registered a {@link StreamsUncaughtExceptionHandler}.
   * 
   * @param assignment:   the final assignment returned to the kafka broker
   * @param subscription: the original subscription passed into the assignor
   * @param error:        the corresponding error type if one was detected while processing the returned assignment,  
   *                      or AssignmentError.NONE if the returned assignment was valid
   */
  default void onAssignmentComputed(GroupAssignment assignment, GroupSubscription subscription, AssignmentError error) {}

  /**
   * Wrapper class for the final assignment of active and standbys tasks to individual 
   * KafkaStreams clients
   */
  class TaskAssignment {

	/**
     * @return the assignment of tasks to kafka streams clients
     */
    public Collection<KafkaStreamsAssignment> assignment();
  }
}

Another reason for introducing the new TaskAssignment and ApplicationState classes is to clean up the way assignment is performed today, as the current API is really not fit for public consumption. Currently, the TaskAssignor is provided a set of ClientState objects representing each KafkaStreams client. The ClientState is however not just the input to the assignor, but also its output – the assignment of tasks to KafkaStreams clients is performed by mutating the ClientStates passed in. The return value of the #assign method is a simple boolean indicating to the StreamsPartitionAssignor whether it should request a followup probing rebalance, a feature associated only with the HighAvailabilityTaskAssignor.

To solve these problems, we plan to refactor the interface with two goals in mind:

  1. To provide a clean separation of input/output by splitting the ClientState into an input-only KafkaStreamsState metadata class and an output-only KafkaStreamsAssignment return value class
  2. To decouple the followup rebalance request from the probing rebalance feature and give the assignor more direct control over the followup rebalance schedule, by allowing it to indicate which KafkaStreams client(s) should trigger a rejoin and when to request the subsequent rebalance

This gives us the following two new top-level public interfaces, KafkaStreamsState  and KafkaStreamsAssignment :

KafkaStreamsAssignment

Next we have the KafkaStreamsAssignment class, representing the output of the assignment to be created by the TaskAssignor:

KafkaStreamsAssignment
package org.apache.kafka.streams.processor.assignment; 

/**
 * A simple container class for the assignor to return the desired placement of active and standby tasks on KafkaStreams clients
  */
public class KafkaStreamsAssignment {

  /* 
   * Construct an instance of KafkaStreamsAssignment with this processId and the given set of
   * assigned tasks. If you want this KafkaStreams client to request a followup rebalance, you
   * can set the followupRebalanceDeadline via the {@link #withFollowupRebalance(Instant)} API.
   *
   * @param processId the processId for the KafkaStreams client that should receive this assignment
   * @param assignment the set of tasks to be assigned to this KafkaStreams client
   *
   * @return a new KafkaStreamsAssignment object with the given processId and assignment
   */
  public static KafkaStreamsAssignment of(final ProcessId processId, final Set<AssignedTask> assignment);

  /**
   * This API can be used to request that a followup rebalance be triggered by the KafkaStreams client 
   * receiving this assignment. The followup rebalance will be initiated after the provided deadline
   * has passed, although it will always wait until it has finished the current rebalance before 
   * triggering a new one. This request will last until the new rebalance, and will be erased if a
   * new rebalance begins before the scheduled followup rebalance deadline has elapsed. The next
   * assignment must request the followup rebalance again if it still wants to schedule one for
   * the given instant, otherwise no additional rebalance will be triggered after that.
   * 
   * @param rebalanceDeadline the instant after which this KafkaStreams client will trigger a followup rebalance
   *
   * @return a new KafkaStreamsAssignment object with the same processId and assignment but with the given rebalanceDeadline
   */
  public KafkaStreamsAssignment withFollowupRebalance(final Instant rebalanceDeadline);

  public ProcessID processId();

  public Set<AssignedTask> assignment();

  /**
   * @return the actual deadline in objective time, after which the followup rebalance will be attempted.
   * Equivalent to {@code 'now + followupRebalanceDelay'}
   */
  public Instant followupRebalanceDeadline();

  public static class AssignedTask {

    public AssignedTask(final TaskId id, final Type taskType);

    enum Type {
        ACTIVE,
        STANDBY
    }
    
    public Type type();

    public TaskId id();
  }
}

Read-only APIs

The following APIs are intended for users to read/use but do not need to be implemented in order to plug in a custom assignor

ProcessID

The  ProcessId  is a new wrapper class around the UUID to make things easier to understand:

ProcessId
package org.apache.kafka.streams.processor.assignment; 

/** A simple wrapper around UUID that abstracts a Process Id */
public class ProcessId {

    public ProcessId(final UUID id) {
        this.id = id;
    }

    public id() {
        return id;
    }
}
 

KafkaStreamsState

Next we have the KafkaStreamsState  interface, representing the input to the assignor:

KafkaStreamsState
package org.apache.kafka.streams.processor.assignment;

/**
 * A read-only metadata class representing the current state of each KafkaStreams client with at least one StreamThread participating in this rebalance
 */
public interface KafkaStreamsState {
  /**
   * @return the processId of the application instance running on this KafkaStreams client
   */
  ProcessID processId();

  /**
   * Returns the number of processing threads available to work on tasks for this KafkaStreams client, 
   * which represents its overall capacity for work relative to other KafkaStreams clients.
   *
   * @return the number of processing threads on this KafkaStreams client
   */
  int numProcessingThreads();

  /**
   * @return the set of consumer client ids for this KafkaStreams client
   */
  SortedSet<String> consumerClientIds();

  /**
   * @return the set of all active tasks owned by consumers on this KafkaStreams client since the previous rebalance
   */
  SortedSet<TaskId> previousActiveTasks();

  /**
   * @return the set of all standby tasks owned by consumers on this KafkaStreams client since the previous rebalance
   */
  SortedSet<TaskId> previousStandbyTasks();

  /**
   * Returns the total lag across all logged stores in the task. Equal to the end offset sum if this client
   * did not have any state for this task on disk.
   *
   * @return end offset sum - offset sum
   *          Task.LATEST_OFFSET if this was previously an active running task on this client
   * @throws UnsupportedOperationException if the user did not request task lags be computed.
    */
  long lagFor(final TaskId task);

  /**
   * @return the previous tasks assigned to this consumer ordered by lag, filtered for any tasks that don't exist in this assignment
   * @throws UnsupportedOperationException if the user did not request task lags be computed.
   */
  SortedSet<TaskId> prevTasksByLag(final String consumerClientId);

  /**
   * Returns a collection containing all (and only) stateful tasks in the topology by {@link TaskId},
   * mapped to its "offset lag sum". This is computed as the difference between the changelog end offset
   * and the current offset, summed across all logged state stores in the task.
   *
   * @return a map from all stateful tasks to their lag sum
   * @throws UnsupportedOperationException if the user did not request task lags be computed.
   */
  Map<TaskId, Long> statefulTasksToLagSums();

  /**
   * The {@link HostInfo} of this KafkaStreams client, if set via the
   * {@link org.apache.kafka.streams.StreamsConfig#APPLICATION_SERVER_CONFIG application.server} config
   *
   * @return the host info for this KafkaStreams client if configured, else {@code Optional.empty()}
   */
  Optional<HostInfo> hostInfo();

  /**
   * The client tags for this KafkaStreams client, if set any have been via configs using the
   * {@link org.apache.kafka.streams.StreamsConfig#clientTagPrefix}
   * <p>
   * Can be used however you want, or passed in to enable the rack-aware standby task assignor.
   *
   * @return all the client tags found in this KafkaStreams client's {@link org.apache.kafka.streams.StreamsConfig}
   */
  Map<String, String> clientTags();

  /**
   * @return the rackId for this KafkaStreams client, or {@link Optional#empty()} if none was configured
   */
  Optional<String> rackId();

  }

ApplicationState

The KafkaStreamsState  will be wrapped up along with the other inputs to the assignor (such as the configuration and set of tasks to be assigned, as well as various utilities that may be useful) in the next new interface, the ApplicationState . The methods on the ApplicationState  are basically just the current inputs to the #assign method:

ApplicationState
package org.apache.kafka.streams.processor.assignment;

/**
 * A read-only metadata class representing the current state of each KafkaStreams client with at least one StreamThread participating in this rebalance
 */
public interface ApplicationState {
    /**
     * @param computeTaskLags whether or not to include task lag information in the returned metadata. Note that passing 
     * in "true" will result in a remote call to fetch changelog topic end offsets and you should pass in "false" unless
     * you specifically need the task lag information.
     *
     * @return a map from the {@code processId} to {@link KafkaStreamsState} for all KafkaStreams clients in this app
     *
     * @throws TaskAssignmentException if a retriable error occurs while computing KafkaStreamsState metadata. Re-throw
     *                                 this exception to have Kafka Streams retry the rebalance by returning the same
     *                                 assignment and scheduling an immediate followup rebalance
     */
    Map<ProcessID, KafkaStreamsState> kafkaStreamsStates(boolean computeTaskLags);

    /**
     * @return a simple container class with the Streams configs relevant to assignment
     */
    AssignmentConfigs assignmentConfigs();

    /**
     * @return the set of all tasks in this topology which must be assigned
     */
    Set<TaskInfo> allTasks();

}

TaskInfo

A small interface with metadata for each task to be assigned will be used to pass along information about stateful vs stateless tasks, the mapping of input and changelog topic partitions to tasks, and other essential info such as the rack ids for each topic partition belonging to a given task.

TaskInfo
/**
 * A simple container class corresponding to a given {@link TaskId}.
 * Includes metadata such as whether it's stateful and the names of all state stores
 * belonging to this task, the set of input topic partitions and changelog topic partitions
 * for all logged state stores, and the rack ids of all replicas of each topic partition
 * in the task.
 */
public interface TaskInfo {

    TaskId id();

    boolean isStateful();

    Set<String> stateStoreNames();

    Set<TopicPartition> inputTopicPartitions();

    Set<TopicPartition> changelogTopicPartitions();

    Map<TopicPartition, Set<String>> partitionToRackIds();
}


TaskAssignmentUtils

We'll also move some of the existing assignment functionality into a utils class that can be called by implementors of the new TaskAssignor . This will allow users to more easily adapt or modify pieces of the complex existing assignment algorithm, without having to re-implement the entire thing from scratch. 

TaskAssignmentUtils
package org.apache.kafka.streams.processor.assignment;

/**
 * A set of utilities to help implement task assignment
 */
public final class TaskAssignmentUtils {
    /**
     * Assign standby tasks to KafkaStreams clients according to the default logic.
     * <p>
     * If rack-aware client tags are configured, the rack-aware standby task assignor will be used
     *
     * @param applicationState        the metadata and other info describing the current application state
     * @param KafkaStreamsAssignments the current assignment of tasks to KafkaStreams clients
     *
     * @return a new map containing the mappings from KafkaStreamsAssignments updated with the default standby assignment
     */
    public static Map<ProcessID, KafkaStreamsAssignment> defaultStandbyTaskAssignment(final ApplicationState applicationState, 
                                                                                      final Map<ProcessID, KafkaStreamsAssignment> KafkaStreamsAssignments);

    /**
     * Optimize the active task assignment for rack-awareness
     *
     * @param applicationState        the metadata and other info describing the current application state
     * @param kafkaStreamsAssignments the current assignment of tasks to KafkaStreams clients
     * @param tasks                   the set of tasks to reassign if possible. Must already be assigned to a KafkaStreams client
     *
     * @return a new map containing the mappings from KafkaStreamsAssignments updated with the default rack-aware assignment for active tasks
     */
    public static Map<ProcessID, KafkaStreamsAssignment> optimizeRackAwareActiveTasks(final ApplicationState applicationState, 
                                                                                      final Map<ProcessID, KafkaStreamsAssignment> kafkaStreamsAssignments, 
                                                                                      final SortedSet<TaskId> tasks);

    /**
     * Optimize the standby task assignment for rack-awareness
     *
     * @param KafkaStreamsAssignments the current assignment of tasks to KafkaStreams clients
     * @param applicationState        the metadata and other info describing the current application state
     *
     * @return a new map containing the mappings from KafkaStreamsAssignments updated with the default rack-aware assignment for standy tasks
     */
    public static Map<ProcessID, KafkaStreamsAssignment> optimizeRackAwareStandbyTasks(final ApplicationState applicationState,
                                                                                       final Map<ProcessID, KafkaStreamsAssignment> kafkaStreamsAssignments);

    /**
     * Return a "no-op" assignment that just copies the previous assignment of tasks to KafkaStreams clients
     *
     * @param applicationState the metadata and other info describing the current application state
     *
     * @return a new map containing an assignment that replicates exactly the previous assignment reported in the applicationState
     */
    public static Map<ProcessID, KafkaStreamsAssignment> identityAssignment(final ApplicationState applicationState);
 }

TaskAssignmentUtils  provides new APIs but pre-existing functionality, essentially presenting a clean way for users to take advantage of the current optimizations and algorithms that are utilized by the built-in assignors, so that users don't have to re-implement complex features such as rack-awareness. The #defaultStandbyTaskAssignment API will just delegate to the appropriate standby task assignor (either basic default or client tag based standby rack awareness, depending on the existence of client tags in the configuration). Similarly, the #optimizeRackAware{Active/Standby}Tasks API will just delegate to the new RackAwareTaskAssignor that is being added in KIP-925.

AssignmentConfigs

Last, we have the AssignmentConfigs, which are (and would remain) just a basic container class, although we will migrate from public fields to standard getters for each of the configs passed into the assignor. Going forward, when a KIP is proposed to introduce a new config intended for the assignor, it should include the appropriate getter(s) in this class as part of the accepted proposal.

AssignmentConfigs
package org.apache.kafka.streams.processor.assignment;

public class AssignmentConfigs {
    public long acceptableRecoveryLag();
    public int maxWarmupReplicas();
    public int numStandbyReplicas();
    public long probingRebalanceIntervalMs();
    public List<String> rackAwareAssignmentTags();
    public int trafficCost();
    public int nonOverlapCost();
}


Finally, as part of this change, we're moving some of the behavior that can fail into the task assignor. In particular, we're moving the bits that compute lags for stateful tasks into the implementation of ApplicationState#kafkaStreamsStates . Users who request the task lags via the computeTaskLags  input flag should make sure to handle failures the way they desire, and can rethrow a thrown TaskAssignmentException  (or just not catch it in the first place) to have Kafka Streams automatically "retry" the rebalance by returning the same assignment and scheduling an immediate followup rebalance. Advanced users who want more control over the "fallback" assignment and/or the timing of immediate followup rebalance(s) can simply swallow the TaskAssignmentException  and use the followupRebalanceDeadline  to schedule followup rebalances, eg to implement a retry/backoff policy


Proposed Changes

On the whole we are not introducing new functionality to Streams itself, but rather are refactoring the existing TaskAssignor interface so that it's (a) pluggable/publicly exposed, and (b) easier to understand, use, and implement. Code-wise the largest change is the breaking up of the ClientState into the new KafkaStreamsState and KafkaStreamsAssignment interfaces, but that will be handled transparently to the user for all existing built-in-assignors, which will continue to work the same as before. 

Observing The Final Assignment

The custom assignor specifies a task to KafkaStreams client mapping, but the final task to group member mapping is controlled by the Streams partition assignor. It may be useful for implementors of the custom assignor to observe the final task to group member assignment that is returned back to Kafka. For example, it may be useful to record the mapping in an event that can be materialized externally to observe the current assignments. To support this, the TaskAssignor  interface also includes a method named #onAssignmentComputed which is called with the final computed GroupAssignment.

This callback can also be used to return an error to the user and notify them in the case of an invalid assignment computed by their TaskAssignor. The specifics are discussed in the following section.

Assignment Validation

As noted in the TaskAssignor  javadocs, the StreamsPartitionAssignor will verify the assignment returned by the task assignor and return an error via #onAssignmentComputed  if any of the following cases are observed while processing the TaskAssignor 's assignment:

  1. ACTIVE_TASK_ASSIGNED_MULTIPLE_TIMES :  multiple KafkaStreams clients assigned with the same active task
  2. ACTIVE_AND_STANDBY_TASK_ASSIGNED_TO_SAME_KAFKASTREAMS : active task and standby task assigned to the same KafkaStreams client
  3. INVALID_STANDBY_TASK: stateless task assigned as a standby task
  4. UNKNOWN_PROCESS_ID : unrecognized ProcessId  not matching any of the participating consumers
  5. UNKNOWN_TASK_ID: unrecognized TaskId not matching any of the tasks to be assigned

If any of these errors are detected, the StreamsPartitionAssignor will throw an exception after returning the error code via the #onAssignmentComputed  callback. This error will be bubbled up through the StreamThread to the uncaught exception handler where the user can choose how to react from there, same as any other exception.

If no error is detected, the AssignmentError  code NONE  will be returned in the #onAssignmentComputed  callback.

Consumer Assignments

One major decision in this KIP was whether to encompass the assignment of tasks to consumers/threads within each KafkaStreams client, or to leave that up to the StreamsPartitionAssignor and only carve out the KafkaStreams-level assignment for pluggability. Ultimately we decided on the latter, for several reasons:

  1. To keep this KIP looking towards the future: we are working on moving away from the consumer-per-StreamThread model and hope to soon have only a single consumer per KafkaStreams client. If we only have to assign tasks at the KafkaStreams client level, we can implement this new feature completely transparently and users who plug in a custom task assignor won't have to do anything to adapt, and we don't need to deprecate any APIs in the assignor after this change.
  2. To shield users from subtle, difficult, and/or internal protocol requirements and constraints: for example with cooperative rebalancing, it is vital that the assignor make sure not to revoke and reassign a partition to a new consumer within the same rebalance. This is a non-trivial constraint and would be easy for users to mess up, and further, we don't want users to have to worry about protocol-level details.
  3. To retain control over the final assignment: right now we apply some niche optimizations/bugfixes like assigning transient standbys to avoid losing in-memory state, as well as some other subtle logic that currently resides in the last leg of the StreamsPartitionAssignor's algorithm that tackles the distribution of KafkaStreams client tasks to threads. We don't want users to have to implement all of this again or else risk running into the same bugs and regressions that we've already found and fixed based on the past years of operating the StreamsPartitionAssignor.
  4. To keep things simple: this is frankly already a complicated enough feature, and adding consumer-level assignment only makes things worse. We didn't see any practical application for adjusting the thread-level assignment since it pretty much always makes sense to do the "sticky-but-balanced" assignment that Streams currently applies. In the end, while there have been many problems and complaints with the current KafkaStreams client assignment, we haven't seen or heard anyone request control over the thread-level assignment and could not think of any use case for this ourselves, and so it simply does not feel worth the added complexity.

This topic is touched on again in the Rejective Alternatives section

Compatibility, Deprecation, and Migration Plan

Since this was formally an internal config and not part of the public API, we don't need to go through the usual deprecation path. See "Rejected Alternatives" for some slightly more nuanced discussion here

Test Plan

Mostly nothing to report here as there should already be tests in place for this config, however I will check the existing test coverage during implementation and fill in any gaps as needed to make sure it's possible to set either of the OOTB assignors (HA or Sticky) as well as a custom assignor.

Rejected Alternatives

  1. One obvious question here is whether we want to still deprecate the old internal config anyway, out of compassion for any who may already be using it despite it not being considered public. Personally I think this would be reasonable but don't feel strongly one way or another.
  2. Another possibility that was considered and ultimately decided against was whether to encompass the thread-level assignment in this KIP, and bring that into the public API and make it pluggable as well. We determined that this did not seem necessary to do as part of the initial KIP, especially considering the large scope we have already reached. However it's worth noting that a followup KIP that builds on the new public API(s) introduced here would become much more feasible should someone wish to customize the thread-level logic at some point in the future.



  • No labels