Sharded Daemon Process

Module info

To use Pekko Sharded Daemon Process, you must add the following dependency in your project:

sbt

val PekkoVersion = "2.0.0-M3+229-dd5e07f6-SNAPSHOT"
libraryDependencies += "org.apache.pekko" %% "pekko-cluster-sharding-typed" % PekkoVersion

Maven

<properties>
  <scala.binary.version>2.13</scala.binary.version>
</properties>
<dependencyManagement>
  <dependencies>
    <dependency>
      <groupId>org.apache.pekko</groupId>
      <artifactId>pekko-bom_${scala.binary.version}</artifactId>
      <version>2.0.0-M3+229-dd5e07f6-SNAPSHOT</version>
      <type>pom</type>
      <scope>import</scope>
    </dependency>
  </dependencies>
</dependencyManagement>
<dependencies>
  <dependency>
    <groupId>org.apache.pekko</groupId>
    <artifactId>pekko-cluster-sharding-typed_${scala.binary.version}</artifactId>
  </dependency>
</dependencies>

Gradle

def versions = [
  ScalaBinary: "2.13"
]
dependencies {
  implementation platform("org.apache.pekko:pekko-bom_${versions.ScalaBinary}:2.0.0-M3+229-dd5e07f6-SNAPSHOT")

  implementation "org.apache.pekko:pekko-cluster-sharding-typed_${versions.ScalaBinary}"
}

Project Info: Pekko Cluster Sharding (typed)
Artifact	org.apache.pekko pekko-cluster-sharding-typed 2.0.0-M3+229-dd5e07f6-SNAPSHOT Snapshots are available
JDK versions	OpenJDK 17 OpenJDK 21 OpenJDK 25
Scala versions	2.13.18, 3.3.8
JPMS module name	pekko.cluster.sharding.typed
License	Apache-2.0
Home page	https://pekko.apache.org/
API documentation	API (Scaladoc) API (Javadoc)
Forums	Apache Pekko User mailing list Apache Pekko Dev mailing list apache/pekko discussion
Release notes	Release Notes
Issues	Github issues
Sources	https://github.com/apache/pekko

Introduction

Sharded Daemon Process provides a way to run N actors, each given a numeric id starting from 0 that are then kept alive and balanced across the cluster. When a rebalance is needed the actor is stopped and, triggered by a keep alive from a Cluster Singleton (the keep alive should be seen as an implementation detail and may change in future versions).

The intended use case is for splitting data processing workloads across a set number of workers that each get to work on a subset of the data that needs to be processed. This is commonly needed to create projections based on the event streams available from all the EventSourcedBehaviors in a CQRS application. Events are tagged with one out of N tags used to split the workload of consuming and updating a projection between N workers.

For cases where a single actor needs to be kept alive see Cluster Singleton

Basic example

To set up a set of actors running with Sharded Daemon process each node in the cluster needs to run the same initialization when starting up:

Scala

sourceval tags = Vector("tag-1", "tag-2", "tag-3")
ShardedDaemonProcess(system).init("TagProcessors", tags.size, id => TagProcessor(tags(id)))

Java

sourceList<String> tags = List.of("tag-1", "tag-2", "tag-3");
ShardedDaemonProcess.get(system)
    .init(
        TagProcessor.Command.class,
        "TagProcessors",
        tags.size(),
        id -> TagProcessor.create(tags.get(id)));

An additional factory method is provided for further configurability and providing a graceful stop message for the actor.

Addressing the actors

In use cases where you need to send messages to the daemon process actors it is recommended to use the system receptionist either with a single ServiceKey which all daemon process actors register themeselves to for broadcasts or individual keys if more fine grained messaging is needed.

Dynamic scaling of number of workers

Starting the sharded daemon process with initWithContext returns an ActorRef[ShardedDaemonProcessCommand] that accepts a ChangeNumberOfProcessesChangeNumberOfProcesses command to rescale the process to a new number of workers.

The rescaling process among other things includes the process actors stopping themselves in response to a stop message so may be a relatively slow operation. If a subsequent request to rescale is sent while one is in progress it is responded to with a failure response.

A rolling upgrade switching from a static number of workers to a dynamic number is possible. It is not safe to do a rolling upgrade from dynamic number of workers to static without a full cluster shutdown.

Colocate processes

When using the default shard allocation strategy the processes for different names are allocated independent of each other, i.e. the same process index for different process names may be allocated to different nodes. Colocating processes can be useful to share resources, such as Projections with EventsBySliceFirehoseQuery

To colocate such processes you can use the ConsistentHashingShardAllocationStrategyConsistentHashingShardAllocationStrategy as shardAllocationStrategy parameter of the init or initWithContext methods.

Note

Create a new instance of the ConsistentHashingShardAllocationStrategy for each Sharded Daemon Process name, i.e. a ConsistentHashingShardAllocationStrategy instance must not be shared.

The shard identifier that is used by Sharded Daemon Process is the same as the process index, i.e. processes with the same index will be colocated.

The allocation strategy is using Consistent Hashing of the Cluster membership ring to assign a shard to a node. When adding or removing nodes it will rebalance according to the new consistent hashing, but that means that only a few shards will be rebalanced and others remain on the same location. When there are changes to the Cluster membership the shards may be on different nodes for a while, but eventually, when the membership is stable, the shards with the same identifier will end up on the same node.

Scalability

This cluster tool is intended for small numbers of consumers and will not scale well to a large set. In large clusters it is recommended to limit the nodes the sharded daemon process will run on using a role.

Edit this page

2.0.0-M3+229-dd5e07f6*