public object OnMessageReceived(object message, Server source, LocalShardHeartbeatReporting heartbeatReport, Membership existingMembership)
{
Election election = message as Election;
if (election != null)
{
HeartbeatInfo localHeartbeat = heartbeatReport.GetHeartbeatInfo(_context.LocalAddress);
if (localHeartbeat != null)
{
return(_manager.CastVote(election, existingMembership, localHeartbeat));
}
}
return(null);
}
/// <summary>
/// Speculative Phase:
/// 1. We search for the node with the latest op-log entry. This detail bears the highest value.
/// 2. Next, if multiple nodes have the same op-log entry, we move onto the next step.
/// 3. All those nodes which have the same op-log entry AND are connected to the CS are considered.
/// 4. Highest priority from amongst these active nodes are taken into consideration.
/// 5. If this node fulfills all of the above, it successfully passes the speculative phase.
/// </summary>
/// <param name="heartbeatReport"></param>
/// <returns></returns>
private bool ShouldIInitiateElection(LocalShardHeartbeatReporting heartbeatReport, Activity activity)
{
IList <Address> activeNodes = null;
if (heartbeatReport != null)
{
activeNodes = heartbeatReport.GetReportTable.Keys.ToList();
OperationId maxOplog = null;
IList <string> matchingOplogServerIPs = new List <string>();
HeartbeatInfo localHeartbeat = heartbeatReport.GetHeartbeatInfo(_context.LocalAddress);
if (localHeartbeat == null)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsWarnEnabled)
{
LoggerManager.Instance.ShardLogger.Warn(
"ElectionBasedMembershipStrategy.ShouldIInititateElections()",
"local node heartbeat is null");
}
return(false);
}
OperationId lastRepId = null;
if (activity.Equals(Activity.TakeoverElectionsTriggered))
{
HeartbeatInfo info = null;
if (_shard != null && _shard.Primary != null)
{
info = heartbeatReport.GetHeartbeatInfo(_shard.Primary.Address);
}
}
else
{
ShardConfiguration sConfig = _clusterConfigMgr.GetShardConfiguration(_context.LocalShardName);
int configuredNodesCount = 0;
OperationId OpIdAtCS = null;
if (sConfig != null && sConfig.Servers != null && sConfig.Servers.Nodes != null)
{
configuredNodesCount = sConfig.Servers.Nodes.Count;
}
if (configuredNodesCount > 0 && activeNodes != null && activeNodes.Count < configuredNodesCount)
{
ShardInfo sInfo = null;
ClusterInfo cInfo = _context.ConfigurationSession.GetDatabaseClusterInfo(_context.ClusterName);
if (cInfo != null)
{
sInfo = cInfo.GetShardInfo(_context.LocalShardName);
}
if (sInfo != null)
{
OpIdAtCS = sInfo.LastOperationId;
}
if (OpIdAtCS > lastRepId)
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsInfoEnabled)
{
LoggerManager.Instance.ShardLogger.Info(
"electBasedMemSt.ShouldIInitElections()",
"CS has an operation newer than my operation. Hence, waiting.");
}
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsInfoEnabled)
{
if (OpIdAtCS != null)
{
LoggerManager.Instance.ShardLogger.Info(
"electBasedMemSt.ShouldIInitElections()",
"Operation ID on the CS:- " + OpIdAtCS.ElectionId + ":"
+ OpIdAtCS.ElectionBasedSequenceId);
}
else
{
LoggerManager.Instance.ShardLogger.Info(
"electBasedMemSt.ShouldIInitElections()", "The operation ID at the CS is set to null.");
}
if (lastRepId != null)
{
LoggerManager.Instance.ShardLogger.Info(
"electBasedMemSt.ShouldIInitElections()", "Local node Operation ID:- " + lastRepId.ElectionId +
":" + lastRepId.ElectionBasedSequenceId);
}
else
{
LoggerManager.Instance.ShardLogger.Info(
"electBasedMemSt.ShouldIInitElections()", "The local node operation ID is set to null.");
}
}
//We maintain the last replicated operation log entry with the CS.
//If a node in a shard with older data (usually the previous secondary) is up before the node with the
//latest data(usually the previous primary), it waits for a configurable amount of time (2 minutes for
//now) before proceeding with the election procedure if it is still unable to detect a primary node.
//This way we give the node with the latest data a chance to become primary and therefore avoid data loss.
lock (_mutexOnWait)
{
Monitor.Wait(_mutexOnWait, _waitTimeout);
}
}
}
for (int i = 0; i < activeNodes.Count; i++)
{
HeartbeatInfo info = heartbeatReport.GetHeartbeatInfo(activeNodes[i]);
OperationId currIndexOplog = info.LastOplogOperationId;
if (currIndexOplog > maxOplog)
{
maxOplog = currIndexOplog;
}
if (((localHeartbeat.LastOplogOperationId == null && info.LastOplogOperationId == null) || localHeartbeat.LastOplogOperationId != null && localHeartbeat.LastOplogOperationId.Equals(info.LastOplogOperationId)) &&
info.CSStatus == ConnectivityStatus.Connected)
{
matchingOplogServerIPs.Add(activeNodes[i].IpAddress.ToString());
}
}
}
if (localHeartbeat.LastOplogOperationId != null && maxOplog != null && maxOplog > localHeartbeat.LastOplogOperationId && (lastRepId == null || maxOplog > lastRepId))
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("electBasedMemSt.ShouldIInitiateElect()", "Local operation log is behind max op log wrt " + _context.LocalShardName + " shard.");
}
if (maxOplog != null && LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("electBasedMemSt.ShouldIInitiateElect()", "maxOplog: " + maxOplog.ElectionId + ":" + maxOplog.ElectionBasedSequenceId);
}
if (localHeartbeat.LastOplogOperationId != null && LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("electBasedMemSt.ShouldIInitiateElect()", "local opLog (from the heartbeat): " + localHeartbeat.LastOplogOperationId.ElectionId + ":" + localHeartbeat.LastOplogOperationId.ElectionBasedSequenceId);
}
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled && lastRepId != null)
{
LoggerManager.Instance.ShardLogger.Debug("electBasedMemSt.ShouldIInitiateElect()", "LastOpFromOpLog (from the replication module): " + lastRepId.ElectionId + ":" + lastRepId.ElectionBasedSequenceId);
}
return(false);
}
else if (maxOplog == localHeartbeat.LastOplogOperationId || (lastRepId != null && lastRepId.Equals(maxOplog)))
{
//if: there are multiple nodes that have the same oplog entry,
//decision will be made on the basis of the priorities.
//else: the node with the highest oplog entry will be considered eligible.
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("electBasedMemSt.ShouldIInitiateElect()", "Local operation log is equal to the max op log wrt " + _context.LocalShardName + " shard.");
}
if (maxOplog != null && (localHeartbeat != null && localHeartbeat.LastOplogOperationId != null))
{
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("electBasedMemSt.ShouldIInitiateElect()", "maxOplog: " + maxOplog.ElectionId + ":" + maxOplog.ElectionBasedSequenceId);
}
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug("electBasedMemSt.ShouldIInitiateElect()", "local opLog (from the heartbeat): " + localHeartbeat.LastOplogOperationId.ElectionId + ":" + localHeartbeat.LastOplogOperationId.ElectionBasedSequenceId);
}
if (LoggerManager.Instance.ShardLogger != null && LoggerManager.Instance.ShardLogger.IsDebugEnabled && lastRepId != null)
{
LoggerManager.Instance.ShardLogger.Debug("electBasedMemSt.ShouldIInitiateElect()", "LastOpFromOpLog (from the replication module): " + lastRepId.ElectionId + ":" + lastRepId.ElectionBasedSequenceId);
}
}
if (matchingOplogServerIPs.Count > 0)
{
int highestRunningNodePriority = _manager.GetHighestNodePriority(activeNodes,
matchingOplogServerIPs);
if (highestRunningNodePriority.Equals(_manager.LocalServerPriority))
{
if (LoggerManager.Instance.ShardLogger != null &&
LoggerManager.Instance.ShardLogger.IsDebugEnabled)
{
LoggerManager.Instance.ShardLogger.Debug(
"ElectionBasedMembershipStrategy.ShouldIInitiateElection()",
"Node : " + _context.LocalAddress.IpAddress.ToString() + " in shard: " +
_context.LocalShardName + " is eligible having priority: " +
highestRunningNodePriority + " .");
}
return(true);
}
else
{
return(false);
}
}
}
}
return(false);
}