// Receives the heartbeats from different nodes which are part of a shard and
// adds them to the report table.
public void ReceiveHeartbeat(Address source, HeartbeatInfo heartbeatInfo)
{
try
{
if (_localShardHeartbeatReporter != null)
{
// We need to verify if the node sending the heartbeat is part of the existing configuration.
// Updating the config is a costly process so we check the node in the existing config.
// If the node was freshly added in an existing (active) shard, we will not add the heartbeat to the report
// until it exists in the local node config instance.
if (_clusterConfigMgr != null)
{
ShardConfiguration sConfig = _clusterConfigMgr.GetShardConfiguration(_context.LocalShardName);
ServerNode sNode = null;
if (sConfig != null && sConfig.Servers != null)
{
sNode = sConfig.Servers.GetServerNode(source.IpAddress.ToString());
}
if (sNode == null)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("LocalShardCheckHeartbeatTask.ReceiveHeartbeat() ", "The node " + source + " is not part of the configuration.");
}
return;
}
}
bool isAnOldNode = _localShardHeartbeatReporter.AddToReport(source, heartbeatInfo);
if (!isAnOldNode)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("LocalShardCheckHeartbeatTask.ReceiveHeartbeat() ", "Node " + source.IpAddress + " added to the table for the first time. ");
}
lock (_membershipManager)
{
_membershipManager.HeartbeatReport = _localShardHeartbeatReporter;
}
_electionExecTask = new ElectionMechanismExecutionTask(_membershipManager, Activity.NodeJoining, new Server(source, Status.Running));
_electionExecTask.Start();
OnActivityCompleted();
}
}
}
catch (Exception e)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsErrorEnabled)
{
LoggerManager.Instance.ShardLogger.Error("LocalShardCheckHeartbeatTask.ReceiveHeartbeat() ", e.ToString());
}
}
}
// the method which runs in a thread.
public void Run()
{
LoggerManager.Instance.SetThreadContext(new LoggerContext()
{
ShardName = _context.LocalShardName != null ? _context.LocalShardName : "", DatabaseName = ""
});
_context.StatusLatch.WaitForAny(NodeStatus.Running);
_startSignal.WaitOne();
while (_running)
{
IDictionary <Address, HeartbeatInfo> reportTable = null;
try
{
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("LocalShardCheckHeartbeatTask.Run() ", "The shard (or the nodes of the shard) " +
_context.LocalShardName +
" does not exist in the configuration.");
}
return;
}
reportTable = _localShardHeartbeatReporter.GetReportTable;
if (reportTable != null && reportTable.Count > 0)
{
CheckHeartbeats(reportTable);
IList <Address> tentativeLostNodes = CheckForLostNodes(reportTable);
if (tentativeLostNodes != null && tentativeLostNodes.Count > 0)
{
// 1. These are the nodes lost.
// 2. The lost node needs to be removed from the heartbeats table and the missing heartbeats
// 3. Elections will be triggered when a node is lost.
foreach (var node in tentativeLostNodes)
{
if (_localShardHeartbeatReporter.GetReportTable.ContainsKey(node) &&
!ChannelExists(sConfig.Servers.GetServerNode(node.IpAddress.ToString())))
{
if (LoggerManager.Instance.ShardLogger != null &&
LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("LocalShard.CheckHeartbeatTask",
"did not receive heart beat from " + node);
}
//_localShardHeartbeatReporter.RemoveFromReport(node);
OnActivityTriggered(Activity.NodeLeaving, node);
lock (_membershipManager)
{
_membershipManager.HeartbeatReport =
(LocalShardHeartbeatReporting)_localShardHeartbeatReporter;
}
_electionExecTask = new ElectionMechanismExecutionTask(_membershipManager,
Activity.NodeLeaving, new Server(node, Status.Stopped));
_electionExecTask.Start();
OnActivityCompleted();
}
else
{
if (LoggerManager.Instance.ShardLogger != null &&
LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("LocalShard.CheckHeartbeatTask",
"did not receive heart beat from " + node + ", however channel is connected");
}
}
}
}
//if no primary is set, we proceed to the following steps:
if (!_shard.NodeRole.Equals(NodeRole.Intermediate) && _membershipManager.LatestMembership == null ||
( /*_membershipManager.LatestMembership != null &&*/
(_membershipManager.LatestMembership.Primary == null ||
( /*_membershipManager.LatestMembership.Primary != null &&*/
!ChannelExists(_membershipManager.LatestMembership.Primary)))))
{
//First, check if the conduction of elections is feasible.
//If yes, the actual election mechanism is triggered.
if (AreElectionsFeasible(_localShardHeartbeatReporter, sConfig))
{
lock (_membershipManager)
{
_membershipManager.HeartbeatReport =
(LocalShardHeartbeatReporting)_localShardHeartbeatReporter;
}
_electionExecTask = new ElectionMechanismExecutionTask(_membershipManager, Activity.GeneralElectionsTriggered, null);
_electionExecTask.Start();
}
//RTD: badddd logic.
//else if (_localShardHeartbeatReporter.PrimaryExists() &&
// _localShardHeartbeatReporter.GetCurrentPrimary() != null &&
// ChannelExists(_localShardHeartbeatReporter.GetCurrentPrimary()))
//{
// ServerNode currentPrimaNode = _localShardHeartbeatReporter.GetCurrentPrimary();
// if (currentPrimaNode != null)
// {
// if (LoggerManager.Instance.ShardLogger != null &&
// LoggerManager.Instance.ShardLogger.IsDebugEnabled)
// LoggerManager.Instance.ShardLogger.Debug("LocalShard.CheckHeartbeatTask",
// "no primary set so far, setting " + currentPrimaNode.Name +
// " as primary as other nodes are reporting");
// MembershipChangeArgs args = new MembershipChangeArgs();
// args.ChangeType = MembershipChangeArgs.MembershipChangeType.PrimarySet;
// args.ElectionId = _localShardHeartbeatReporter.GetCurrentElectionId();
// args.ServerName = new Address(currentPrimaNode.Name, sConfig.Port);
// ((LocalShard)_shard).OnMembershipChanged(args);
// }
//}
}
LogMembership();
}
}
catch (ThreadAbortException e)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsErrorEnabled && _checkHbThread != null)
{
LoggerManager.Instance.ShardLogger.Error(_checkHbThread.Name, "Task aborted.");
}
break;
}
catch (Exception e)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsErrorEnabled)
{
LoggerManager.Instance.ShardLogger.Error("LocalShardCheckHeartbeatTask.Run() ", e.ToString());
}
}
lock (_syncMutex)
{
Monitor.Wait(_syncMutex, _poolingThreshold);
}
_startSignal.WaitOne();
}
}