Raft leader uses the same previous LogEntryIndex and Term to all batched AppendEntries message

[xirc]

Leader uses the same previous values (prevLogIndex and prevLogTerm) to all batched AppendEntries messages.

akka-entity-replication/src/main/scala/lerna/akka/entityreplication/raft/Leader.scala

Lines 217 to 256 in 8692aa9

	private[this] def publishAppendEntries(): Unit = {
	resetHeartbeatTimeoutTimer()
	otherMemberIndexes.foreach { memberIndex =>
	val nextIndex = currentData.nextIndexFor(memberIndex)
	val prevLogIndex = nextIndex.prev()
	val prevLogTerm = currentData.replicatedLog.termAt(prevLogIndex)
	val messages =
	prevLogTerm match {
	case Some(prevLogTerm) =>
	val batchEntries = currentData.replicatedLog.getFrom(
	nextIndex,
	settings.maxAppendEntriesSize,
	settings.maxAppendEntriesBatchSize,
	)
	batchEntries match {
	case batchEntries if batchEntries.isEmpty =>
	Seq(
	AppendEntries(
	shardId,
	currentData.currentTerm,
	selfMemberIndex,
	prevLogIndex,
	prevLogTerm,
	entries = Seq.empty,
	currentData.commitIndex,
	),
	)
	case batchEntries =>
	batchEntries.map { entries =>
	AppendEntries(
	shardId,
	currentData.currentTerm,
	selfMemberIndex,
	prevLogIndex,
	prevLogTerm,
	entries,
	currentData.commitIndex,
	)
	}
	}

Expected

Raft leader should set appropriate previous values (prevLogIndex and prevLogTerm) to batched AppendEntries. The expected AppendEntries is the following:

[xirc]

Possible workaround:

• Use lerna.akka.entityreplication.raft.max-append-entries-batch-size = 1