/// <exception cref="System.Exception"/>
protected override void ServiceInit(Configuration conf)
{
conf = conf is YarnConfiguration ? conf : new YarnConfiguration(conf);
string zkQuorum = conf.Get(YarnConfiguration.RmZkAddress);
if (zkQuorum == null)
{
throw new YarnRuntimeException("Embedded automatic failover " + "is enabled, but "
+ YarnConfiguration.RmZkAddress + " is not set");
}
string rmId = HAUtil.GetRMHAId(conf);
string clusterId = YarnConfiguration.GetClusterId(conf);
localActiveNodeInfo = CreateActiveNodeInfo(clusterId, rmId);
string zkBasePath = conf.Get(YarnConfiguration.AutoFailoverZkBasePath, YarnConfiguration
.DefaultAutoFailoverZkBasePath);
string electionZNode = zkBasePath + "/" + clusterId;
long zkSessionTimeout = conf.GetLong(YarnConfiguration.RmZkTimeoutMs, YarnConfiguration
.DefaultRmZkTimeoutMs);
IList <ACL> zkAcls = RMZKUtils.GetZKAcls(conf);
IList <ZKUtil.ZKAuthInfo> zkAuths = RMZKUtils.GetZKAuths(conf);
int maxRetryNum = conf.GetInt(CommonConfigurationKeys.HaFcElectorZkOpRetriesKey,
CommonConfigurationKeys.HaFcElectorZkOpRetriesDefault);
elector = new ActiveStandbyElector(zkQuorum, (int)zkSessionTimeout, electionZNode
, zkAcls, zkAuths, this, maxRetryNum);
elector.EnsureParentZNode();
if (!IsParentZnodeSafe(clusterId))
{
NotifyFatalError(electionZNode + " znode has invalid data! " + "Might need formatting!"
);
}
base.ServiceInit(conf);
}
/// <exception cref="System.Exception"/>
public virtual void TestHandleSessionExpiration()
{
ActiveStandbyElector.ActiveStandbyElectorCallback cb = cbs[0];
byte[] appData = appDatas[0];
ActiveStandbyElector elector = electors[0];
// Let the first elector become active
elector.EnsureParentZNode();
elector.JoinElection(appData);
ZooKeeperServer zks = GetServer(serverFactory);
ActiveStandbyElectorTestUtil.WaitForActiveLockData(null, zks, ParentDir, appData);
Org.Mockito.Mockito.Verify(cb, Org.Mockito.Mockito.Timeout(1000)).BecomeActive();
CheckFatalsAndReset();
Log.Info("========================== Expiring session");
zks.CloseSession(elector.GetZKSessionIdForTests());
// Should enter neutral mode when disconnected
Org.Mockito.Mockito.Verify(cb, Org.Mockito.Mockito.Timeout(1000)).EnterNeutralMode
();
// Should re-join the election and regain active
ActiveStandbyElectorTestUtil.WaitForActiveLockData(null, zks, ParentDir, appData);
Org.Mockito.Mockito.Verify(cb, Org.Mockito.Mockito.Timeout(1000)).BecomeActive();
CheckFatalsAndReset();
Log.Info("========================== Quitting election");
elector.QuitElection(false);
ActiveStandbyElectorTestUtil.WaitForActiveLockData(null, zks, ParentDir, null);
// Double check that we don't accidentally re-join the election
// due to receiving the "expired" event.
Thread.Sleep(1000);
Org.Mockito.Mockito.Verify(cb, Org.Mockito.Mockito.Never()).BecomeActive();
ActiveStandbyElectorTestUtil.WaitForActiveLockData(null, zks, ParentDir, null);
CheckFatalsAndReset();
}
/// <exception cref="System.Exception"/>
public override void SetUp()
{
base.SetUp();
zkServer = GetServer(serverFactory);
for (int i = 0; i < NumElectors; i++)
{
cbs[i] = Org.Mockito.Mockito.Mock <ActiveStandbyElector.ActiveStandbyElectorCallback
>();
appDatas[i] = Ints.ToByteArray(i);
electors[i] = new ActiveStandbyElector(hostPort, 5000, ParentDir, ZooDefs.Ids.OpenAclUnsafe
, Collections.EmptyList <ZKUtil.ZKAuthInfo>(), cbs[i], CommonConfigurationKeys.HaFcElectorZkOpRetriesDefault
);
}
}
/// <summary>Randomly expire the ZK sessions of the two ZKFCs.</summary>
/// <remarks>
/// Randomly expire the ZK sessions of the two ZKFCs. This differs
/// from the above test in that it is not a controlled failover -
/// we just do random expirations and expect neither one to ever
/// generate fatal exceptions.
/// </remarks>
/// <exception cref="System.Exception"/>
public virtual void TestRandomExpirations()
{
cluster.Start();
long st = Time.Now();
long runFor = StressRuntimeSecs * 1000;
Random r = new Random();
while (Time.Now() - st < runFor)
{
cluster.GetTestContext().CheckException();
int targetIdx = r.Next(2);
ActiveStandbyElector target = cluster.GetElector(targetIdx);
long sessId = target.GetZKSessionIdForTests();
if (sessId != -1)
{
Log.Info(string.Format("Expiring session %x for svc %d", sessId, targetIdx));
GetServer(serverFactory).CloseSession(sessId);
}
Thread.Sleep(r.Next(300));
}
}
/// <exception cref="Org.Apache.Hadoop.HadoopIllegalArgumentException"/>
/// <exception cref="System.IO.IOException"/>
/// <exception cref="Org.Apache.Zookeeper.KeeperException"/>
private void InitZK()
{
zkQuorum = conf.Get(ZkQuorumKey);
int zkTimeout = conf.GetInt(ZkSessionTimeoutKey, ZkSessionTimeoutDefault);
// Parse ACLs from configuration.
string zkAclConf = conf.Get(ZkAclKey, ZkAclDefault);
zkAclConf = ZKUtil.ResolveConfIndirection(zkAclConf);
IList <ACL> zkAcls = ZKUtil.ParseACLs(zkAclConf);
if (zkAcls.IsEmpty())
{
zkAcls = ZooDefs.Ids.CreatorAllAcl;
}
// Parse authentication from configuration.
string zkAuthConf = conf.Get(ZkAuthKey);
zkAuthConf = ZKUtil.ResolveConfIndirection(zkAuthConf);
IList <ZKUtil.ZKAuthInfo> zkAuths;
if (zkAuthConf != null)
{
zkAuths = ZKUtil.ParseAuth(zkAuthConf);
}
else
{
zkAuths = Collections.EmptyList();
}
// Sanity check configuration.
Preconditions.CheckArgument(zkQuorum != null, "Missing required configuration '%s' for ZooKeeper quorum"
, ZkQuorumKey);
Preconditions.CheckArgument(zkTimeout > 0, "Invalid ZK session timeout %s", zkTimeout
);
int maxRetryNum = conf.GetInt(CommonConfigurationKeys.HaFcElectorZkOpRetriesKey,
CommonConfigurationKeys.HaFcElectorZkOpRetriesDefault);
elector = new ActiveStandbyElector(zkQuorum, zkTimeout, GetParentZnode(), zkAcls,
zkAuths, new ZKFailoverController.ElectorCallbacks(this), maxRetryNum);
}
/// <exception cref="System.Exception"/>
public static void WaitForElectorState(MultithreadedTestUtil.TestContext ctx, ActiveStandbyElector
elector, ActiveStandbyElector.State state)
{
while (elector.GetStateForTests() != state)
{
if (ctx != null)
{
ctx.CheckException();
}
Thread.Sleep(50);
}
}