RaftMemberData.resolveNewLogEntries got requirement failed: A new entry conflicted with a committed entry

[xirc]

The following error occurred in some fault injection tests:

04:48:20.417 xxx        ERROR   akka.actor.OneForOneStrategy    xxx akka://xxx/system/sharding/raft-shard-xxx-replica-group-3/30/30       -       requirement failed: The entry with index [583] should not conflict with the committed entry (commitIndex [584]) java.lang.IllegalArgumentException: requirement failed: The entry with index [583] should not conflict with the committed entry (commitIndex [584])
    at scala.Predef$.require(Predef.scala:337)
    at lerna.akka.entityreplication.raft.FollowerData.resolveNewLogEntries(RaftMemberData.scala:147)
    at lerna.akka.entityreplication.raft.FollowerData.resolveNewLogEntries$(RaftMemberData.scala:119)
    at lerna.akka.entityreplication.raft.RaftMemberDataImpl.resolveNewLogEntries(RaftMemberData.scala:616)
    at lerna.akka.entityreplication.raft.Follower.lerna$akka$entityreplication$raft$Follower$$receiveAppendEntries(Follower.scala:111)
    at lerna.akka.entityreplication.raft.Follower$$anonfun$followerBehavior$1.applyOrElse(Follower.scala:25)
    at akka.actor.Actor.aroundReceive(Actor.scala:537)
    at akka.actor.Actor.aroundReceive$(Actor.scala:535)
    at lerna.akka.entityreplication.raft.RaftActor.akka$persistence$Eventsourced$$super$aroundReceive(RaftActor.scala:141)
    at akka.persistence.Eventsourced$$anon$4.stateReceive(Eventsourced.scala:923)
    at akka.persistence.Eventsourced.aroundReceive$$original(Eventsourced.scala:251)
    at akka.persistence.Eventsourced.aroundReceive(Eventsourced.scala:148)
    at akka.persistence.Eventsourced.aroundReceive$(Eventsourced.scala:250)
    at lerna.akka.entityreplication.raft.RaftActor.aroundReceive(RaftActor.scala:141)
    at akka.actor.ActorCell.receiveMessage(ActorCell.scala:580)
    at akka.actor.ActorCell.invoke$$original(ActorCell.scala:548)
    at akka.actor.ActorCell.invoke(ActorCell.scala:61)
    at akka.dispatch.Mailbox.processMailbox(Mailbox.scala:270)
    at akka.dispatch.Mailbox.run$$original(Mailbox.scala:231)
    at akka.dispatch.Mailbox.run(Mailbox.scala:32)
    at akka.dispatch.Mailbox.exec(Mailbox.scala:243)
    at java.base/java.util.concurrent.ForkJoinTask.doExec(ForkJoinTask.java:290)
    at java.base/java.util.concurrent.ForkJoinPool$WorkQueue.topLevelExec(ForkJoinPool.java:1020)
    at java.base/java.util.concurrent.ForkJoinPool.scan(ForkJoinPool.java:1656)
    at java.base/java.util.concurrent.ForkJoinPool.runWorker(ForkJoinPool.java:1594)
    at java.base/java.util.concurrent.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:183)

[xirc]

From diagnosing logs of another test that also got the requirement failed error, the following situation could be a cause of this error:

RaftActor of replica-group-1

• 06:15:50.049: It appended log entry (index 788) into ReplicatedLog
• 06:15:50.262: It updated the commit index to 788
• 06:16:13.451: It voted other RaftActor in term 7
• 06:16:13.681: It appended log entry (index 789) into ReplicatedLog
• 06:16:14.140: It appended log entry (index 790) into ReplicatedLog
• 06:16:14.140: It updated the commit index to 789
• ...

RaftActor of replica-group-2

• 06:06:04.726: It became the leader in term 2
• 06:15:50.050 ~ 06:15:50.051: It updated the commit index to 788
  • 06:15:50.050: It updated indices for replica-group-1 (next index = 788 -> 789, match index = 787 -> 788) by applying an event SucceededAppendEntries
  • 06:15:50.051: It updated indices for replica-group-3 (next index = 788 -> 789, match index = 787 -> 788) by applying an event SucceededAppendEntries
• 06:15:56.314: It started a replication of log entry (index 789)
• 06:16:01.388: It(leader) became a follower by detecting a new term (term 3)
• ... electing a new leader
• 06:16:18.102: It restarted due to a persistence failure
• ... electing a new leader
• 06:16:33.837: It voted the RaftActor of replica-group-3 on term 11
• 06:16:36.075: It updated the commit index (0 -> 789)
• 06:16:36.076: It got the requirement failed error

RaftActor of replica-group-3

• 06:15:50.049: It appended log entry (index 788) into ReplicatedLog
• 06:15:50.262: It updated the commit index to 788
• 06:16:13.448: It became the leader in term 7
  • The RaftActor of replica-group-1 voted this RaftActor (of replica-group-3)
• 06:16:13.630: It succeeded in replication of NoOp (index 789) and then updated the commit index to 789
  • It updated indices for replica-group-1 (next index = None -> 790, match index = None -> 789) by applying an event SucceededAppendEntries
• ... electing a new leader
• 06:16:21.930: It updated the commit index to 832
• 06:16:22.176: It became a follower by detecting a new term (term 8)
• 06:16:24.727: It became the leader in term 11
• 06:16:24.727: It updated indices for all replica groups (next index = None, match index = None)
• 06:16:24.737 ~ 06:16:24.824:
  • It updated indices for replica-group-1 (next index = None -> 833 -> 835, match index = None 834)
  • It updated the commit index (832 -> 834)
• 06:16:32.924 ~ 06:16:33.700:
  • 06:16:32.924: It updated the next index for replica-group-2 (None -> 833)
  • 06:16:32.924: It decremented the next index for replica-group-2 (833 -> 832)
  • ... decrementing the next index for replica-group-2
  • 06:16:33.700: It decremented the next index for replica-group-2 (736 -> 735)
• 06:16:34.037:
  • It updated the next index for replica-group-2 (735 -> 790)
  • It updated the match index for replica-group-2 (None -> 789)

[xirc]

There might be at least two possible solutions:

1. Improve a mechanism for decrementing the next index
2. Improve a mechanism for advancing the commit index

There might be a reason behind the next index being lower than expected:

• If a follower is slow temporary, the leader sends multiple AppendEntries messages at some heartbeat interval, and the follower replies with multiple AppendEntriesFaield messages, which decrease the next index than expected
• If the leader sends multiple AppendEntries (on batch mode), and the follower replies with multiple AppendEntriesFaield, which decreases the next index than expected