For unicast zen discovery don't overwrite a ping response for a node if the previous ping response has a set master and the current response hasn't.

Per single main ping request we maintain the received ping response per node. Each node level ping response is mapped into that. If from a previous node level ping request the response has already been set for a node, it will be overwritten. We give higher value to the latest response. This change makes sure that this doesn't happen if the previous response has a set master and the current response hasn't a set master. Otherwise a node will lose the fact that another node has elected itself as master, the result of that would be that there would multiple master nodes in a single cluster.

Closes #5413
This commit is contained in:
Martijn van Groningen 2014-03-13 13:51:05 +07:00
parent ca6a2bb790
commit 669a7ec498
2 changed files with 123 additions and 1 deletions

View File

@ -342,7 +342,17 @@ public class UnicastZenPing extends AbstractLifecycleComponent<ZenPing> implemen
if (responses == null) {
logger.warn("received ping response {} with no matching id [{}]", pingResponse, response.id);
} else {
PingResponse existingResponse = responses.get(pingResponse.target());
if (existingResponse == null) {
responses.put(pingResponse.target(), pingResponse);
} else {
// try and merge the best ping response for it, i.e. if the new one
// doesn't have the master node set, and the existing one does, then
// the existing one is better, so we keep it
if (pingResponse.master() != null) {
responses.put(pingResponse.target(), pingResponse);
}
}
}
}
} finally {

View File

@ -0,0 +1,112 @@
/*
* Licensed to Elasticsearch under one or more contributor
* license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright
* ownership. Elasticsearch licenses this file to you under
* the Apache License, Version 2.0 (the "License"); you may
* not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing,
* software distributed under the License is distributed on an
* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
* KIND, either express or implied. See the License for the
* specific language governing permissions and limitations
* under the License.
*/
package org.elasticsearch.cluster;
import org.apache.lucene.util.LuceneTestCase;
import org.elasticsearch.action.admin.cluster.health.ClusterHealthResponse;
import org.elasticsearch.cluster.node.DiscoveryNode;
import org.elasticsearch.common.Priority;
import org.elasticsearch.common.settings.Settings;
import org.elasticsearch.common.util.concurrent.AtomicArray;
import org.elasticsearch.test.ElasticsearchIntegrationTest;
import org.elasticsearch.test.junit.annotations.TestLogging;
import org.junit.Test;
import java.util.concurrent.CountDownLatch;
import static org.elasticsearch.common.settings.ImmutableSettings.settingsBuilder;
import static org.hamcrest.Matchers.equalTo;
/**
*/
@LuceneTestCase.Slow
@ElasticsearchIntegrationTest.ClusterScope(scope = ElasticsearchIntegrationTest.Scope.TEST, numNodes = 0)
public class ZenUnicastDiscoveryTests extends ElasticsearchIntegrationTest {
@Test
@TestLogging("discovery.zen:TRACE")
// The bug zen unicast ping override bug, may rarely manifest itself, it is very timing dependant.
// Without the fix in UnicastZenPing, this test fails roughly 1 out of 10 runs from the command line.
public void testMasterElectionNotMissed() throws Exception {
final Settings settings = settingsBuilder()
// Failure only manifests if multicast ping is disabled!
.put("discovery.zen.ping.multicast.ping.enabled", false)
.put("discovery.zen.minimum_master_nodes", 2)
// Can't use this, b/c at the moment all node will only ping localhost:9300
// .put("discovery.zen.ping.unicast.hosts", "localhost")
.put("discovery.zen.ping.unicast.hosts", "localhost:15300,localhost:15301,localhost:15302")
.put("transport.tcp.port", "15300-15400")
.build();
final CountDownLatch latch = new CountDownLatch(3);
final AtomicArray<String> nodes = new AtomicArray<String>(3);
Runnable r1 = new Runnable() {
@Override
public void run() {
logger.info("--> start first node");
nodes.set(0, cluster().startNode(settings));
latch.countDown();
}
};
new Thread(r1).start();
sleep(between(500, 3000));
Runnable r2 = new Runnable() {
@Override
public void run() {
logger.info("--> start second node");
nodes.set(1, cluster().startNode(settings));
latch.countDown();
}
};
new Thread(r2).start();
sleep(between(500, 3000));
Runnable r3 = new Runnable() {
@Override
public void run() {
logger.info("--> start third node");
nodes.set(2, cluster().startNode(settings));
latch.countDown();
}
};
new Thread(r3).start();
latch.await();
ClusterHealthResponse clusterHealthResponse = client().admin().cluster().prepareHealth().setWaitForEvents(Priority.LANGUID).setWaitForNodes("3").execute().actionGet();
assertThat(clusterHealthResponse.isTimedOut(), equalTo(false));
DiscoveryNode masterDiscoNode = null;
for (String node : nodes.toArray(new String[3])) {
ClusterState state = cluster().client(node).admin().cluster().prepareState().setLocal(true).execute().actionGet().getState();
assertThat(state.nodes().size(), equalTo(3));
if (masterDiscoNode == null) {
masterDiscoNode = state.nodes().masterNode();
} else {
assertThat(masterDiscoNode.equals(state.nodes().masterNode()), equalTo(true));
}
}
}
}