Merge branch 'master' into refactor-snapshot-store-failure-spec

Author: xirc

CHANGELOG.md

@@ -15,6 +15,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   [PR#164](https://github.com/lerna-stack/akka-entity-replication/pull/164)
 - Add function extracting shard id from entity id to lerna.akka.entityreplication.typed.ClusterReplication
   [PR#172](https://github.com/lerna-stack/akka-entity-replication/pull/172)
+- Add function of Disabling raft actor [PR#173](https://github.com/lerna-stack/akka-entity-replication/pull/173)
 
 ### Fixed
 - RaftActor might delete committed entries

src/main/resources/reference.conf

@@ -22,6 +22,10 @@ lerna.akka.entityreplication {
     // Changing this value will cause data inconsistency.
     number-of-shards = 100
 
+    // Shard Ids of raft actors to disable.
+    // e.g. ["2", "5"]
+    disabled-shards = []
+
     // Maximum number of entries which AppendEntries contains.
     // The too large size will cause message serialization failure.
     max-append-entries-size = 16
@@ -111,6 +115,7 @@ lerna.akka.entityreplication {
       // Snapshot synchronization reads events that related to Raft.
       query.plugin = ""
     }
+
   }
 
   raft.eventsourced {

src/main/scala/lerna/akka/entityreplication/ReplicationRegion.scala

@@ -137,6 +137,8 @@ private[entityreplication] class ReplicationRegion(
   private[this] val regions: Map[MemberIndex, mutable.Set[Member]] =
     allMemberIndexes.map(i => i -> mutable.Set.empty[Member]).toMap
 
+  private val disabledShards: Set[ShardId] = settings.raftSettings.disabledShards
+
   // TODO 変数名を実態にあったものに変更
   private[this] val shardingRouters: Map[MemberIndex, ActorRef] = allMemberIndexes.map { memberIndex =>
     def clusterReplicationShardId(message: Any): String = extractNormalizedShardIdInternal(message).raw
@@ -255,11 +257,19 @@ private[entityreplication] class ReplicationRegion(
   def deliverMessage(message: Any): Unit = {
     if (extractShardId.isDefinedAt(message)) {
       val shardId = extractShardId(message)
-      shardingRouters.values.foreach(
-        // Don't forward StartEntity to prevent leaking StartEntityAck
-        _.tell(ShardRegion.StartEntity(shardId), context.system.deadLetters),
-      )
-      handleRoutingCommand(DeliverSomewhere(Command(message)))
+      if (!disabledShards.contains(shardId)) {
+        shardingRouters.values.foreach(
+          // Don't forward StartEntity to prevent leaking StartEntityAck
+          _.tell(ShardRegion.StartEntity(shardId), context.system.deadLetters),
+        )
+        handleRoutingCommand(DeliverSomewhere(Command(message)))
+      } else if (log.isWarningEnabled) {
+        log.warning(
+          s"Following command had sent to disabled shards was dropped: {}(shardId={})",
+          message.getClass.getName,
+          shardId,
+        )
+      }
     } else {
       if (log.isWarningEnabled)
         log.warning("The message [{}] was dropped because its shard ID could not be extracted", message)

src/main/scala/lerna/akka/entityreplication/ReplicationRegionRaftActorStarter.scala

@@ -35,6 +35,12 @@ private[entityreplication] object ReplicationRegionRaftActorStarter {
   ): Behavior[Nothing] = {
     Behaviors
       .setup[Command] { context =>
+        val (disableIds, enableIds) = ids.partition(settings.disabledShards.contains)
+        if (disableIds.nonEmpty) {
+          context.log.info(
+            s"Following disabled shard ids are excluded from ids which is scheduled launching: ${disableIds}",
+          )
+        }
         val startEntityAckAdapter =
           context.messageAdapter[ShardRegion.StartEntityAck](ack => ClassicStartEntityAck(ack.entityId))
         Behaviors.withTimers { timers =>
@@ -46,7 +52,7 @@ private[entityreplication] object ReplicationRegionRaftActorStarter {
             startEntityAckAdapter,
             settings.raftActorAutoStartNumberOfActors,
           )
-          starter.behavior(ids, Set.empty)
+          starter.behavior(enableIds, Set.empty)
         }
       }.narrow[Nothing]
   }

src/main/scala/lerna/akka/entityreplication/raft/RaftActor.scala

@@ -1,7 +1,7 @@
 package lerna.akka.entityreplication.raft
 
 import akka.actor.{ ActorRef, Cancellable, Props, Stash }
-import akka.persistence.RuntimePluginConfig
+import akka.persistence.{ Recovery, RuntimePluginConfig }
 import com.typesafe.config.{ Config, ConfigFactory }
 import lerna.akka.entityreplication.ClusterReplication.EntityPropsProvider
 import lerna.akka.entityreplication.ReplicationRegion.Msg
@@ -214,6 +214,22 @@ private[raft] class RaftActor(
 
   override def snapshotPluginConfig: Config = ConfigFactory.empty()
 
+  private def isDisabled: Boolean = settings.disabledShards.contains(shardId.raw)
+
+  override def recovery: Recovery = {
+    if (isDisabled) {
+      Recovery.none
+    } else {
+      Recovery()
+    }
+  }
+
+  override def preStart(): Unit = {
+    if (isDisabled) {
+      context.stop(self)
+    }
+  }
+
   val numberOfMembers: Int = settings.replicationFactor
 
   @nowarn("msg=Use RaftMemberData.truncateAndAppendEntries instead.")

src/main/scala/lerna/akka/entityreplication/raft/RaftSettings.scala

@@ -28,6 +28,8 @@ trait RaftSettings {
 
   def numberOfShards: Int
 
+  def disabledShards: Set[String]
+
   def maxAppendEntriesSize: Int
 
   def maxAppendEntriesBatchSize: Int

src/main/scala/lerna/akka/entityreplication/raft/RaftSettingsImpl.scala

@@ -17,6 +17,7 @@ private[entityreplication] final case class RaftSettingsImpl(
     replicationFactor: Int,
     quorumSize: Int,
     numberOfShards: Int,
+    disabledShards: Set[String],
     maxAppendEntriesSize: Int,
     maxAppendEntriesBatchSize: Int,
     compactionSnapshotCacheTimeToLive: FiniteDuration,
@@ -215,6 +216,8 @@ private[entityreplication] object RaftSettingsImpl {
       s"snapshot-every ($eventSourcedSnapshotEvery) should be greater than 0.",
     )
 
+    val disabledShards: Set[String] = config.getStringList("disabled-shards").asScala.toSet
+
     RaftSettingsImpl(
       config = config,
       electionTimeout = electionTimeout,
@@ -224,6 +227,7 @@ private[entityreplication] object RaftSettingsImpl {
       replicationFactor = replicationFactor,
       quorumSize = quorumSize,
       numberOfShards = numberOfShards,
+      disabledShards = disabledShards,
       maxAppendEntriesSize = maxAppendEntriesSize,
       maxAppendEntriesBatchSize = maxAppendEntriesBatchSize,
       compactionSnapshotCacheTimeToLive = compactionSnapshotCacheTimeToLive,

src/main/scala/lerna/akka/entityreplication/raft/eventsourced/CommitLogStoreActor.scala

@@ -4,7 +4,14 @@ import akka.Done
 import akka.actor.{ ActorLogging, ActorRef, ActorSystem, Props }
 import akka.cluster.sharding.ShardRegion.HashCodeMessageExtractor
 import akka.cluster.sharding.{ ClusterSharding, ClusterShardingSettings }
-import akka.persistence.{ PersistentActor, RecoveryCompleted, SaveSnapshotFailure, SaveSnapshotSuccess, SnapshotOffer }
+import akka.persistence.{
+  PersistentActor,
+  Recovery,
+  RecoveryCompleted,
+  SaveSnapshotFailure,
+  SaveSnapshotSuccess,
+  SnapshotOffer,
+}
 import akka.util.ByteString
 import lerna.akka.entityreplication.{ ClusterReplicationSerializable, ClusterReplicationSettings }
 import lerna.akka.entityreplication.model.{ NormalizedShardId, TypeName }
@@ -307,4 +314,19 @@ private[entityreplication] class CommitLogStoreActor(typeName: TypeName, setting
     }
   }
 
+  private def isDisabled: Boolean = settings.raftSettings.disabledShards.contains(shardId)
+
+  override def recovery: Recovery = {
+    if (isDisabled) {
+      Recovery.none
+    } else {
+      Recovery()
+    }
+  }
+
+  override def preStart(): Unit = {
+    if (isDisabled) {
+      context.stop(self)
+    }
+  }
 }

src/multi-jvm/scala/lerna/akka/entityreplication/ReplicationRegionSpec.scala

@@ -1,6 +1,8 @@
 package lerna.akka.entityreplication
 
 import akka.Done
+import akka.actor.testkit.typed.scaladsl.LoggingTestKit
+import akka.actor.typed.scaladsl.adapter.ClassicActorSystemOps
 
 import java.util.concurrent.atomic.AtomicInteger
 import akka.actor.{ Actor, ActorRef, ActorSelection, DiagnosticActorLogging, Props, RootActorPath, Terminated }
@@ -120,6 +122,7 @@ object ReplicationRegionSpecConfig extends MultiNodeConfig {
       lerna.akka.entityreplication.raft.compaction.log-size-threshold = 2
       lerna.akka.entityreplication.raft.compaction.preserve-log-size = 1
       lerna.akka.entityreplication.raft.compaction.log-size-check-interval = 0.1s
+      lerna.akka.entityreplication.raft.disabled-shards = ["12"]
       """))
       .withValue(
         "lerna.akka.entityreplication.raft.multi-raft-roles",
@@ -334,6 +337,27 @@ class ReplicationRegionSpec extends MultiNodeSpec(ReplicationRegionSpecConfig) w
       }
     }
 
+    "drop all messages sent to disabled shards" in {
+      val typeName = createSeqTypeName()
+
+      runOn(node4, node5, node6) {
+        clusterReplication = createReplication(typeName)
+      }
+      enterBarrier("ReplicationRegion created")
+
+      // "disable-id"'s shard id is "12" and the shard id have been defined as disabled id.
+      val entityId = "disabled-id"
+      runOn(node4) {
+        LoggingTestKit
+          .warn(
+            "Following command had sent to disabled shards was dropped: lerna.akka.entityreplication.ReplicationRegionSpec$DummyReplicationActor$Cmd(shardId=12)",
+          ).expect {
+            clusterReplication ! Cmd(entityId)
+          }(system.toTyped)
+      }
+      enterBarrier("Some command sent")
+    }
+
     "Broadcast" when {
 
       "全 MemberIndex に配信される" in {

src/test/scala/lerna/akka/entityreplication/ReplicationRegionRaftActorStarterSpec.scala

@@ -74,6 +74,33 @@ final class ReplicationRegionRaftActorStarterSpec
 
     }
 
+    "trigger all actor starts without disabled actor" in {
+      val shardRegionProbe = TestProbe()
+
+      val customSettings = RaftSettings(
+        ConfigFactory
+          .parseString("""
+              |lerna.akka.entityreplication.raft.disabled-shards = ["2"]
+              |""".stripMargin).withFallback(system.settings.config),
+      )
+      val raftActorStarter =
+        spawnRaftActorStarter(shardRegionProbe.ref, Set("1", "2", "3"), customSettings)
+
+      assume(customSettings.raftActorAutoStartNumberOfActors >= 3)
+
+      // The starter will stop at the end of this test.
+      watch(raftActorStarter)
+
+      // The starter should trigger all actor starts without disabled shards.
+      val startedRaftActorIds = (1 to 2).map { _ =>
+        expectStartEntityAndThenAck(shardRegionProbe)
+      }.toSet
+      startedRaftActorIds shouldBe Set("1", "3")
+
+      // The starter should stop itself.
+      expectTerminated(raftActorStarter)
+    }
+
     "retry all actor starts with no ACK" in {
 
       val shardRegionProbe = TestProbe()