private void ChangeMembershipShardwide(MembershipChangeArgs args)
{
Message message = new Message();
message.NeedsResponse = false;
message.MessageType = MessageType.MembershipOperation;
//Since, in order to handle deadlock scenarios, the message cannot be sent to the local node,
// this method is being called in its stead.
((LocalShard)_shard).OnMembershipChanged(args);
message.Payload = args;
IList <Server> activeList = new List <Server>();
foreach (var node in _shard.ActiveChannelsList)
{
if (!node.Address.Equals(_context.LocalAddress))
{
activeList.Add(node);
}
}
ShardMulticastRequest <ResponseCollection <object>, object> request = _shard.CreateMulticastRequest <ResponseCollection <object>, object>(activeList, message);
IAsyncResult asyncResult = request.BeginExecute();
request.EndExecute(asyncResult);
}
/// <summary>
/// We need to ask the current primary to quit taking write ops
/// so that the left over ops can first be replicated
/// before the actual election mechanism begins.
/// </summary>
/// <returns></returns>
private bool RequestPrimaryToStopOperations()
{
ShardConfiguration sConfg = null;
bool endResult = false;
if (_clusterConfigMgr != null)
{
sConfg = _clusterConfigMgr.GetShardConfiguration(_context.LocalShardName);
}
if (sConfg != null)
{
MembershipChangeArgs args = new MembershipChangeArgs();
args.ChangeType = MembershipChangeArgs.MembershipChangeType.RestrictPrimary;
args.ServerName = _context.LocalAddress;
args.ElectionId = _latestMembership.ElectionId;
DatabaseMessage msg = new DatabaseMessage();
msg.Payload = args;
msg.NeedsResponse = true;
msg.OpCode = OpCode.RestrictPrimary;
msg.MessageType = MessageType.DBOperation;
ShardRequestBase <bool> request = _shard.CreateUnicastRequest <bool>(new Server(new Address(_latestMembership.Primary.Name, sConfg.Port), Status.Running), msg);
IAsyncResult result = request.BeginExecute();
endResult = request.EndExecute(result);
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsInfoEnabled)
{
LoggerManager.Instance.ShardLogger.Info("MembershipMgr.RequestPrimaryToStopOperations()", "Requested primary to stop taking write operations. Primary response: " + endResult.ToString());
}
}
return(endResult);
}
public object Clone()
{
MembershipChangeArgs args = new MembershipChangeArgs();
args.ServerName = ServerName;
args.ElectionId = ElectionId;
args.ChangeType = ChangeType;
return(args);
}
internal bool AbortTakeoverMechanismTask(MembershipChangeArgs args)
{
if (args != null && args.ChangeType == MembershipChangeArgs.MembershipChangeType.TimeoutOnRestrictedPrimary)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsInfoEnabled)
{
LoggerManager.Instance.ShardLogger.Info("MembershipMgr.AbortTakeoverMechanismTask()", "Takeover unsuccessful. Beginning takeover retry task.");
}
if (_retryTask == null)
{
_retryTask = new TakeoverRetryTask(this);
}
_retryTask.Start();
if (_takeOverElectionTask != null && _takeOverElectionTask.IsStarted())
{
_takeOverElectionTask.Stop();
}
return(true);
}
return(false);
}
internal bool OnForcefulPrimaryDemotion(MembershipChangeArgs args)
{
try
{
if (args != null && args.ChangeType == MembershipChangeArgs.MembershipChangeType.ForcefullyDemotePrimary && args.ServerName != null)
{
if (_strategy != null)
{
lock (_mutex)
{
_strategy.TriggerElectionMechanism(Activity.ForcefulPrimaryDemotion, new Server(args.ServerName, Status.Running), _heartbeatReporting, _latestMembership);
if (_latestMembership != null && (_latestMembership.Primary == null || _latestMembership.Primary.Name != _context.LocalAddress.IpAddress.ToString()))
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("MembershipManager.OnForcefulPrimaryDemotion()", "Call for forceful demotion of the primary node received. Primary demoted successfully.");
}
return(true);
}
}
}
}
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsInfoEnabled)
{
LoggerManager.Instance.ShardLogger.Info("MembershipManager.OnForcefulPrimaryDemotion()", "Forceful primary demotion unsuccessful.");
}
return(false);
}
catch (Exception e)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsErrorEnabled)
{
LoggerManager.Instance.ShardLogger.Error("MembershipMgr.OnForcefulPrimaryDemotion(): ", e.StackTrace);
}
return(false);
}
}
public void TriggerElectionMechanism(Activity activity, Server server, LocalShardHeartbeatReporting heartbeatReport, Membership existingMembership)
{
LoggerManager.Instance.SetThreadContext(new LoggerContext()
{
ShardName = _context.LocalShardName != null ? _context.LocalShardName : "", DatabaseName = ""
});
if (existingMembership == null)
{
existingMembership = new Membership();
}
ShardConfiguration sConfig = null;
//Get the shard configuration
if (_clusterConfigMgr != null)
{
sConfig = _clusterConfigMgr.GetShardConfiguration(_context.LocalShardName);
}
IList <Address> activeNodes = null;
MembershipChangeArgs args = new MembershipChangeArgs();
ServerNodes staticServerNodes = null;
if (sConfig == null || sConfig.Servers == null)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsWarnEnabled)
{
LoggerManager.Instance.ShardLogger.Warn("ElectionBasedMembershipStrategy.TiggerElectionMechanism() ", "The shard " + _context.LocalShardName + " does not exist in the configuration.");
}
return;
}
staticServerNodes = sConfig.Servers;
ElectionResult result = null;
if (heartbeatReport != null)
{
activeNodes = heartbeatReport.GetReportTable.Keys.ToList();
}
Address activityNode = null;
if (server == null)
{
activityNode = _context.LocalAddress;
}
else
{
activityNode = server.Address;
}
switch (activity)
{
case Activity.NodeJoining:
if (server == null)
{
return;
}
//On node join, we need to get membership from the config server for the first time.
Membership csMembership = _context.ConfigurationSession.GetMembershipInfo(_context.ClusterName, _context.LocalShardName);
ServerNode joiningNode = sConfig.Servers.GetServerNode(server.Address.IpAddress.ToString());
// If the added node is configured while the cluster is up and running, do the following.
if (joiningNode == null)
{
if (_clusterConfigMgr != null)
{
_clusterConfigMgr.UpdateClusterConfiguration();
sConfig = _clusterConfigMgr.GetShardConfiguration(_context.LocalShardName);
}
if (sConfig == null)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsWarnEnabled)
{
LoggerManager.Instance.ShardLogger.Warn("ElectionBasedMembershipStrategy.TriggerElectionMechanism() ", "The shard " + _context.LocalShardName + " does not exist in the configuration.");
}
return;
}
joiningNode = sConfig.Servers.GetServerNode(server.Address.IpAddress.ToString());
}
if (joiningNode == null)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsWarnEnabled)
{
LoggerManager.Instance.ShardLogger.Warn("ElectionBasedMembershipStrategy.TriggerElectionMechanism() ", "The node " + server.Address + " is not part of the configuration.");
}
return;
}
if (existingMembership == null || existingMembership.Servers == null || !existingMembership.Servers.Contains(joiningNode))
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("ElectBasedMemSt.TriggerElectMech", "Node joining activity triggered for " + activityNode);
}
}
bool thisNodeIsPrimary = false;
OperationId lastOpId = null;
if (heartbeatReport != null && heartbeatReport.GetReportTable.ContainsKey(server.Address))
{
args.ServerName = _context.LocalAddress;
args.ElectionId = null;
args.ChangeType = MembershipChangeArgs.MembershipChangeType.NodeJoined;
if (server.Address.Equals(_context.LocalAddress))
{
_context.ConfigurationSession.ReportNodeJoining(_context.ClusterName, _context.LocalShardName, sConfig.Servers.GetServerNode(_context.LocalAddress.IpAddress.ToString()));
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsInfoEnabled)
{
LoggerManager.Instance.ShardLogger.Info("electBasedMemSt.TriggerElectMech", server.Address + " reported its joining to the config server. ");
}
//if the primary is not null and the channel is not disconnected, it can be set here.
if ((existingMembership == null || existingMembership.Primary == null) && csMembership.Primary != null && _shard.ActiveChannelsList.Contains(new Server(new Address(csMembership.Primary.Name, sConfig.Port), Status.Initializing)) && ObeysMajorityRule(_shard.ActiveChannelsList.Count, sConfig.Servers.Nodes.Count))
{
args.ServerName = new Address(csMembership.Primary.Name, sConfig.Port);
args.ElectionId = csMembership.ElectionId;
args.ChangeType = MembershipChangeArgs.MembershipChangeType.PrimarySet;
//if the node which was lost comes back up before the CS or the nodes can declare it dead,
//it should resume its status as a primary. There should be no need for an election in this case.
if (args.ServerName.Equals(_context.LocalAddress))
{
thisNodeIsPrimary = true;
}
}
}
if (thisNodeIsPrimary)
{
if (csMembership.ElectionId != null && LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsInfoEnabled)
{
LoggerManager.Instance.ShardLogger.Info("electBasedMemSt.TriggerElectMech", "election_id: " + csMembership.ElectionId.Id + " election time :" + csMembership.ElectionId.ElectionTime);
}
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsInfoEnabled)
{
LoggerManager.Instance.ShardLogger.Info("electBasedMemSt.TriggerElectMech", "I am already declared primary");
}
lastOpId = LastOperationId;
ChangeMembershipShardwide(args);
}
else
{
((LocalShard)_shard).OnMembershipChanged(args);
}
if (server.Address.Equals(_context.LocalAddress))
{
ServerNode sNode = sConfig.Servers.GetServerNode(_context.LocalAddress.IpAddress.ToString());
if (sNode == null)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsWarnEnabled)
{
LoggerManager.Instance.ShardLogger.Warn("ElectionBasedMembershipStrategy.TriggerElectionMechanism() ", "The node " + sNode.Name + " does not exist in the configuration.");
}
return;
}
_context.ConfigurationSession.ReportHeartbeat(_context.ClusterName, _context.LocalShardName, sNode, existingMembership, lastOpId);
}
}
else
{
if (existingMembership.Primary != null && existingMembership.Primary.Name.Equals(server.Address.IpAddress.ToString()))
{
if (sConfig.Servers == null || sConfig.Servers.Nodes == null || !ObeysMajorityRule(activeNodes.Count, sConfig.Servers.Nodes.Count))
{
_context.ConfigurationSession.SetNodeStatus(_context.ClusterName, _context.LocalShardName, existingMembership.Primary, NodeRole.None);
args.ChangeType = MembershipChangeArgs.MembershipChangeType.PrimaryDemoted;
args.ServerName = _context.LocalAddress;
ChangeMembershipShardwide(args);
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsInfoEnabled)
{
LoggerManager.Instance.ShardLogger.Info("electBasedMemSt.TriggerElectMech", " Node addition activity occured. Primary node " + _context.LocalAddress.IpAddress.ToString() + " demoted.");
}
return;
}
}
}
break;
case Activity.NodeLeaving:
if (server == null)
{
return;
}
bool hasMajority = ObeysMajorityRule(activeNodes.Count, staticServerNodes.Nodes.Count);
args.ServerName = server.Address;
args.ChangeType = MembershipChangeArgs.MembershipChangeType.NodeLeft;
_clusterConfigMgr.UpdateClusterConfiguration();
if (existingMembership.Primary != null)
{
// if the existing primary is actually the node lost, we need to update the configuration.
if (existingMembership.Primary.Name == server.Address.IpAddress.ToString())
{
//if Primary leaves, it should be updated locally.
args.ChangeType = MembershipChangeArgs.MembershipChangeType.PrimaryLost;
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("electBasedMemSt.TriggerElectMech", "Node leaving activity triggered for " + server.Address + " . Primary lost.");
}
}
else if (existingMembership.Primary.Name == _context.LocalAddress.IpAddress.ToString()) // if the existing primary is the local node, we need to check for possible demotion of the current primary.
{
if (!hasMajority)
{
_context.ConfigurationSession.SetNodeStatus(_context.ClusterName, _context.LocalShardName, existingMembership.Primary, NodeRole.None);
args.ChangeType = MembershipChangeArgs.MembershipChangeType.PrimaryDemoted;
args.ServerName = _context.LocalAddress;
ChangeMembershipShardwide(args);
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("electBasedMemSt.TriggerElectMech", " Node leaving activity occurred. Primary node " + _context.LocalAddress.IpAddress.ToString() + " demoted.");
}
return;
}
}
}
((LocalShard)_shard).OnMembershipChanged(args);
break;
case Activity.GeneralElectionsTriggered:
case Activity.TakeoverElectionsTriggered:
// this is where the actual election mechanism takes place.
//Step 1: if no node in the heartbeat table has a primary and there is no primary in the local node's membership, we proceed forward.
//Else if there is a primary but this looks like the takeover election mechanism, we proceed along as well.
if ((activity.Equals(Activity.GeneralElectionsTriggered) && !heartbeatReport.PrimaryExists() && existingMembership.Primary == null) || (activity.Equals(Activity.TakeoverElectionsTriggered) && heartbeatReport.PrimaryExists()))
{
//Step 2: we verify that this node has a majority of the shard nodes connected to it.
if (activeNodes != null && ObeysMajorityRule(activeNodes.Count, staticServerNodes.Nodes.Count))
{
//Step 3: Perform the initial sanity check. (Speculative phase)
if (ShouldIInitiateElection(heartbeatReport, activity))
{
if (existingMembership != null && existingMembership.Primary != null && activity == Activity.GeneralElectionsTriggered)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsInfoEnabled)
{
LoggerManager.Instance.ShardLogger.Info("electBasedMemSt.TriggerElectMech", "A primary has already been selected for "
+ _context.LocalShardName + " hence exiting the election mechanism.");
}
return;
}
//Step 4: The elections take place in real. (Authoritative Phase)
result = HoldElection(heartbeatReport, activity);
if (result != null)
{
if (result.PollingResult == ElectionResult.Result.PrimarySelected)
{
//if the shard is undergoing the takeover election mechanism, the old primary needs to
//be demoted first.
bool oldPrimaryDemoted = false;
if (activity == Activity.TakeoverElectionsTriggered)
{
MembershipChangeArgs args2 = new MembershipChangeArgs();
args2.ChangeType = MembershipChangeArgs.MembershipChangeType.ForcefullyDemotePrimary;
args2.ServerName = _context.LocalAddress;
args2.ElectionId = existingMembership.ElectionId;
Message msg = new Message();
msg.Payload = args2;
msg.MessageType = MessageType.MembershipOperation;
msg.NeedsResponse = true;
ShardRequestBase <bool> request = _shard.CreateUnicastRequest <bool>(new Server(new Address(existingMembership.Primary.Name, sConfig.Port), Status.Running), msg);
IAsyncResult result2 = request.BeginExecute();
oldPrimaryDemoted = request.EndExecute(result2);
}
//Submit the result to the CS.
if (activity == Activity.GeneralElectionsTriggered || (activity == Activity.TakeoverElectionsTriggered && oldPrimaryDemoted))
{
_context.ConfigurationSession.SubmitElectionResult(_context.ClusterName.ToLower(), _context.LocalShardName.ToLower(), result);
}
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("electBasedMemSt.TriggerElectMech", "Election result submitted for shard " + _context.LocalShardName.ToString());
}
_context.ElectionResult = result;
args.ServerName = _context.LocalAddress;
args.ElectionId = result.ElectionId;
args.ChangeType = MembershipChangeArgs.MembershipChangeType.PrimarySelected;
//Once, the result is submitted, inform the shard nodes.
ChangeMembershipShardwide(args);
_context.ConfigurationSession.ReportHeartbeat(_context.ClusterName, _context.LocalShardName, result.ElectedPrimary, existingMembership, LastOperationId);
}
//Finally, end this round of elections.
_context.ConfigurationSession.EndElection(_context.ClusterName, _context.LocalShardName, result.ElectionId);
}
}
}
}
break;
case Activity.CSDisconnected:
//this is called whenever a node loses connection with the config server.
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("ElectionBasedMembershipStrategy.TriggerElectionMechanism() ", "Config Server disconnected. ");
}
//if the number of configured nodes are even and the primary loses connection with the CS, it needs to demote itself.
if (existingMembership != null && existingMembership.Primary != null && existingMembership.Primary.Name == _context.LocalAddress.IpAddress.ToString() && staticServerNodes.Nodes.Count % 2 == 0)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("electBasedMemSt.TriggerElectMech", " Connection of the node " + _context.LocalAddress.ToString() + " with the config server is lost.");
}
args.ServerName = _context.LocalAddress;
args.ElectionId = existingMembership.ElectionId;
args.ChangeType = MembershipChangeArgs.MembershipChangeType.PrimaryDemoted;
ChangeMembershipShardwide(args);
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsInfoEnabled)
{
LoggerManager.Instance.ShardLogger.Info("electBasedMemSt.TriggerElectMech", " Primary node " + _context.LocalAddress.IpAddress.ToString() + " demoted because the primary lost connection with the CS.");
}
}
break;
case Activity.ForcefulPrimaryDemotion:
if (existingMembership != null && existingMembership.Primary != null && existingMembership.Primary.Name == _context.LocalAddress.IpAddress.ToString())
{
_context.ConfigurationSession.SetNodeStatus(_context.ClusterName, _context.LocalShardName, existingMembership.Primary, NodeRole.None);
args.ChangeType = MembershipChangeArgs.MembershipChangeType.PrimaryDemoted;
args.ServerName = _context.LocalAddress;
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsInfoEnabled)
{
LoggerManager.Instance.ShardLogger.Info("electBasedMemSt.TriggerElectMech", "Primary node " + _context.LocalAddress.IpAddress.ToString() + " demoted in order to complete the take over election mechanism. ");
}
((LocalShard)_shard).OnMembershipChanged(args);
}
break;
}
}
internal void UpdateLocalMembership(MembershipChangeArgs args)
{
LoggerManager.Instance.SetThreadContext(new LoggerContext()
{
ShardName = _context.LocalShardName != null ? _context.LocalShardName : "", DatabaseName = ""
});
ShardConfiguration sConfig = null;
if (_clusterConfigMgr != null)
{
sConfig = _clusterConfigMgr.GetShardConfiguration(_context.LocalShardName);
}
if (sConfig == null || sConfig.Servers == null)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsWarnEnabled)
{
LoggerManager.Instance.ShardLogger.Warn("MembershipManager.UpdateLocalMembership() ", "The shard (or the nodes of the shard) " + _context.LocalShardName + " does not exist in the configuration.");
}
return;
}
ServerNode affectedServer = sConfig.Servers.GetServerNode(args.ServerName.IpAddress.ToString());
switch (args.ChangeType)
{
case MembershipChangeArgs.MembershipChangeType.NodeJoined:
lock (_mutexOnUpdateConfig)
{
_latestMembership.AddServer(affectedServer);
}
break;
case MembershipChangeArgs.MembershipChangeType.NodeLeft:
lock (_mutexOnUpdateConfig)
{
_latestMembership.RemoveServer(affectedServer);
}
break;
case MembershipChangeArgs.MembershipChangeType.PrimarySelected:
case MembershipChangeArgs.MembershipChangeType.PrimarySet:
lock (_mutexOnUpdateConfig)
{
_latestMembership.AddServer(affectedServer);
if (args.ServerName != null)
{
_latestMembership.Primary = sConfig.Servers.GetServerNode(args.ServerName.IpAddress.ToString());
}
if (args.ElectionId != null)
{
_latestMembership.ElectionId = args.ElectionId;
}
SanityCheckForTakeoverElect();
}
_strategy.OnPrimaryChanged();
break;
case MembershipChangeArgs.MembershipChangeType.PrimaryLost:
case MembershipChangeArgs.MembershipChangeType.PrimaryDemoted:
lock (_mutexOnUpdateConfig)
{
if (args.ChangeType.Equals(MembershipChangeArgs.MembershipChangeType.PrimaryLost))
{
_latestMembership.RemoveServer(affectedServer);
}
_latestMembership.Primary = null;
_latestMembership.ElectionId = null;
}
break;
}
if (args.ServerName != null && args.ChangeType != MembershipChangeArgs.MembershipChangeType.None)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("MembershipManager.UpdateLocalMembership() ", "Membership updated: " + args.ChangeType.ToString() + " of node " + args.ServerName.IpAddress.ToString());
}
}
}