Apache
/
activemq-artemis
mirror of https://github.com/apache/activemq-artemis.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

ARTEMIS-2713 Master failback can trigger a useless quorum vote on slave failover

This commit is contained in:
Francesco Nigro 2020-04-14 09:04:40 +02:00 committed by Clebert Suconic
parent e4c5ea719f
commit 5a829ff1f4
2 changed files with 13 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
artemis-server/src/main/java/org/apache/activemq/artemis/core/server/cluster/qourum/SharedNothingBackupQuorum.java
View File

@ -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

2
artemis-server/src/main/java/org/apache/activemq/artemis/core/server/impl/SharedNothingBackupActivation.java
View File

@ -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));
}