n, // number of nodes
3, // replicas
"MasterSlave", true); // do rebalance
// start controller
ClusterControllerManager controller =
new ClusterControllerManager(ZK_ADDR, clusterName, "controller_0");
controller.syncStart();
Map<String, Set<String>> errPartitions = new HashMap<String, Set<String>>() {
{
put("SLAVE-MASTER", TestHelper.setOf("TestDB0_4"));
put("OFFLINE-SLAVE", TestHelper.setOf("TestDB0_8"));
}
};
// start mock participants
MockParticipantManager[] participants = new MockParticipantManager[n];
for (int i = 0; i < n; i++) {
String instanceName = "localhost_" + (12918 + i);
if (i == 0) {
participants[i] =
new MockParticipantManager(ZK_ADDR, clusterName, instanceName);
participants[i].setTransition(new ErrTransition(errPartitions));
} else {
participants[i] = new MockParticipantManager(ZK_ADDR, clusterName, instanceName);
}
participants[i].syncStart();
}
// verify cluster
Map<String, Map<String, String>> errStateMap = new HashMap<String, Map<String, String>>();
errStateMap.put("TestDB0", new HashMap<String, String>());
errStateMap.get("TestDB0").put("TestDB0_4", "localhost_12918");
errStateMap.get("TestDB0").put("TestDB0_8", "localhost_12918");
boolean result =
ClusterStateVerifier
.verifyByZkCallback((new ClusterStateVerifier.BestPossAndExtViewZkVerifier(ZK_ADDR,
clusterName, errStateMap)));
Assert.assertTrue(result, "Cluster verification fails");
// reset a non-exist partition, should throw exception
try {
String command =
"--zkSvr " + ZK_ADDR + " --resetPartition " + clusterName
+ " localhost_12918 TestDB0 TestDB0_nonExist";
ClusterSetup.processCommandLineArgs(command.split("\\s+"));
Assert.fail("Should throw exception on reset a non-exist partition");
} catch (Exception e) {
// OK
}
// reset one error partition
errPartitions.remove("SLAVE-MASTER");
participants[0].setTransition(new ErrTransitionWithResetCnt(errPartitions));
clearStatusUpdate(clusterName, "localhost_12918", "TestDB0", "TestDB0_4");
_errToOfflineInvoked = 0;
String command =
"--zkSvr " + ZK_ADDR + " --resetPartition " + clusterName
+ " localhost_12918 TestDB0 TestDB0_4";
ClusterSetup.processCommandLineArgs(command.split("\\s+"));
Thread.sleep(200); // wait reset to be done
try {
ClusterSetup.processCommandLineArgs(command.split("\\s+"));
Assert.fail("Should throw exception on reset a partition not in ERROR state");
} catch (Exception e) {
// OK
}
errStateMap.get("TestDB0").remove("TestDB0_4");
result =
ClusterStateVerifier
.verifyByZkCallback((new ClusterStateVerifier.BestPossAndExtViewZkVerifier(ZK_ADDR,
clusterName, errStateMap)));
Assert.assertTrue(result, "Cluster verification fails");
Assert.assertEquals(_errToOfflineInvoked, 1);
// reset the other error partition
participants[0].setTransition(new ErrTransitionWithResetCnt(null));
clearStatusUpdate(clusterName, "localhost_12918", "TestDB0", "TestDB0_8");
command =
"--zkSvr " + ZK_ADDR + " --resetPartition " + clusterName
+ " localhost_12918 TestDB0 TestDB0_8";
ClusterSetup.processCommandLineArgs(command.split("\\s+"));
result =
ClusterStateVerifier.verifyByPolling(new ClusterStateVerifier.BestPossAndExtViewZkVerifier(
ZK_ADDR, clusterName));
Assert.assertTrue(result, "Cluster verification fails");
Assert.assertEquals(_errToOfflineInvoked, 2, "Should reset 2 partitions");
// clean up
controller.syncStop();
for (int i = 0; i < 5; i++) {
participants[i].syncStop();
}
System.out.println("END " + clusterName + " at " + new Date(System.currentTimeMillis()));