🔀Merge branch 'master' into fix-lost-commited-events-by-leader-transfer

Author: negokaz

CHANGELOG.md

@@ -31,6 +31,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - The Raft leader uses the same previous `LogEntryIndex` and `Term` to all batched `AppendEntries` messages [#123](https://github.com/lerna-stack/akka-entity-replication/issues/123)
 - Raft Actors doesn't accept a `RequestVote(lastLogIndex < log.lastLogIndex, lastLogTerm > log.lastLogTerm)` message [#125](https://github.com/lerna-stack/akka-entity-replication/issues/125)
 - A new event is created even though all past events have not been applied [#130](https://github.com/lerna-stack/akka-entity-replication/issues/130)
+- `InstallSnapshot` can miss snapshots to copy [PR#128](https://github.com/lerna-stack/akka-entity-replication/pull/128)
+
+  ⚠️ This change adds a new persistence event. This might don't allow downgrading after upgrading.
 - Moving a leader during snapshot synchronization can delete committed log entries [#133](https://github.com/lerna-stack/akka-entity-replication/issues/133)
 
   ⚠️ This change adds a new persistence event. This might don't allow downgrading after upgrading.

build.sbt

@@ -55,6 +55,7 @@ lazy val lerna = (project in file("."))
         // TODO 2.6.x 系に対応できる方法に変更する。
         "com.github.dnvriend" %% "akka-persistence-inmemory" % "2.5.15.2"  % Test,
         "com.typesafe.akka"   %% "akka-persistence-testkit"  % akkaVersion % Test,
+        "com.typesafe.akka"   %% "akka-stream-testkit"       % akkaVersion % Test,
       ),
     inConfig(MultiJvm)(
       // multi-jvm ディレクトリをフォーマットするために必要

src/main/mima-filters/2.0.0.backwards.excludes/pr-128-fix-inconsistency-after-installsnapshot.excludes

@@ -0,0 +1,8 @@
+# It is safe to exclude the following since CompactionCompletedTag (that is renamed to EntitySnapshotsUpdatedTag) is package-private.
+ProblemFilters.exclude[MissingClassProblem]("lerna.akka.entityreplication.raft.persistence.CompactionCompletedTag")
+ProblemFilters.exclude[MissingClassProblem]("lerna.akka.entityreplication.raft.persistence.CompactionCompletedTag$")
+# It is safe to exclude the following since SnapshotSyncManager is package-private.
+ProblemFilters.exclude[MissingClassProblem]("lerna.akka.entityreplication.raft.snapshot.sync.SnapshotSyncManager$CompactionEnvelope")
+ProblemFilters.exclude[MissingClassProblem]("lerna.akka.entityreplication.raft.snapshot.sync.SnapshotSyncManager$CompactionEnvelope$")
+ProblemFilters.exclude[MissingClassProblem]("lerna.akka.entityreplication.raft.snapshot.sync.SnapshotSyncManager$SyncCompleteAll")
+ProblemFilters.exclude[MissingClassProblem]("lerna.akka.entityreplication.raft.snapshot.sync.SnapshotSyncManager$SyncCompleteAll$")

src/main/protobuf/cluster_replication.proto

@@ -206,6 +206,14 @@ message TimeBasedUUID {
   required int64 least_sig_bits = 2;
 }
 
+message SnapshotCopied {
+  required Offset offset = 1;
+  required MemberIndex member_index = 2;
+  required NormalizedShardId shard_id = 3;
+  required Term snapshot_last_log_term = 4;
+  required LogEntryIndex snapshot_last_log_index = 5;
+  repeated NormalizedEntityId entity_ids = 6;
+}
 
 // ===
 // model

src/main/resources/reference.conf

@@ -57,6 +57,13 @@ lerna.akka.entityreplication {
 
       // Time to abort operations related to persistence
       persistence-operation-timeout = 10s
+
+      // Maximum size of a snapshot batch copied from leader's snapshot store to local snapshot store
+      // Note:
+      // If the event that updated the snapshots contains more than this batch size of entityId,
+      // only the snapshots the single event indicates will be copied over this limit.
+      // Copying snapshot should be executed atomically per event.
+      max-snapshot-batch-size = 1000
     }
 
     sharding = ${akka.cluster.sharding} {
@@ -92,6 +99,7 @@ lerna.akka.entityreplication {
         }
         event-adapter-bindings {
           "lerna.akka.entityreplication.raft.RaftActor$CompactionCompleted" = akka-entity-replication-raft-event-adapter
+          "lerna.akka.entityreplication.raft.snapshot.sync.SnapshotSyncManager$SnapshotCopied" = akka-entity-replication-raft-event-adapter
         }
       }
 

src/main/scala/lerna/akka/entityreplication/protobuf/ClusterReplicationSerializer.scala

@@ -48,11 +48,12 @@ private[entityreplication] final class ClusterReplicationSerializer(val system:
   // raft.snapshot
   val EntitySnapshotManifest = "DA"
   // raft.snapshot.sync
-  val SyncCompletedManifest = "EA"
-  val SyncProgressManifest  = "EB"
-  val NoOffsetManifest      = "EC"
-  val SequenceManifest      = "ED"
-  val TimeBasedUUIDManifest = "EE"
+  val SyncCompletedManifest  = "EA"
+  val SyncProgressManifest   = "EB"
+  val NoOffsetManifest       = "EC"
+  val SequenceManifest       = "ED"
+  val TimeBasedUUIDManifest  = "EE"
+  val SnapshotCopiedManifest = "EF"
   // raft.model
   val NoOpManifest = "FA"
   // typed
@@ -90,11 +91,12 @@ private[entityreplication] final class ClusterReplicationSerializer(val system:
     // raft.snapshot
     EntitySnapshotManifest -> entitySnapshotFromBinary,
     // raft.snapshot.sync
-    SyncCompletedManifest -> syncCompletedFromBinary,
-    SyncProgressManifest  -> syncProgressFromBinary,
-    NoOffsetManifest      -> noOffsetEnvelopeFromBinary,
-    SequenceManifest      -> sequenceEnvelopeFromBinary,
-    TimeBasedUUIDManifest -> timeBasedUUIDEnvelopeFromBinary,
+    SyncCompletedManifest  -> syncCompletedFromBinary,
+    SyncProgressManifest   -> syncProgressFromBinary,
+    NoOffsetManifest       -> noOffsetEnvelopeFromBinary,
+    SequenceManifest       -> sequenceEnvelopeFromBinary,
+    TimeBasedUUIDManifest  -> timeBasedUUIDEnvelopeFromBinary,
+    SnapshotCopiedManifest -> snapshotCopiedFromBinary,
     // raft.model
     NoOpManifest -> noOpFromBinary,
     // typed
@@ -160,11 +162,12 @@ private[entityreplication] final class ClusterReplicationSerializer(val system:
     // raft.snapsnot
     case _: raft.snapshot.SnapshotProtocol.EntitySnapshot => EntitySnapshotManifest
     // raft.snapshot.sync
-    case _: raft.snapshot.sync.SnapshotSyncManager.SyncCompleted => SyncCompletedManifest
-    case _: raft.snapshot.sync.SnapshotSyncManager.SyncProgress  => SyncProgressManifest
-    case _: NoOffsetEnvelope.type                                => NoOffsetManifest
-    case _: SequenceEnvelope                                     => SequenceManifest
-    case _: TimeBasedUUIDEnvelope                                => TimeBasedUUIDManifest
+    case _: raft.snapshot.sync.SnapshotSyncManager.SyncCompleted  => SyncCompletedManifest
+    case _: raft.snapshot.sync.SnapshotSyncManager.SyncProgress   => SyncProgressManifest
+    case _: NoOffsetEnvelope.type                                 => NoOffsetManifest
+    case _: SequenceEnvelope                                      => SequenceManifest
+    case _: TimeBasedUUIDEnvelope                                 => TimeBasedUUIDManifest
+    case _: raft.snapshot.sync.SnapshotSyncManager.SnapshotCopied => SnapshotCopiedManifest
     // raft.model
     case _: raft.model.NoOp.type => NoOpManifest
     // typed
@@ -202,11 +205,12 @@ private[entityreplication] final class ClusterReplicationSerializer(val system:
     // raft.snapshot
     case m: raft.snapshot.SnapshotProtocol.EntitySnapshot => entitySnapShotToBinary(m)
     // raft.snapshot.sync
-    case m: raft.snapshot.sync.SnapshotSyncManager.SyncCompleted => syncCompletedToBinary(m)
-    case m: raft.snapshot.sync.SnapshotSyncManager.SyncProgress  => syncProgressToBinary(m)
-    case m: NoOffsetEnvelope.type                                => noOffsetEnvelopeToBinary(m)
-    case m: SequenceEnvelope                                     => sequenceEnvelopeToBinary(m)
-    case m: TimeBasedUUIDEnvelope                                => timeBasedUUIDEnvelopeToBinary(m)
+    case m: raft.snapshot.sync.SnapshotSyncManager.SyncCompleted  => syncCompletedToBinary(m)
+    case m: raft.snapshot.sync.SnapshotSyncManager.SyncProgress   => syncProgressToBinary(m)
+    case m: NoOffsetEnvelope.type                                 => noOffsetEnvelopeToBinary(m)
+    case m: SequenceEnvelope                                      => sequenceEnvelopeToBinary(m)
+    case m: TimeBasedUUIDEnvelope                                 => timeBasedUUIDEnvelopeToBinary(m)
+    case m: raft.snapshot.sync.SnapshotSyncManager.SnapshotCopied => snapshotCopiedToBinary(m)
     // raft.model
     case m: raft.model.NoOp.type => noOpToBinary(m)
     // typed
@@ -811,6 +815,30 @@ private[entityreplication] final class ClusterReplicationSerializer(val system:
     )
   }
 
+  private def snapshotCopiedToBinary(message: raft.snapshot.sync.SnapshotSyncManager.SnapshotCopied): Array[Byte] = {
+    msg.SnapshotCopied
+      .of(
+        offset = offsetToProto(message.offset),
+        memberIndex = memberIndexToProto(message.memberIndex),
+        shardId = normalizedShardIdToProto(message.shardId),
+        snapshotLastLogTerm = termToProto(message.snapshotLastLogTerm),
+        snapshotLastLogIndex = logEntryIndexToProto(message.snapshotLastLogIndex),
+        entityIds = message.entityIds.map(normalizedEntityIdToProto).toSeq,
+      ).toByteArray
+  }
+
+  private def snapshotCopiedFromBinary(bytes: Array[Byte]): raft.snapshot.sync.SnapshotSyncManager.SnapshotCopied = {
+    val proto = msg.SnapshotCopied.parseFrom(bytes)
+    raft.snapshot.sync.SnapshotSyncManager.SnapshotCopied(
+      offset = offsetFromProto(proto.offset),
+      memberIndex = memberIndexFromProto(proto.memberIndex),
+      shardId = normalizedShardIdFromProto(proto.shardId),
+      snapshotLastLogTerm = termFromProto(proto.snapshotLastLogTerm),
+      snapshotLastLogIndex = logEntryIndexFromProto(proto.snapshotLastLogIndex),
+      entityIds = proto.entityIds.map(normalizedEntityIdFromProto).toSet,
+    )
+  }
+
   // ===
   // model
   // ===

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

@@ -161,6 +161,12 @@ private[raft] class RaftActor(
   override val persistenceId: String =
     ActorIds.persistenceId("raft", typeName.underlying, shardId.underlying, selfMemberIndex.role)
 
+  /**
+    * NOTE:
+    * [[RaftActor]] has to use the same journal plugin as [[SnapshotSyncManager]]
+    * because snapshot synchronization is achieved by reading both the events
+    * [[CompactionCompleted]] which [[RaftActor]] persisted and SnapshotCopied which [[SnapshotSyncManager]] persisted.
+    */
   override def journalPluginId: String = settings.journalPluginId
 
   override def journalPluginConfig: Config = settings.journalPluginAdditionalConfig

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

@@ -46,6 +46,8 @@ trait RaftSettings {
 
   def snapshotSyncPersistenceOperationTimeout: FiniteDuration
 
+  def snapshotSyncMaxSnapshotBatchSize: Int
+
   def clusterShardingConfig: Config
 
   def raftActorAutoStartFrequency: FiniteDuration

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

@@ -25,6 +25,7 @@ private[entityreplication] final case class RaftSettingsImpl(
     compactionLogSizeCheckInterval: FiniteDuration,
     snapshotSyncCopyingParallelism: Int,
     snapshotSyncPersistenceOperationTimeout: FiniteDuration,
+    snapshotSyncMaxSnapshotBatchSize: Int,
     clusterShardingConfig: Config,
     raftActorAutoStartFrequency: FiniteDuration,
     raftActorAutoStartNumberOfActors: Int,
@@ -140,6 +141,13 @@ private[entityreplication] object RaftSettingsImpl {
     val snapshotSyncPersistenceOperationTimeout: FiniteDuration =
       config.getDuration("snapshot-sync.persistence-operation-timeout").toScala
 
+    val snapshotSyncMaxSnapshotBatchSize: Int =
+      config.getInt("snapshot-sync.max-snapshot-batch-size")
+    require(
+      snapshotSyncMaxSnapshotBatchSize > 0,
+      s"snapshot-sync.max-snapshot-batch-size (${snapshotSyncMaxSnapshotBatchSize}) should be larger than 0",
+    )
+
     val clusterShardingConfig: Config = config.getConfig("sharding")
 
     val raftActorAutoStartFrequency: FiniteDuration =
@@ -196,6 +204,7 @@ private[entityreplication] object RaftSettingsImpl {
       compactionLogSizeCheckInterval = compactionLogSizeCheckInterval,
       snapshotSyncCopyingParallelism = snapshotSyncCopyingParallelism,
       snapshotSyncPersistenceOperationTimeout = snapshotSyncPersistenceOperationTimeout,
+      snapshotSyncMaxSnapshotBatchSize = snapshotSyncMaxSnapshotBatchSize,
       clusterShardingConfig = clusterShardingConfig,
       raftActorAutoStartFrequency = raftActorAutoStartFrequency,
       raftActorAutoStartNumberOfActors = raftActorAutoStartNumberOfActors,

…persistence/CompactionCompletedTag.scala …sistence/EntitySnapshotsUpdatedTag.scalasrc/main/scala/lerna/akka/entityreplication/raft/persistence/CompactionCompletedTag.scala renamed to src/main/scala/lerna/akka/entityreplication/raft/persistence/EntitySnapshotsUpdatedTag.scala src/main/scala/lerna/akka/entityreplication/raft/persistence/CompactionCompletedTag.scala renamed to src/main/scala/lerna/akka/entityreplication/raft/persistence/EntitySnapshotsUpdatedTag.scala

@@ -3,11 +3,12 @@ package lerna.akka.entityreplication.raft.persistence
 import lerna.akka.entityreplication.model.NormalizedShardId
 import lerna.akka.entityreplication.raft.routing.MemberIndex
 
-private[entityreplication] final case class CompactionCompletedTag(
+private[entityreplication] final case class EntitySnapshotsUpdatedTag(
     memberIndex: MemberIndex,
     shardId: NormalizedShardId,
 ) {
   private[this] val delimiter = ":"
 
+  // Do not change this tag format for compatibility
   override def toString: String = s"CompactionCompleted${delimiter}${shardId.underlying}${delimiter}${memberIndex.role}"
 }