raboof/pekko
1
0
Fork 0
Code Issues Pull requests Projects Releases 6 Packages Wiki Activity Actions 11

Nodes not part of cluster have marked the Gossip as seen, see #3031

Browse source 
* Problem may occur when joining member with same hostname:port again,
  after downing.
* Reproduced with StressSpec exerciseJoinRemove with fixed port that
  joins and shutdown several times.
* Real solution for this will be covered by ticket #2788 by adding
  uid to member identifier, but as first step we need to support
  this scenario with current design.
* Use unique node identifier for vector clock to avoid mixup of
  old and new member instance.
* Support transition from Down to Joining in Gossip merge
* Don't gossip to unknown or unreachable members.
This commit is contained in:
Patrik Nordwall 2013-02-12 21:45:41 +01:00
parent cab78e5174
commit b349ad8d87
4 changed files with 28 additions and 8 deletions
Download patch file Download diff file Expand all files Collapse all files

15
akka-cluster/src/main/scala/akka/cluster/ClusterDaemon.scala
View file

@ -9,6 +9,7 @@ import scala.collection.immutable
import scala.concurrent.duration._
import scala.concurrent.forkjoin.ThreadLocalRandom
import scala.util.control.NonFatal
import java.util.UUID
import akka.actor.{ Actor, ActorLogging, ActorRef, Address, Cancellable, Props, PoisonPill, ReceiveTimeout, RootActorPath, Scheduler }
import akka.actor.OneForOneStrategy
import akka.actor.Status.Failure
@ -219,7 +220,8 @@ private[cluster] final class ClusterCoreDaemon(publisher: ActorRef) extends Acto
import cluster.{ selfAddress, scheduler, failureDetector }
import cluster.settings._
val vclockNode = VectorClock.Node(selfAddress.toString)
// FIXME the UUID should not be needed when Address contains uid, ticket #2788
val vclockNode = VectorClock.Node(selfAddress.toString + "-" + UUID.randomUUID())
// note that self is not initially member,
// and the Gossip is not versioned for this 'Node' yet
@ -507,10 +509,10 @@ private[cluster] final class ClusterCoreDaemon(publisher: ActorRef) extends Acto
val localGossip = latestGossip
if (remoteGossip.overview.unreachable.exists(_.address == selfAddress)) {
// FIXME how should we handle this situation?
log.debug("Received gossip with self as unreachable, from [{}]", from)
} else if (!localGossip.overview.isNonDownUnreachable(from)) {
log.debug("Ignoring received gossip with self [{}] as unreachable, from [{}]", selfAddress, from)
} else if (localGossip.overview.isNonDownUnreachable(from)) {
log.debug("Ignoring received gossip from unreachable [{}] ", from)
} else {
// leader handles merge conflicts, or when they have different views of how is leader
val handleMerge = localGossip.leader == Some(selfAddress) || localGossip.leader != remoteGossip.leader
@ -830,7 +832,8 @@ private[cluster] final class ClusterCoreDaemon(publisher: ActorRef) extends Acto
gossipTo(address, GossipEnvelope(selfAddress, latestGossip, conversation = false))
def gossipTo(address: Address, gossipMsg: GossipEnvelope): Unit =
if (address != selfAddress) clusterCore(address) ! gossipMsg
if (address != selfAddress && gossipMsg.gossip.members.exists(_.address == address))
clusterCore(address) ! gossipMsg
def publish(newGossip: Gossip): Unit = {
publisher ! PublishChanges(newGossip)