honeymoose
/
OpenSearch
mirror of https://github.com/honeymoose/OpenSearch.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Test: clear unicast temporal response list before we stop the elected master node. 

Original commit: elastic/x-pack-elasticsearch@0b2675f9ff
This commit is contained in:
Martijn van Groningen 2015-05-18 13:06:42 +02:00
parent 1db5d1e37d
commit 6404b2c7e5
1 changed files with 17 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
src/test/java/org/elasticsearch/watcher/test/integration/NoMasterNodeTests.java
View File

@ -16,6 +16,9 @@ import org.elasticsearch.common.unit.TimeValue;
import org.elasticsearch.discovery.DiscoverySettings;
import org.elasticsearch.discovery.MasterNotDiscoveredException;
import org.elasticsearch.discovery.zen.elect.ElectMasterService;
import org.elasticsearch.discovery.zen.ping.ZenPing;
import org.elasticsearch.discovery.zen.ping.ZenPingService;
import org.elasticsearch.discovery.zen.ping.unicast.UnicastZenPing;
import org.elasticsearch.test.ElasticsearchIntegrationTest.ClusterScope;
import org.elasticsearch.test.discovery.ClusterDiscoveryConfiguration;
import org.elasticsearch.test.junit.annotations.TestLogging;
@ -51,7 +54,7 @@ import static org.hamcrest.core.Is.is;
/**
*/
@Slow
@TestLogging("watcher:DEBUG,discovery:TRACE")
@TestLogging("discovery:TRACE")
@ClusterScope(scope = TEST, numClientNodes = 0, transportClientRatio = 0, randomDynamicTemplates = false, numDataNodes = 0)
public class NoMasterNodeTests extends AbstractWatcherIntegrationTests {
@ -225,6 +228,19 @@ public class NoMasterNodeTests extends AbstractWatcherIntegrationTests {
}
private void stopElectedMasterNodeAndWait() throws Exception {
// Due to the fact that the elected master node gets stopped before the ping timeout has passed,
// the unicast ping may have cached ping responses. This can cause the existing node to think
// that there is another node and if its node id is lower than the node that has been stopped it
// will elect itself as master. This is bad and should be fixed in core. What I think that should happen is that
// if a node detects that is has lost a node, a node should clear its unicast temporal responses or at least
// remove the node that has been removed. This is a workaround:
for (ZenPingService pingService : internalCluster().getInstances(ZenPingService.class)) {
for (ZenPing zenPing : pingService.zenPings()) {
if (zenPing instanceof UnicastZenPing) {
((UnicastZenPing) zenPing).clearTemporalResponses();
}
}
}
internalTestCluster().stopCurrentMasterNode();
// Can't use ensureWatcherStopped, b/c that relies on the watcher stats api which requires an elected master node
assertBusy(new Runnable() {