Akka DDDD template using CQRS/ES with a Distributed Domain
Scala Version = 2.11.6
Akka Version = 2.3.9
Spray Version = 1.3.1
Distributed Domain Driven Design takes the existing DDD concept and applies it ot an application intended to have each domain instance represented by an actor, as opposed to a class instance that must be synchronized via a backing persistent store. In this pattern, each domain instance is a cluster singleton, meaning updates can be made to it's state without fear of conflict.
This is a pattern that uses Command and Query objects to apply the CQS principle for modifying and retrieving data.
Event Sourcing is an architectural pattern in which state is tracked with an immutable event log instead of destructive updates (mutable).
To get this application going, you will need to:
- Set up the datastore
- Boot the cluster nodes
- Boot the Http microservice node
This application requires a distributed journal. Storage backends for journals and snapshot stores are pluggable in Akka persistence. In this case we are using Cassandra. You can find other journal plugins here.
The datastore is specified in application.conf
cassandra-journal.contact-points = ["127.0.0.1"]
cassandra-snapshot-store.contact-points = ["127.0.0.1"]
As you can see, the default is localhost. In a cloud deployment, you could add several addresses to a cassandra cluster.
This application uses a simple domain to demonstrate CQRS and event sourcing with Akka Persistence. This domain is an online auction:
final case class Bid(price:Double, buyer:String, timeStamp:Long)
final case class Auction(auctionId:String,
startTime:Long,
endTime:Long,
initialPrice:Double,
acceptedBids:List[Bid],
refusedBids:List[Bid],
ended:Boolean)
This is a distributed application, leveraging Akka Cluster.
The Command path of this application is illustrated by the creation of an auction, and placing bids.
The Query path of this application is illustrated by the querying of winning bid and bid history.
In order to distribute and segregate these paths, we leverage Akka Cluster, as well as Cluster Sharding.
Cluster Sharding enables the distribution of the command and query actors across several nodes in the cluster, supporting interaction using their logical identifier, without having to care about their physical location in the cluster.
You must first boot some cluster nodes (as many as you want). Running locally, these are distinguished by port eg:[2551,2552,...].
This cluster must specify one or more seed nodes in application.conf
akka.cluster {
seed-nodes = [
"akka.tcp://[email protected]:2551",
"akka.tcp://[email protected]:2552"]
}
The Cluster Nodes are bootstrapped in ClusterNode.scala.
To boot each cluster node locally:
sbt 'runMain com.boldradius.cqrs.ClusterNodeApp nodeIpAddress port'
for example:
sbt 'runMain com.boldradius.cqrs.ClusterNodeApp 127.0.0.1 2551'
The HTTP front end is implemented as a Spray microservice and is bootstrapped in HttpApp.scala.It participates in the Cluster, but as a proxy.
To run the microservice locally:
sbt 'runMain com.boldradius.cqrs.HttpApp httpIpAddress httpPort akkaIpAddres akkaPort'
for example:
sbt 'runMain com.boldradius.cqrs.HttpApp 127.0.0.1 9000 127.0.0.1 0'
The HTTP API enables the user to:
- Create an Auction
- Place a bid
- Query for the current winning bid
- Query for the bid history
POST http://127.0.0.1:9000/startAuction
{"auctionId":"123",
"start":"2015-01-20-16:25",
"end":"2015-07-20-16:35",
"initialPrice" : 2,
"prodId" : "3"}
POST http://127.0.0.1:9000/bid
{"auctionId":"123",
"buyer":"dave",
"bidPrice":6}
GET http://127.0.0.1:9000/winningBid/123
http://127.0.0.1:9000/bidHistory/123
The trait HttpAuctionServiceRoute.scala implements a route that takes ActorRefs (one for command and query) as input. Upon receiving an Http request, it either sends a command message to the command actor, or a query message to the query actor.
def route(command: ActorRef, query:ActorRef) = {
post {
path("startAuction") {
extract(_.request) { e =>
entity(as[StartAuctionDto]) {
auction => onComplete(
(command ? StartAuctionCmd(auction.auctionId,....
The command path is implemented in BidProcessor.scala. This is a PersistentActor that receives commands:
def initial: Receive = {
case a@StartAuctionCmd(id, start, end, initialPrice, prodId) => ...
}
def takingBids(state: Auction): Receive = {
case a@PlaceBidCmd(id, buyer, bidPrice) => ...
}
and produces events, writing them to the event journal, and notifying the Query Path of the updated journal:
def handleProcessedCommand(sendr: ActorRef, processedCommand: ProcessedCommand): Unit = {
// ack whether there is an event or not
processedCommand.event.fold(sender() ! processedCommand.ack) { evt =>
persist(evt) { persistedEvt =>
readRegion ! Update(await = true)
sendr ! processedCommand.ack
processedCommand.newReceive.fold({})(context.become)
}
}
}
This actor is cluster sharded on auctionId as follows:
val idExtractor: ShardRegion.IdExtractor = {
case m: AuctionCmd => (m.auctionId, m)
}
val shardResolver: ShardRegion.ShardResolver = msg => msg match {
case m: AuctionCmd => (math.abs(m.auctionId.hashCode) % 100).toString
}
val shardName: String = "BidProcessor"
This means, there is only one instance of this actor in the cluster, and all commands with the same auctionId will be routed to the same actor.
If this actor receives no commands for 1 minute, it will passivate ( a pattern enabling the parent to stop the actor, in order to reduce memory consumption without losing any commands it is currently processing):
/** passivate the entity when no activity */
context.setReceiveTimeout(1 minute) // this will send a ReceiveTimeout message after one minute, if no other messages come in
The timeout is handled in the Passivation.scala trait:
protected def withPassivation(receive: Receive): Receive = receive.orElse{
// tell parent actor to send us a poisinpill
case ReceiveTimeout => context.parent ! Passivate(stopMessage = PoisonPill)
// stop
case PoisonPill => context.stop(self)
}
If this actor fails, or is passivated, and then is required again (to handle a command), the cluster will spin it up, and it will replay the event journal. In this case we make use a var: auctionRecoverStateMaybe to capture the state while we replay. When the replay is finished, the actor is notified with the RecoveryCompleted message and we can then "become" appropriately to reflect this state.
def receiveRecover: Receive = {
case evt:AuctionStartedEvt =>
auctionRecoverStateMaybe = Some(Auction(evt.auctionId,evt.started,evt.end,evt.initialPrice,Nil,Nil,false))
case evt: AuctionEvt => {
auctionRecoverStateMaybe = auctionRecoverStateMaybe.map(state =>
updateState(evt.logInfo("receiveRecover" + _.toString),state))
}
// Once recovery is complete, check the state to become the appropriate behaviour
case RecoveryCompleted => {
auctionRecoverStateMaybe.fold[Unit]({}) { auctionState =>
if (auctionState.logInfo("receiveRecover RecoveryCompleted state: " + _.toString).ended)
context.become(passivate(auctionClosed(auctionState)).orElse(unknownCommand))
else {
launchLifetime(auctionState.endTime)
context.become(passivate(takingBids(auctionState)).orElse(unknownCommand))
}
}
}
The Queries are handled in a different Actor: BidView.scala. This is a PersistentView that handles query messages, or prompts from it's companion PersistentActor to update itself.
BidView.scala is linked to the BidProcessor.scala event journal via it's persistenceId
override val persistenceId: String = "BidProcessor" + "-" + self.path.name
This means it has access to this event journal, and can maintain, and recover state from this journal.
It is possible for a PersistentView to save it's own snapshots, but, in our case, it isn't required.
This PersistentView is sharded in the same way the PersistentActor is:
val idExtractor: ShardRegion.IdExtractor = {
case m : AuctionEvt => (m.auctionId,m)
case m : BidQuery => (m.auctionId,m)
}
val shardResolver: ShardRegion.ShardResolver = {
case m: AuctionEvt => (math.abs(m.auctionId.hashCode) % 100).toString
case m: BidQuery => (math.abs(m.auctionId.hashCode) % 100).toString
}
One could have used a different shard strategy here, but a consequence of the above strategy is that the Query Path will reside in the same Shard Region as the command path, reducing latency of the Update() message from Command to Query.
The PersistentView maintains the following model in memory:
final case class BidState(auctionId:String,
start:Long,
end:Long,
product:Double,
acceptedBids:List[Bid],
rejectedBids:List[Bid],
closed:Boolean)
This model is sufficient to satisfy both queries: Winning Bid, and Bid History:
def auctionInProgress(currentState:BidState, prodId:String):Receive = {
case GetBidHistoryQuery(auctionId) => sender ! BidHistoryResponse(auctionId,currentState.acceptedBids)
case WinningBidPriceQuery(auctionId) =>
currentState.acceptedBids.headOption.fold(
sender ! WinningBidPriceResponse(auctionId,currentState.product))(b =>
sender ! WinningBidPriceResponse(auctionId,b.price))
....
}