Merge pull request #1785 from akka/wip-3458-adjust-biased-gossip-patriknw

+clu #3458 Adjust biased gossip for large cluster

akka-cluster/src/main/scala/akka/cluster/ClusterDaemon.scala

@@ -698,7 +698,8 @@ private[cluster] class ClusterCoreDaemon(publisher: ActorRef) extends Actor with
       val localGossip = latestGossip
 
       val preferredGossipTargets: Vector[UniqueAddress] =
-        if (ThreadLocalRandom.current.nextDouble() < GossipDifferentViewProbability) { // If it's time to try to gossip to some nodes with a different view
+        if (ThreadLocalRandom.current.nextDouble() < adjustedGossipDifferentViewProbability) {
+          // If it's time to try to gossip to some nodes with a different view
           // gossip to a random alive member with preference to a member with older gossip version
           localGossip.members.collect {
             case m if !localGossip.seenByNode(m.uniqueAddress) && validNodeForGossip(m.uniqueAddress) ⇒
@@ -723,6 +724,33 @@ private[cluster] class ClusterCoreDaemon(publisher: ActorRef) extends Actor with
     }
   }
 
+  /**
+   * For large clusters we should avoid shooting down individual
+   * nodes. Therefore the probability is reduced for large clusters.
+   */
+  def adjustedGossipDifferentViewProbability: Double = {
+    val size = latestGossip.members.size
+    val low = ReduceGossipDifferentViewProbability
+    val high = low * 3
+    // start reduction when cluster is larger than configured ReduceGossipDifferentViewProbability
+    if (size <= low)
+      GossipDifferentViewProbability
+    else {
+      // don't go lower than 1/10 of the configured GossipDifferentViewProbability
+      val minP = GossipDifferentViewProbability / 10
+      if (size >= high)
+        minP
+      else {
+        // linear reduction of the probability with increasing number of nodes
+        // from ReduceGossipDifferentViewProbability at ReduceGossipDifferentViewProbability nodes
+        // to ReduceGossipDifferentViewProbability / 10 at ReduceGossipDifferentViewProbability * 3 nodes
+        // i.e. default from 0.8 at 400 nodes, to 0.08 at 1600 nodes
+        val k = (minP - GossipDifferentViewProbability) / (high - low)
+        GossipDifferentViewProbability + (size - low) * k
+      }
+    }
+  }
+
   /**
    * Runs periodic leader actions, such as member status transitions, assigning partitions etc.
    */