public void SiloStatusChangeNotification(SiloAddress updatedSilo, SiloStatus status)
{
// This silo's status has changed
if (Equals(updatedSilo, MyAddress))
{
if (status == SiloStatus.Stopping || status.Equals(SiloStatus.ShuttingDown))
{
// QueueAction up the "Stop" to run on a system turn
Scheduler.QueueAction(() => Stop(true), CacheValidator.SchedulingContext).Ignore();
}
else if (status == SiloStatus.Dead)
{
// QueueAction up the "Stop" to run on a system turn
Scheduler.QueueAction(() => Stop(false), CacheValidator.SchedulingContext).Ignore();
}
}
else // Status change for some other silo
{
if (status.IsTerminating())
{
// QueueAction up the "Remove" to run on a system turn
Scheduler.QueueAction(() => RemoveServer(updatedSilo, status), CacheValidator.SchedulingContext).Ignore();
}
else if (status.Equals(SiloStatus.Active)) // do not do anything with SiloStatus.Starting -- wait until it actually becomes active
{
// QueueAction up the "Remove" to run on a system turn
Scheduler.QueueAction(() => AddServer(updatedSilo), CacheValidator.SchedulingContext).Ignore();
}
}
}
public void SiloStatusChangeNotification(SiloAddress updatedSilo, SiloStatus status)
{
if (!status.IsTerminating())
{
return;
}
SiloRuntimeStatistics ignore;
periodicStats.TryRemove(updatedSilo, out ignore);
NotifyAllStatisticsChangeEventsSubscribers(updatedSilo, null);
}
private void AssertStatus(SiloAddress address, SiloStatus expected)
{
var localStatus = this.oracle.GetApproximateSiloStatus(address);
Assert.Equal(expected, localStatus);
if (address.Equals(this.siloDetails.SiloAddress))
{
Assert.Equal(localStatus, this.oracle.CurrentStatus);
}
Assert.Equal(!address.Equals(this.siloDetails.SiloAddress) && expected == SiloStatus.Dead, this.oracle.IsDeadSilo(address));
Assert.Equal(address.Equals(this.siloDetails.SiloAddress) || !expected.IsTerminating(), this.oracle.IsFunctionalDirectory(address));
}
public void SiloStatusChangeNotification(SiloAddress updatedSilo, SiloStatus status)
{
// This silo's status has changed
if (updatedSilo.Equals(MyAddress))
{
if (status.IsTerminating())
{
Stop();
}
}
else // Status change for some other silo
{
if (status.IsTerminating())
{
RemoveServer(updatedSilo);
}
else if (status.Equals(SiloStatus.Active)) // do not do anything with SiloStatus.Created or SiloStatus.Joining -- wait until it actually becomes active
{
AddServer(updatedSilo);
}
}
}
public void SiloStatusChangeNotification(SiloAddress updatedSilo, SiloStatus status)
{
if (status.IsTerminating() && updatedSilo.Equals(this.Silo))
{
refreshTimer?.Dispose();
}
else if (status != SiloStatus.Dead)
{
return;
}
scheduler.QueueTask(() => OnClientRefreshTimer(null), SchedulingContext).Ignore();
}
private void OnSiloStatusChange(SiloAddress updatedSilo, SiloStatus status)
{
if (!status.IsTerminating())
{
return;
}
if (Equals(updatedSilo, this.Silo))
{
this.publishTimer.Dispose();
}
periodicStats.TryRemove(updatedSilo, out _);
NotifyAllStatisticsChangeEventsSubscribers(updatedSilo, null);
}
public void SiloStatusChangeNotification(SiloAddress updatedSilo, SiloStatus status)
{
// This silo's status has changed
if (!Equals(updatedSilo, MyAddress)) // Status change for some other silo
{
if (status.IsTerminating())
{
// QueueAction up the "Remove" to run on a system turn
CacheValidator.WorkItemGroup.QueueAction(() => RemoveServer(updatedSilo, status));
}
else if (status == SiloStatus.Active) // do not do anything with SiloStatus.Starting -- wait until it actually becomes active
{
// QueueAction up the "Remove" to run on a system turn
CacheValidator.WorkItemGroup.QueueAction(() => AddServer(updatedSilo));
}
}
}
public void SiloStatusChangeNotification(SiloAddress updatedSilo, SiloStatus status)
{
// This silo's status has changed
if (updatedSilo.Equals(myAddress))
{
if (status.IsTerminating())
{
Stop();
}
}
else // Status change for some other silo
{
if (status.IsTerminating())
{
RemoveServer(updatedSilo);
}
else if (status.Equals(SiloStatus.Active)) // do not do anything with SiloStatus.Created or SiloStatus.Joining -- wait until it actually becomes active
{
AddServer(updatedSilo);
}
}
}
public async Task UpdateStatus(SiloStatus status)
{
string errorString = null;
int numCalls = 0;
try
{
Func <int, Task <bool> > updateMyStatusTask = async counter =>
{
numCalls++;
if (log.IsEnabled(LogLevel.Debug))
{
log.Debug("-Going to try to TryUpdateMyStatusGlobalOnce #{0}", counter);
}
return(await TryUpdateMyStatusGlobalOnce(status)); // function to retry
};
if (status == SiloStatus.Dead && this.membershipTableProvider is SystemTargetBasedMembershipTable)
{
// SystemTarget-based membership may not be accessible at this stage, so allow for one quick attempt to update
// the status before continuing regardless of the outcome.
var updateTask = updateMyStatusTask(0);
updateTask.Ignore();
var result = await Task.WhenAny(Task.Delay(TimeSpan.FromSeconds(5)), updateTask);
if (ReferenceEquals(result, updateTask))
{
await result;
}
else
{
this.log.LogWarning(
"Failed to update status to dead in the alotted time during shutdown");
}
this.CurrentStatus = status;
return;
}
bool ok = await MembershipExecuteWithRetries(updateMyStatusTask, this.clusterMembershipOptions.MaxJoinAttemptTime);
if (ok)
{
if (log.IsEnabled(LogLevel.Debug))
{
log.Debug("-Silo {0} Successfully updated my Status in the Membership table to {1}", myAddress, status);
}
var gossipTask = this.GossipToOthers(this.myAddress, status);
var timeoutTask = Task.Delay(GossipTimeout);
var task = await Task.WhenAny(gossipTask, timeoutTask);
if (ReferenceEquals(task, timeoutTask))
{
if (status.IsTerminating())
{
this.log.LogWarning("Timed out while gossiping status to other silos after {Timeout}", GossipTimeout);
}
else if (this.log.IsEnabled(LogLevel.Debug))
{
this.log.LogDebug("Timed out while gossiping status to other silos after {Timeout}", GossipTimeout);
}
}
}
else
{
errorString = $"-Silo {myAddress} failed to update its status to {status} in the Membership table due to write contention on the table after {numCalls} attempts.";
log.Error(ErrorCode.MembershipFailedToWriteConditional, errorString);
throw new OrleansException(errorString);
}
}
catch (Exception exc)
{
if (errorString == null)
{
errorString = $"-Silo {this.myAddress} failed to update its status to {status} in the table due to failures (socket failures or table read/write failures) after {numCalls} attempts: {exc.Message}";
log.Error(ErrorCode.MembershipFailedToWrite, errorString);
throw new OrleansException(errorString, exc);
}
throw;
}
}
// TODO move this logic in the LocalGrainDirectory
private void OnSiloStatusChange(SiloAddress updatedSilo, SiloStatus status)
{
// ignore joining events and also events on myself.
if (updatedSilo.Equals(LocalSilo))
{
return;
}
// We deactivate those activations when silo goes either of ShuttingDown/Stopping/Dead states,
// since this is what Directory is doing as well. Directory removes a silo based on all those 3 statuses,
// thus it will only deliver a "remove" notification for a given silo once to us. Therefore, we need to react the fist time we are notified.
// We may review the directory behavior in the future and treat ShuttingDown differently ("drain only") and then this code will have to change a well.
if (!status.IsTerminating())
{
return;
}
if (status == SiloStatus.Dead)
{
this.RuntimeClient.BreakOutstandingMessagesToDeadSilo(updatedSilo);
}
var activationsToShutdown = new List <IGrainContext>();
try
{
// scan all activations in activation directory and deactivate the ones that the removed silo is their primary partition owner.
lock (activations)
{
foreach (var activation in activations)
{
try
{
var activationData = activation.Value;
if (!activationData.PlacementStrategy.IsUsingGrainDirectory || grainDirectoryResolver.HasNonDefaultDirectory(activationData.GrainId.Type))
{
continue;
}
if (!updatedSilo.Equals(directory.GetPrimaryForGrain(activationData.GrainId)))
{
continue;
}
activationsToShutdown.Add(activationData);
}
catch (Exception exc)
{
logger.LogError(
(int)ErrorCode.Catalog_SiloStatusChangeNotification_Exception,
exc,
"Catalog has thrown an exception while handling removal of silo {Silo}", updatedSilo.ToStringWithHashCode());
}
}
}
logger.LogInformation(
(int)ErrorCode.Catalog_SiloStatusChangeNotification,
"Catalog is deactivating {Count} activations due to a failure of silo {Silo}, since it is a primary directory partition to these grain ids.",
activationsToShutdown.Count,
updatedSilo.ToStringWithHashCode());
}
finally
{
// outside the lock.
if (activationsToShutdown.Count > 0)
{
DeactivateActivations(activationsToShutdown).Ignore();
}
}
}
public void SiloStatusChangeNotification(SiloAddress updatedSilo, SiloStatus status)
{
// ignore joining events and also events on myself.
if (updatedSilo.Equals(LocalSilo)) return;
// We deactivate those activations when silo goes either of ShuttingDown/Stopping/Dead states,
// since this is what Directory is doing as well. Directory removes a silo based on all those 3 statuses,
// thus it will only deliver a "remove" notification for a given silo once to us. Therefore, we need to react the fist time we are notified.
// We may review the directory behaiviour in the future and treat ShuttingDown differently ("drain only") and then this code will have to change a well.
if (!status.IsTerminating()) return;
var activationsToShutdown = new List<ActivationData>();
try
{
// scan all activations in activation directory and deactivate the ones that the removed silo is their primary partition owner.
lock (activations)
{
foreach (var activation in activations)
{
try
{
var activationData = activation.Value;
if (!directory.GetPrimaryForGrain(activationData.Grain).Equals(updatedSilo)) continue;
lock (activationData)
{
// adapted from InsideGarinClient.DeactivateOnIdle().
activationData.ResetKeepAliveRequest();
activationsToShutdown.Add(activationData);
}
}
catch (Exception exc)
{
logger.Error(ErrorCode.Catalog_SiloStatusChangeNotification_Exception,
String.Format("Catalog has thrown an exception while executing SiloStatusChangeNotification of silo {0}.", updatedSilo.ToStringWithHashCode()), exc);
}
}
}
logger.Info(ErrorCode.Catalog_SiloStatusChangeNotification,
String.Format("Catalog is deactivating {0} activations due to a failure of silo {1}, since it is a primary directory partiton to these grain ids.",
activationsToShutdown.Count, updatedSilo.ToStringWithHashCode()));
}
finally
{
// outside the lock.
if (activationsToShutdown.Count > 0)
{
DeactivateActivations(activationsToShutdown).Ignore();
}
}
}
/// <summary>
/// Return true if silo is terminating or missing.
/// </summary>
public static bool IsUnavailable(this SiloStatus siloStatus)
{
return(siloStatus.IsTerminating() || siloStatus == SiloStatus.None);
}
public void SiloStatusChangeNotification(SiloAddress updatedSilo, SiloStatus status)
{
// This silo's status has changed
if (Equals(updatedSilo, MyAddress))
{
if (status == SiloStatus.Stopping || status.Equals(SiloStatus.ShuttingDown))
{
// QueueAction up the "Stop" to run on a system turn
Scheduler.QueueAction(() => Stop(true), CacheValidator.SchedulingContext).Ignore();
}
else if (status == SiloStatus.Dead)
{
// QueueAction up the "Stop" to run on a system turn
Scheduler.QueueAction(() => Stop(false), CacheValidator.SchedulingContext).Ignore();
}
}
else // Status change for some other silo
{
if (status.IsTerminating())
{
// QueueAction up the "Remove" to run on a system turn
Scheduler.QueueAction(() => RemoveServer(updatedSilo, status), CacheValidator.SchedulingContext).Ignore();
}
else if (status.Equals(SiloStatus.Active)) // do not do anything with SiloStatus.Starting -- wait until it actually becomes active
{
// QueueAction up the "Remove" to run on a system turn
Scheduler.QueueAction(() => AddServer(updatedSilo), CacheValidator.SchedulingContext).Ignore();
}
}
}
public void SiloStatusChangeNotification(SiloAddress updatedSilo, SiloStatus status)
{
if (!status.IsTerminating()) return;
SiloRuntimeStatistics ignore;
periodicStats.TryRemove(updatedSilo, out ignore);
NotifyAllStatisticsChangeEventsSubscribers(updatedSilo, null);
}
public async Task UpdateStatus(SiloStatus status)
{
if (status == SiloStatus.Joining)
{
// first, cleanup all outdated entries of myself from the table
Func <int, Task <bool> > cleanupTableEntriesTask = async counter =>
{
if (log.IsEnabled(LogLevel.Debug))
{
log.Debug("-Attempting CleanupTableEntries #{0}", counter);
}
var table = await this.membershipTableProvider.ReadAll();
log.Info(ErrorCode.MembershipReadAll_Cleanup, "-CleanupTable called on silo startup. Membership table {0}",
table.ToString());
return(await CleanupMyTableEntries(table));
};
await MembershipExecuteWithRetries(cleanupTableEntriesTask, this.clusterMembershipOptions.MaxJoinAttemptTime);
}
if (status == SiloStatus.Dead && this.membershipTableProvider is SystemTargetBasedMembershipTable)
{
this.CurrentStatus = status;
// SystemTarget-based clustering does not support transitioning to Dead locally since at this point app scheduler turns have been stopped.
return;
}
string errorString = null;
int numCalls = 0;
try
{
Func <int, Task <bool> > updateMyStatusTask = async counter =>
{
numCalls++;
if (log.IsEnabled(LogLevel.Debug))
{
log.Debug("-Going to try to TryUpdateMyStatusGlobalOnce #{0}", counter);
}
return(await TryUpdateMyStatusGlobalOnce(status)); // function to retry
};
bool ok = await MembershipExecuteWithRetries(updateMyStatusTask, this.clusterMembershipOptions.MaxJoinAttemptTime);
if (ok)
{
if (log.IsEnabled(LogLevel.Debug))
{
log.Debug("-Silo {0} Successfully updated my Status in the Membership table to {1}", myAddress, status);
}
var gossipTask = this.GossipToOthers(this.myAddress, status);
if (status.IsTerminating())
{
var timeoutTask = Task.Delay(ShutdownGossipTimeout);
var task = await Task.WhenAny(gossipTask, timeoutTask);
if (ReferenceEquals(task, timeoutTask))
{
this.log.LogWarning("Timed out while gossiping status to other silos after {Timeout}", ShutdownGossipTimeout);
}
}
}
else
{
errorString = $"-Silo {myAddress} failed to update its status to {status} in the Membership table due to write contention on the table after {numCalls} attempts.";
log.Error(ErrorCode.MembershipFailedToWriteConditional, errorString);
throw new OrleansException(errorString);
}
}
catch (Exception exc)
{
if (errorString == null)
{
errorString = $"-Silo {this.myAddress} failed to update its status to {status} in the table due to failures (socket failures or table read/write failures) after {numCalls} attempts: {exc.Message}";
log.Error(ErrorCode.MembershipFailedToWrite, errorString);
throw new OrleansException(errorString, exc);
}
throw;
}
}
public async Task UpdateStatus(SiloStatus status)
{
bool wasThrownLocally = false;
int numCalls = 0;
try
{
Func <int, Task <bool> > updateMyStatusTask = async counter =>
{
numCalls++;
if (log.IsEnabled(LogLevel.Debug))
{
log.LogDebug("Going to try to TryUpdateMyStatusGlobalOnce #{Attempt}", counter);
}
return(await TryUpdateMyStatusGlobalOnce(status)); // function to retry
};
if (status == SiloStatus.Dead && this.membershipTableProvider is SystemTargetBasedMembershipTable)
{
// SystemTarget-based membership may not be accessible at this stage, so allow for one quick attempt to update
// the status before continuing regardless of the outcome.
var updateTask = updateMyStatusTask(0);
updateTask.Ignore();
await Task.WhenAny(Task.Delay(TimeSpan.FromMilliseconds(500)), updateTask);
var gossipTask = this.GossipToOthers(this.myAddress, status);
gossipTask.Ignore();
await Task.WhenAny(Task.Delay(TimeSpan.FromMilliseconds(500)), gossipTask);
this.CurrentStatus = status;
return;
}
bool ok = await MembershipExecuteWithRetries(updateMyStatusTask, this.clusterMembershipOptions.MaxJoinAttemptTime);
if (ok)
{
if (log.IsEnabled(LogLevel.Debug))
{
log.LogDebug("Silo {SiloAddress} Successfully updated my Status in the membership table to {Status}", myAddress, status);
}
var gossipTask = this.GossipToOthers(this.myAddress, status);
gossipTask.Ignore();
var cancellation = new CancellationTokenSource();
var timeoutTask = Task.Delay(GossipTimeout, cancellation.Token);
var task = await Task.WhenAny(gossipTask, timeoutTask);
if (ReferenceEquals(task, timeoutTask))
{
if (status.IsTerminating())
{
this.log.LogWarning("Timed out while gossiping status to other silos after {Timeout}", GossipTimeout);
}
else if (this.log.IsEnabled(LogLevel.Debug))
{
this.log.LogDebug("Timed out while gossiping status to other silos after {Timeout}", GossipTimeout);
}
}
else
{
cancellation.Cancel();
}
}
else
{
wasThrownLocally = true;
log.LogError(
(int)ErrorCode.MembershipFailedToWriteConditional,
"Silo {MyAddress} failed to update its status to {Status} in the membership table due to write contention on the table after {NumCalls} attempts.",
myAddress,
status,
numCalls);
throw new OrleansException($"Silo {myAddress} failed to update its status to {status} in the membership table due to write contention on the table after {numCalls} attempts.");
}
}
catch (Exception exc) when(!wasThrownLocally)
{
log.LogError(
(int)ErrorCode.MembershipFailedToWrite,
exc,
"Silo {MyAddress} failed to update its status to {Status} in the table due to failures (socket failures or table read/write failures) after {NumCalls} attempts",
myAddress,
status,
numCalls);
throw new OrleansException($"Silo {myAddress} failed to update its status to {status} in the table due to failures (socket failures or table read/write failures) after {numCalls} attempts", exc);
}
}
