[activemq-artemis] 02/03: ARTEMIS-2713 Master failback can trigger a useless quorum vote on slave failover

[cl...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

clebertsuconic pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/activemq-artemis.git

commit 5a829ff1f4b33540e6ddb6f1f0e99ed2a8204f4e
Author: Francesco Nigro <ni...@gmail.com>
AuthorDate: Tue Apr 14 09:04:40 2020 +0200

    ARTEMIS-2713 Master failback can trigger a useless quorum vote on slave failover
---
 .../server/cluster/qourum/SharedNothingBackupQuorum.java   | 14 ++++++++++++--
 .../core/server/impl/SharedNothingBackupActivation.java    |  2 +-
 2 files changed, 13 insertions(+), 3 deletions(-)

diff --git a/artemis-server/src/main/java/org/apache/activemq/artemis/core/server/cluster/qourum/SharedNothingBackupQuorum.java b/artemis-server/src/main/java/org/apache/activemq/artemis/core/server/cluster/qourum/SharedNothingBackupQuorum.java
index 9da89d0..77d38f6 100644
--- a/artemis-server/src/main/java/org/apache/activemq/artemis/core/server/cluster/qourum/SharedNothingBackupQuorum.java
+++ b/artemis-server/src/main/java/org/apache/activemq/artemis/core/server/cluster/qourum/SharedNothingBackupQuorum.java
@@ -72,6 +72,8 @@ public class SharedNothingBackupQuorum implements Quorum, SessionFailureListener
    private CountDownLatch latch;
 
    private final Object onConnectionFailureGuard = new Object();
+
+   private final boolean failback;
    /**
     * This is a safety net in case the live sends the first {@link ReplicationLiveIsStoppingMessage}
     * with code {@link org.apache.activemq.artemis.core.protocol.core.impl.wireformat.ReplicationLiveIsStoppingMessage.LiveStopping#STOP_CALLED} and crashes before sending the second with
@@ -87,7 +89,8 @@ public class SharedNothingBackupQuorum implements Quorum, SessionFailureListener
                                     int quorumSize,
                                     int voteRetries,
                                     long voteRetryWait,
-                                    int quorumVoteWait) {
+                                    int quorumVoteWait,
+                                    boolean failback) {
       this.scheduledPool = scheduledPool;
       this.quorumSize = quorumSize;
       this.latch = new CountDownLatch(1);
@@ -96,6 +99,7 @@ public class SharedNothingBackupQuorum implements Quorum, SessionFailureListener
       this.voteRetries = voteRetries;
       this.voteRetryWait = voteRetryWait;
       this.quorumVoteWait = quorumVoteWait;
+      this.failback = failback;
    }
 
    @Override
@@ -161,7 +165,13 @@ public class SharedNothingBackupQuorum implements Quorum, SessionFailureListener
     */
    @Override
    public void nodeDown(Topology topology, long eventUID, String nodeID) {
-      //noop: we are NOT interested on topology info coming from connections != this.connection
+      // ignore it during a failback:
+      // a failing slave close all connections but the one used for replication
+      // triggering a nodeDown before the restarted master receive a STOP_CALLED from it.
+      // This can make master to fire a useless quorum vote during a normal failback.
+      if (!failback && targetServerID.equals(nodeID)) {
+         onConnectionFailure();
+      }
    }
 
    @Override
diff --git a/artemis-server/src/main/java/org/apache/activemq/artemis/core/server/impl/SharedNothingBackupActivation.java b/artemis-server/src/main/java/org/apache/activemq/artemis/core/server/impl/SharedNothingBackupActivation.java
index bbfc2b3..3c199d2 100644
--- a/artemis-server/src/main/java/org/apache/activemq/artemis/core/server/impl/SharedNothingBackupActivation.java
+++ b/artemis-server/src/main/java/org/apache/activemq/artemis/core/server/impl/SharedNothingBackupActivation.java
@@ -131,7 +131,7 @@ public final class SharedNothingBackupActivation extends Activation {
             logger.trace("Entered a synchronized");
             if (closed)
                return;
-            backupQuorum = new SharedNothingBackupQuorum(activeMQServer.getNodeManager(), activeMQServer.getScheduledPool(), networkHealthCheck, replicaPolicy.getQuorumSize(), replicaPolicy.getVoteRetries(), replicaPolicy.getVoteRetryWait(), replicaPolicy.getQuorumVoteWait());
+            backupQuorum = new SharedNothingBackupQuorum(activeMQServer.getNodeManager(), activeMQServer.getScheduledPool(), networkHealthCheck, replicaPolicy.getQuorumSize(), replicaPolicy.getVoteRetries(), replicaPolicy.getVoteRetryWait(), replicaPolicy.getQuorumVoteWait(), attemptFailBack);
             activeMQServer.getClusterManager().getQuorumManager().registerQuorum(backupQuorum);
             activeMQServer.getClusterManager().getQuorumManager().registerQuorumHandler(new ServerConnectVoteHandler(activeMQServer));
          }