Duplicated Messages during failover and NullPersistence

[clebert.suconic]

I have made AutomaticFailoverWithDiscoveryTest to use real files on the journal.


If you make it use NullPersistence, and place testFailover on a loop, messages will be duplicated.


I moved the setupGroup to a super class (FailoverTestBase).

To replicate this issue, change the setUp to:

protected void setUp() throws Exception
 {
 super.setUp();
 setupGroupServers(true, "bc1", 5432, groupAddress, groupPort);
 }

[clebert.suconic]

Just to complete the thread, this failure I was talking about is related to an intermitent failure that has happened on hudson.

It doesn't fail if using real files (*probably* duplicate detection behaves different when using real files).

This is the diff to replicate the issue.

Index: tests/src/org/jboss/messaging/tests/integration/cluster/failover/AutomaticFailoverWithDiscoveryTest.java
===================================================================
--- tests/src/org/jboss/messaging/tests/integration/cluster/failover/AutomaticFailoverWithDiscoveryTest.java (revision 6120)
+++ tests/src/org/jboss/messaging/tests/integration/cluster/failover/AutomaticFailoverWithDiscoveryTest.java (working copy)
@@ -63,6 +63,19 @@
 // Constructors --------------------------------------------------

 // Public --------------------------------------------------------
+
+ public void testRepeat() throws Exception
+ {
+ for (int i = 0; i < 100; i++)
+ {
+ if (i > 0)
+ {
+ tearDown();
+ setUp();
+ }
+ testFailover();
+ }
+ }

 public void testFailover() throws Exception
 {
@@ -173,7 +186,7 @@
 protected void setUp() throws Exception
 {
 super.setUp();
- setupGroupServers(true, "bc1", 5432, groupAddress, groupPort);
+ setupGroupServers(false, "bc1", 5432, groupAddress, groupPort);
 }

 @Override

[clebert.suconic]

https://jira.jboss.org/jira/browse/JBMESSAGING-1552

[clebert.suconic]

Actually.. this doesn' t have anything to do with Persistence & NullPersistence


If I wait 17ms (as done on MultiThreadfailoverTest), the test never fails.

[clebert.suconic]

I mean...


If I wait 17 ms between the backup and live start, this issue never happens.

[clebert.suconic]

The issue was related to the time-components on the IDs for sure.

I added a test PreserveOrderDuringFailoverTest, which is based on AutomaticFailoverWithDiscoveryTest.

If you uncomment some code on PreserveOrderDuringFailoverTest, this issue will aways happen:

// This test would fail if both servers have the same time component
// NullStorageManager storageManagerLive = (NullStorageManager)liveService.getServer().getStorageManager();
// TimeAndCounterIDGenerator idgeneratorlive = (TimeAndCounterIDGenerator)storageManagerLive.getIDGenerator();
//
// NullStorageManager storageManagerBackup = (NullStorageManager)backupService.getServer().getStorageManager();
// TimeAndCounterIDGenerator idgeneratorBackup = (TimeAndCounterIDGenerator)storageManagerBackup.getIDGenerator();
//
// idgeneratorBackup.setInternalDate(0);
// idgeneratorlive.setInternalDate(0);

I would expect the IDs not affecting failover any more, so I would debug this but since this part is already changed at Tim's workspace, I will leave this alone.

Tim: If you could please remove the wait on FailoverTestBase during your commit. Since you have changed the ID logic, we won't need the wait any more.

 backupService.start();

- Thread.sleep(20);

 Configuration liveConf = new ConfigurationImpl();