// Select a quorum of P and S that are not Down or Dropped
internal static List <StatefulServiceReplica> GetReplicasForPartialLoss(Guid operationId, List <StatefulServiceReplica> replicaList)
{
List <StatefulServiceReplica> tempReplicas = new List <StatefulServiceReplica>();
foreach (StatefulServiceReplica replica in replicaList)
{
if (FaultAnalysisServiceUtility.IsPrimaryOrSecondary(replica) && FaultAnalysisServiceUtility.IsReplicaUp(replica))
{
TestabilityTrace.TraceSource.WriteInfo(TraceType, "DEBUG {0} temp adding {1},{2},{3}", operationId, replica.Id, replica.ReplicaRole, replica.ReplicaStatus);
tempReplicas.Add(replica);
}
}
int replicasToRestartWithoutPrimary = tempReplicas.Count / 2;
StatefulServiceReplica primary = tempReplicas.Where(r => r.ReplicaRole == ReplicaRole.Primary).FirstOrDefault();
if (primary == null)
{
return(null);
}
List <StatefulServiceReplica> targetReplicas = new List <StatefulServiceReplica>(replicasToRestartWithoutPrimary + 1);
TestabilityTrace.TraceSource.WriteInfo(TraceType, "DEBUG {0} target adding primary {1},{2},{3}", operationId, primary.Id, primary.ReplicaRole, primary.ReplicaStatus);
targetReplicas.Add(primary);
tempReplicas.Remove(primary);
for (int i = 0; i < replicasToRestartWithoutPrimary; i++)
{
TestabilityTrace.TraceSource.WriteInfo(TraceType, "DEBUG {0} target adding {1},{2},{3}", operationId, tempReplicas[i].Id, tempReplicas[i].ReplicaRole, tempReplicas[i].ReplicaStatus);
targetReplicas.Add(tempReplicas[i]);
}
return(targetReplicas);
}
// Replica does not have default constructor so need a method.
private Replica CreateReplica()
{
Replica result;
if (this.random.Next() % 2 == 0)
{
result = new StatefulServiceReplica(
this.random.CreateRandom <ServiceReplicaStatus>(),
this.random.CreateRandom <HealthState>(),
ReplicaRole.IdleSecondary,
this.random.CreateRandom <Uri>().ToString(),
this.random.CreateRandom <string>(),
this.random.CreateRandom <long>(),
TimeSpan.FromSeconds(2));
}
else
{
result = new StatelessServiceInstance(
this.random.CreateRandom <ServiceReplicaStatus>(),
this.random.CreateRandom <HealthState>(),
this.random.CreateRandom <Uri>().ToString(),
this.random.CreateRandom <string>(),
this.random.CreateRandom <long>(),
TimeSpan.FromSeconds(4));
}
return(result);
}
private void KillPrimaryReplica()
{
// Kill the primary
Application application =
_fabricClient.QueryManager.GetApplicationListAsync()
.Result.Single(a => a.ApplicationTypeName == DefaultApplicationTypeName);
Service service =
_fabricClient.QueryManager.GetServiceListAsync(application.ApplicationName).Result.Single();
Partition partition =
_fabricClient.QueryManager.GetPartitionListAsync(service.ServiceName).Result.Single();
StatefulServiceReplica primaryReplica =
_fabricClient.QueryManager.GetReplicaListAsync(partition.PartitionInformation.Id)
.Result.Select(replica => replica as StatefulServiceReplica)
.Single(statefulServiceReplica => statefulServiceReplica.ReplicaRole == ReplicaRole.Primary);
LogHelper.Log("Killing the primary replica at node {0}", primaryReplica.NodeName);
ReplicaSelector replicaSelector = ReplicaSelector.ReplicaIdOf(PartitionSelector.PartitionIdOf(service.ServiceName, partition.PartitionInformation.Id), primaryReplica.Id);
_fabricClient.FaultManager.RemoveReplicaAsync(replicaSelector, CompletionMode.DoNotVerify, false);
}
public async Task <Dictionary <Partition, StatefulServiceReplica[]> > QueryPartitionAndReplicaResultAsyncStateful(CancellationToken ct)
{
var servicePartitionMap = await this.QueryLocationsAsync(ct).ConfigureAwait(false);
var allServiceReplicas =
new Dictionary <Partition, StatefulServiceReplica[]>();
foreach (var partition in servicePartitionMap)
{
List <StatefulServiceReplica> statefulReplicas = new List <StatefulServiceReplica>();
foreach (Replica replica in partition.Value)
{
StatefulServiceReplica statefulReplica = replica as StatefulServiceReplica;
ReleaseAssert.AssertIf(statefulReplica == null, "Replica {0} should be of type stateful for Partition {1}", replica.Id, partition.Key.PartitionId());
statefulReplicas.Add(statefulReplica);
}
allServiceReplicas.Add(partition.Key, statefulReplicas.ToArray());
}
return(allServiceReplicas);
}
protected override async Task ExecuteActionAsync(FabricTestContext testContext, RestartPartitionAction action, CancellationToken cancellationToken)
{
ThrowIf.Null(action.PartitionSelector, "partitionSelector");
this.helper = new TimeoutHelper(action.ActionTimeout);
// get service info so we can validate if the operation is valid
ServiceDescription result = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.ServiceManager.GetServiceDescriptionAsync(
action.PartitionSelector.ServiceName,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
if (result.Kind != ServiceDescriptionKind.Stateful && action.RestartPartitionMode == RestartPartitionMode.OnlyActiveSecondaries)
{
throw new InvalidOperationException(StringHelper.Format(StringResources.Error_InvalidServiceTypeTestability, "RestartPartitionMode.OnlyActiveSecondaries", "Stateful", action.PartitionSelector.ServiceName, "Stateless"));
}
bool hasPersistedState = false;
if (result.Kind == ServiceDescriptionKind.Stateful)
{
StatefulServiceDescription statefulDescription = result as StatefulServiceDescription;
ReleaseAssert.AssertIf(statefulDescription == null, "Stateful service description is not WinFabricStatefulServiceDescription");
hasPersistedState = statefulDescription.HasPersistedState;
}
// now actually select a partition
var getPartitionStateAction = new GetSelectedPartitionStateAction(action.PartitionSelector)
{
RequestTimeout = action.RequestTimeout,
ActionTimeout = helper.GetRemainingTime()
};
await testContext.ActionExecutor.RunAsync(getPartitionStateAction, cancellationToken);
Guid partitionId = getPartitionStateAction.Result.PartitionId;
// get replicas for target
ServiceReplicaList replicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetReplicaListAsync(
partitionId,
0,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
// get replicas for fm in order to get the primary
ServiceReplicaList fmReplicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetReplicaListAsync(
Constants.FmPartitionId,
0,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
string fmPrimaryNodeName = string.Empty;
var readyFMReplicas = fmReplicasResult.Where(r => r.ReplicaStatus == ServiceReplicaStatus.Ready).ToArray();
foreach (var replica in readyFMReplicas)
{
StatefulServiceReplica statefulReplica = replica as StatefulServiceReplica;
ReleaseAssert.AssertIf(statefulReplica == null, "FM Replica is not a stateful replica");
if (statefulReplica.ReplicaRole == ReplicaRole.Primary)
{
fmPrimaryNodeName = replica.NodeName;
}
}
if (string.IsNullOrEmpty(fmPrimaryNodeName))
{
throw new FabricException(StringHelper.Format(StringResources.Error_PartitionPrimaryNotReady, "FailoverManager"), FabricErrorCode.NotReady);
}
////------------------------------------------------------
// target ut at the fm primary only
UnreliableTransportBehavior behavior = new UnreliableTransportBehavior("*", "DoReconfiguration");
behavior.AddFilterForPartitionId(partitionId);
string behaviorName = "BlockDoReconfiguration";
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.TestManager.AddUnreliableTransportBehaviorAsync(
fmPrimaryNodeName,
behaviorName,
behavior,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
// TODO: Wait for some time so that the unreliable transport behavior can be read from the files.
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successfully applied
await Task.Delay(TimeSpan.FromSeconds(5.0), cancellationToken).ConfigureAwait(false);
bool triedToRemovedBehavior = false;
// inspect the actual replicas to restart, only operate on stable ones
try
{
var stableReplicasToRestart = replicasResult.Where(r => r.ReplicaStatus == ServiceReplicaStatus.Ready).ToArray();
foreach (var replica in stableReplicasToRestart)
{
var currentReplica = replica;
if (action.RestartPartitionMode == RestartPartitionMode.OnlyActiveSecondaries)
{
StatefulServiceReplica statefulReplica = currentReplica as StatefulServiceReplica;
ReleaseAssert.AssertIf(statefulReplica == null, "Stateful service replica is not StatefulServiceReplica");
if (statefulReplica.ReplicaRole == ReplicaRole.Primary)
{
continue;
}
}
if (hasPersistedState)
{
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.FaultManager.RestartReplicaAsync(
currentReplica.NodeName,
partitionId,
currentReplica.Id,
CompletionMode.DoNotVerify,
action.RequestTimeout.TotalSeconds,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
}
else
{
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.FaultManager.RemoveReplicaAsync(
currentReplica.NodeName,
partitionId,
currentReplica.Id,
CompletionMode.DoNotVerify,
false, /*force remove*/
action.RequestTimeout.TotalSeconds,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
}
}
triedToRemovedBehavior = true;
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.TestManager.RemoveUnreliableTransportBehaviorAsync(
fmPrimaryNodeName,
behaviorName,
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.RemoveUnreliableTransportBehaviorErrors.Value,
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
// TODO: Wait for some time so that the unreliable transport behavior can be read from the files.
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successfully applied
await Task.Delay(TimeSpan.FromSeconds(5.0)).ConfigureAwait(false);
}
finally
{
// TODO: Provide a way to clear all behaviors just in case.
if (!triedToRemovedBehavior)
{
ActionTraceSource.WriteWarning(TraceType, "Exception after adding behavior to block messages. Removing behavior synchronously");
FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.TestManager.RemoveUnreliableTransportBehaviorAsync(
fmPrimaryNodeName,
behaviorName,
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.RemoveUnreliableTransportBehaviorErrors.Value,
this.helper.GetRemainingTime(),
cancellationToken).GetAwaiter().GetResult();
// TODO: Wait for some time so that the unreliable transport behavior can be read from the files.
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successfully applied
Task.Delay(TimeSpan.FromSeconds(5.0)).GetAwaiter().GetResult();
}
}
// -- note there's no explict validation
// action result
action.Result = new RestartPartitionResult(getPartitionStateAction.Result);
ResultTraceString = StringHelper.Format("RestartPartitionAction succeeded for {0} with RestartPartitionMode = {1}", partitionId, action.RestartPartitionMode);
}
protected override async Task ExecuteActionAsync(FabricTestContext testContext, MoveSecondaryAction action, CancellationToken cancellationToken)
{
ThrowIf.Null(action.PartitionSelector, "PartitionSelector");
this.helper = new TimeoutHelper(action.ActionTimeout);
string newSecondaryNode = action.NewSecondaryNodeName;
string currentSecondaryNode = action.CurrentSecondaryNodeName;
var getPartitionStateAction = new GetSelectedPartitionStateAction(action.PartitionSelector)
{
RequestTimeout = action.RequestTimeout,
ActionTimeout = this.helper.GetRemainingTime()
};
await testContext.ActionExecutor.RunAsync(getPartitionStateAction, cancellationToken).ConfigureAwait(false);
Guid partitionId = getPartitionStateAction.Result.PartitionId;
if (!action.IgnoreConstraints)
{
// get current primary replica node name.
ServiceReplicaList replicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetReplicaListAsync(
partitionId,
0,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
string currentPrimaryNodeInfo = string.Empty;
List <string> currentSecReplicaNodes = new List <string>();
foreach (var replica in replicasResult)
{
StatefulServiceReplica statefulReplica = replica as StatefulServiceReplica;
if (statefulReplica == null)
{
throw new InvalidOperationException(StringHelper.Format(StringResources.Error_InvalidServiceTypeTestability, "MoveSecondary", "Stateful", action.PartitionSelector.ServiceName, "Stateless"));
}
if (statefulReplica.ReplicaRole == ReplicaRole.Primary)
{
currentPrimaryNodeInfo = statefulReplica.NodeName;
if (!string.IsNullOrEmpty(newSecondaryNode) && newSecondaryNode == statefulReplica.NodeName)
{
throw new FabricException(
StringHelper.Format(StringResources.Error_InvalidNodeNameProvided, newSecondaryNode, "MoveSecondary", "Primary exists on node"),
FabricErrorCode.AlreadyPrimaryReplica);
}
}
else if (statefulReplica.ReplicaRole == ReplicaRole.ActiveSecondary)
{
currentSecReplicaNodes.Add(statefulReplica.NodeName);
if (!string.IsNullOrEmpty(newSecondaryNode) && newSecondaryNode == statefulReplica.NodeName)
{
throw new FabricException(
StringHelper.Format(StringResources.Error_InvalidNodeNameProvided, newSecondaryNode, "MoveSecondary", "Secondary exists on node"),
FabricErrorCode.AlreadySecondaryReplica);
}
}
}
if (currentSecReplicaNodes.Count == 0)
{
throw new InvalidOperationException(StringResources.Error_NoSecondariesInReplicaSet);
}
if (string.IsNullOrEmpty(currentSecondaryNode))
{
int num = testContext.Random.Next(currentSecReplicaNodes.Count);
currentSecondaryNode = currentSecReplicaNodes.ElementAt(num);
}
if (!currentSecReplicaNodes.Contains(currentSecondaryNode))
{
throw new FabricException(
StringHelper.Format(StringResources.Error_InvalidNodeNameProvided, newSecondaryNode, "MoveSecondary", "Current node does not have a secondary replica"),
FabricErrorCode.InvalidReplicaStateForReplicaOperation);
}
}
ReleaseAssert.AssertIf(string.IsNullOrEmpty(currentSecondaryNode), "Current node name cannot be null or empty.");
ReleaseAssert.AssertIf(newSecondaryNode == currentSecondaryNode, "Current and New node names are same.");
ActionTraceSource.WriteInfo(TraceSource, "Calling move secondary with current node {0}, new node {1}, partition {2}", currentSecondaryNode, string.IsNullOrEmpty(newSecondaryNode) ? "Random" : newSecondaryNode, partitionId);
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.FaultManager.MoveSecondaryUsingNodeNameAsync(
currentSecondaryNode,
newSecondaryNode,
getPartitionStateAction.Result.ServiceName,
partitionId,
action.IgnoreConstraints,
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.MoveSecondaryFabricErrors.Value,
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
action.Result = new MoveSecondaryResult(currentSecondaryNode, newSecondaryNode, getPartitionStateAction.Result);
this.ResultTraceString = StringHelper.Format(
"MoveSecondaryAction succeeded for moving Primary for {0} from {1} to {2}.",
partitionId,
currentSecondaryNode,
newSecondaryNode);
}
public override async Task <ActionStateBase> RunAsync(CancellationToken cancellationToken, ServiceInternalFaultInfo serviceInternalFaultInfo)
{
InvokeDataLossState state = Convert(this.State);
ServiceDescription result = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.ServiceManager.GetServiceDescriptionAsync(
this.partitionSelector.ServiceName,
this.RequestTimeout,
cancellationToken),
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
if (result.Kind != ServiceDescriptionKind.Stateful)
{
// The message in the first arg is only for debugging, it is not returned to the user.
throw new FabricInvalidForStatelessServicesException("FabricInvalidForStatelessServicesException", FabricErrorCode.InvalidForStatelessServices);
}
int targetReplicaSetSize = (result as StatefulServiceDescription).TargetReplicaSetSize;
SelectedPartition targetPartition = await FaultAnalysisServiceUtility.GetSelectedPartitionStateAsync(
this.FabricClient,
this.partitionSelector,
this.RequestTimeout,
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
Guid partitionId = targetPartition.PartitionId;
long preDataLossNumber = 0;
ServicePartitionList partitionsResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.QueryManager.GetPartitionListAsync(
this.partitionSelector.ServiceName,
null,
this.RequestTimeout,
cancellationToken),
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
bool partitionFound = false;
foreach (StatefulServicePartition partition in partitionsResult)
{
if (partition.PartitionInformation.Id == partitionId)
{
preDataLossNumber = partition.PrimaryEpoch.DataLossNumber;
partitionFound = true;
break;
}
}
if (!partitionFound)
{
throw new FabricException(StringHelper.Format(StringResources.Error_PartitionNotFound), FabricErrorCode.PartitionNotFound);
}
ServiceReplicaList failoverManagerReplicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.QueryManager.GetReplicaListAsync(
FASConstants.FmPartitionId,
0,
this.RequestTimeout,
cancellationToken),
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
string failoverManagerPrimaryNodeName = string.Empty;
var readyFMReplicas = failoverManagerReplicasResult.Where(r => r.ReplicaStatus == ServiceReplicaStatus.Ready).ToArray();
foreach (var replica in readyFMReplicas)
{
StatefulServiceReplica statefulReplica = replica as StatefulServiceReplica;
ReleaseAssert.AssertIf(statefulReplica == null, "FM Replica is not a stateful replica");
if (statefulReplica.ReplicaRole == ReplicaRole.Primary)
{
failoverManagerPrimaryNodeName = replica.NodeName;
}
}
if (string.IsNullOrEmpty(failoverManagerPrimaryNodeName))
{
throw new FabricException(StringHelper.Format(StringResources.Error_PartitionPrimaryNotReady, "FailoverManager"), FabricErrorCode.NotReady);
}
TestabilityTrace.TraceSource.WriteInfo(StepBase.TraceType, "{0} - FM primary location={1}", this.State.OperationId, failoverManagerPrimaryNodeName);
string behaviorName = "BlockDoReconfiguration_" + this.State.OperationId;
List <Tuple <string, string> > unreliableTransportInfo = new List <Tuple <string, string> >();
unreliableTransportInfo.Add(new Tuple <string, string>(failoverManagerPrimaryNodeName, behaviorName));
state.StateProgress.Push(StepStateNames.PerformingActions);
state.Info.DataLossNumber = preDataLossNumber;
state.Info.NodeName = failoverManagerPrimaryNodeName;
state.Info.PartitionId = partitionId;
state.Info.UnreliableTransportInfo = unreliableTransportInfo;
state.Info.TargetReplicaSetSize = targetReplicaSetSize;
return(state);
}
public override async Task <ActionStateBase> RunAsync(CancellationToken cancellationToken, ServiceInternalFaultInfo serviceInternalFaultInfo)
{
InvokeDataLossState state = Convert(this.State);
PartitionSelector partitionSelector = state.Info.PartitionSelector;
DataLossMode dataLossMode = state.Info.DataLossMode;
long preDataLossNumber = state.Info.DataLossNumber;
string failoverManagerPrimaryNodeName = state.Info.NodeName;
Guid partitionId = state.Info.PartitionId;
string behaviorName = state.Info.UnreliableTransportInfo.First().Item2;
int targetReplicaSetSize = state.Info.TargetReplicaSetSize;
TestabilityTrace.TraceSource.WriteInfo(StepBase.TraceType, "{0} - applying UT, partitionId={1}", this.State.OperationId, partitionId);
System.Fabric.Common.UnreliableTransportBehavior behavior = new System.Fabric.Common.UnreliableTransportBehavior("*", "DoReconfiguration");
behavior.AddFilterForPartitionId(partitionId);
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.TestManager.AddUnreliableTransportBehaviorAsync(
failoverManagerPrimaryNodeName,
behaviorName,
behavior,
this.RequestTimeout,
cancellationToken),
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
// TODO: Wait for some time so that the unreliable transport behavior can be read from the files.
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successfully applied
await Task.Delay(TimeSpan.FromSeconds(5.0), cancellationToken).ConfigureAwait(false);
ServiceReplicaList replicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.QueryManager.GetReplicaListAsync(
partitionId,
0,
this.RequestTimeout,
cancellationToken),
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
List <StatefulServiceReplica> replicaList = new List <StatefulServiceReplica>();
foreach (var replica in replicasResult)
{
StatefulServiceReplica statefulReplica = replica as StatefulServiceReplica;
ReleaseAssert.AssertIf(statefulReplica == null, "Service Replica is not of stateful type even though service is stateful");
replicaList.Add(statefulReplica);
}
// Select target replicas based on the DataLosMode
List <StatefulServiceReplica> targets = null;
if (dataLossMode == DataLossMode.FullDataLoss)
{
targets = GetReplicasForFullDataLoss(replicaList);
}
else if (dataLossMode == DataLossMode.PartialDataLoss)
{
targets = FaultAnalysisServiceUtility.GetReplicasForPartialLoss(state.OperationId, replicaList);
}
else
{
throw FaultAnalysisServiceUtility.CreateException(StepBase.TraceType, Interop.NativeTypes.FABRIC_ERROR_CODE.E_INVALIDARG, Strings.StringResources.Error_UnsupportedDataLossMode);
}
if (targets == null)
{
// This will cause the command to rollback and retry
throw new FabricTransientException("The operation could not be performed, please retry", FabricErrorCode.NotReady);
}
foreach (var replica in targets)
{
TestabilityTrace.TraceSource.WriteInfo(
StepBase.TraceType,
"{0} - Removing replica {1} in partition {2} with role {3} and status {4} to induce data loss",
this.State.OperationId,
replica.Id,
partitionId,
replica.ReplicaRole,
replica.ReplicaStatus);
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.ServiceManager.RemoveReplicaAsync(
replica.NodeName,
partitionId,
replica.Id,
this.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.RemoveReplicaErrors.Value,
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
}
ActionTest.PerformInternalServiceFaultIfRequested(this.State.OperationId, serviceInternalFaultInfo, this.State, cancellationToken, true);
await this.WaitForAllTargetReplicasToGetDroppedAsync(partitionId, targets, cancellationToken).ConfigureAwait(false);
await RemoveUnreliableTransportAsync(this.FabricClient, failoverManagerPrimaryNodeName, behaviorName, this.RequestTimeout, this.OperationTimeout, cancellationToken).ConfigureAwait(false);
bool dataLossWasSuccessful = false;
TimeoutHelper timeoutHelper = new TimeoutHelper(TimeSpan.FromSeconds(30));
do
{
ServicePartitionList partitionsResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.QueryManager.GetPartitionListAsync(
this.partitionSelector.ServiceName,
null,
this.RequestTimeout,
cancellationToken),
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
bool partitionFound = false;
long postDataLossNumber = 0;
foreach (StatefulServicePartition partition in partitionsResult)
{
if (partition.PartitionInformation.Id == partitionId)
{
postDataLossNumber = partition.PrimaryEpoch.DataLossNumber;
partitionFound = true;
break;
}
}
if (!partitionFound)
{
throw new FabricException(StringHelper.Format(StringResources.Error_PartitionNotFound), FabricErrorCode.PartitionNotFound);
}
TestabilityTrace.TraceSource.WriteInfo(
StepBase.TraceType,
"{0} - Checking data loss numbers for partition {1} with remaining time {2}. Current numbers {3}:{4}",
this.State.OperationId,
partitionId,
timeoutHelper.GetRemainingTime(),
preDataLossNumber,
postDataLossNumber);
if (postDataLossNumber != preDataLossNumber)
{
dataLossWasSuccessful = true;
break;
}
await System.Fabric.Common.AsyncWaiter.WaitAsync(TimeSpan.FromSeconds(this.dataLossCheckPollIntervalInSeconds), cancellationToken).ConfigureAwait(false);
}while (timeoutHelper.GetRemainingTime() > TimeSpan.Zero);
if (!dataLossWasSuccessful)
{
// This is only viewable internally for debug. This will cause a retry of the whole flow.
string error = string.Format(
CultureInfo.InvariantCulture,
"{0} - Service could not induce data loss for service '{1}' partition '{2}' in '{3}' Please retry",
this.State.OperationId,
partitionSelector.ServiceName,
partitionId,
this.dataLossCheckWaitDurationInSeconds);
TestabilityTrace.TraceSource.WriteWarning(StepBase.TraceType, error);
throw new FabricTransientException("The operation could not be performed, please retry", FabricErrorCode.NotReady);
}
state.StateProgress.Push(StepStateNames.CompletedSuccessfully);
return(state);
}
protected override async Task ExecuteActionAsync(FabricTestContext testContext, InvokeDataLossAction action, CancellationToken cancellationToken)
{
ThrowIf.Null(action.PartitionSelector, "PartitionSelector");
var helper = new TimeoutHelper(action.ActionTimeout);
ServiceDescription result = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.ServiceManager.GetServiceDescriptionAsync(
action.PartitionSelector.ServiceName,
action.RequestTimeout,
cancellationToken),
helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
if (result.Kind != ServiceDescriptionKind.Stateful)
{
throw new InvalidOperationException(StringHelper.Format(StringResources.Error_InvalidServiceTypeTestability, "DataLoss", "Stateful", action.PartitionSelector.ServiceName, "Stateless"));
}
var getPartitionStateAction = new GetSelectedPartitionStateAction(action.PartitionSelector)
{
RequestTimeout = action.RequestTimeout,
ActionTimeout = helper.GetRemainingTime()
};
await testContext.ActionExecutor.RunAsync(getPartitionStateAction, cancellationToken).ConfigureAwait(false);
Guid partitionId = getPartitionStateAction.Result.PartitionId;
long preDataLossNumber = 0;
ServicePartitionList partitionsResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetPartitionListAsync(
action.PartitionSelector.ServiceName,
null,
action.RequestTimeout,
cancellationToken),
helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
bool partitionFound = false;
foreach (StatefulServicePartition partition in partitionsResult)
{
if (partition.PartitionInformation.Id == partitionId)
{
preDataLossNumber = partition.PrimaryEpoch.DataLossNumber;
partitionFound = true;
break;
}
}
if (!partitionFound)
{
throw new FabricException(StringHelper.Format(StringResources.Error_PartitionNotFound), FabricErrorCode.PartitionNotFound);
}
long postDataLossNumber = preDataLossNumber;
do
{
ActionTraceSource.WriteInfo(
TraceType,
"InvokeDataLossAction action pending time:{0}",
helper.GetRemainingTime());
if (helper.GetRemainingTime() <= TimeSpan.Zero)
{
throw new TimeoutException(StringHelper.Format(StringResources.Error_TestabilityActionTimeout, "InvokeDataLoss", partitionId));
}
ServiceReplicaList replicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetReplicaListAsync(
partitionId,
0,
action.RequestTimeout,
cancellationToken),
helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
ServiceReplicaList fmReplicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetReplicaListAsync(
Constants.FmPartitionId,
0,
action.RequestTimeout,
cancellationToken),
helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
string fmPrimaryNodeName = string.Empty;
var readyFMReplicas = fmReplicasResult.Where(r => r.ReplicaStatus == ServiceReplicaStatus.Ready).ToArray();
foreach (var replica in readyFMReplicas)
{
StatefulServiceReplica statefulReplica = replica as StatefulServiceReplica;
ReleaseAssert.AssertIf(statefulReplica == null, "FM Replica is not a stateful replica");
if (statefulReplica.ReplicaRole == ReplicaRole.Primary)
{
fmPrimaryNodeName = replica.NodeName;
}
}
if (string.IsNullOrEmpty(fmPrimaryNodeName))
{
throw new FabricException(StringHelper.Format(StringResources.Error_PartitionPrimaryNotReady, "FailoverManager"), FabricErrorCode.NotReady);
}
UnreliableTransportBehavior behavior = new UnreliableTransportBehavior("*", "DoReconfiguration");
behavior.AddFilterForPartitionId(partitionId);
string behaviorName = "BlockDoReconfiguration";
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.TestManager.AddUnreliableTransportBehaviorAsync(
fmPrimaryNodeName,
behaviorName,
behavior,
action.RequestTimeout,
cancellationToken),
helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
// TODO: Wait for some time so that the unreliable transport behavior can be read from the files.
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successfully applied
await Task.Delay(TimeSpan.FromSeconds(5.0), cancellationToken).ConfigureAwait(false);
bool triedToRemovedBehavior = false;
try
{
var stableReplicasToRemove = replicasResult.Where(r => r.ReplicaStatus == ServiceReplicaStatus.Ready).ToArray();
ActionTraceSource.WriteInfo(TraceType, "Total number of replicas found {0}:{1}", replicasResult.Count(), stableReplicasToRemove.Count());
int replicasToRestartWithoutPrimary =
action.DataLossMode == DataLossMode.FullDataLoss
? stableReplicasToRemove.Length - 1
: (stableReplicasToRemove.Length + 1) / 2 - 1;
foreach (var replica in stableReplicasToRemove)
{
var currentReplica = replica;
StatefulServiceReplica statefulReplica = currentReplica as StatefulServiceReplica;
ReleaseAssert.AssertIf(statefulReplica == null, "Service Replica is not of stateful type even though service is stateful");
ActionTraceSource.WriteInfo(
TraceType,
"Inspecting replica {0}:{1} with role {2} and status {3} to induce data loss",
currentReplica.Id,
partitionId,
statefulReplica.ReplicaRole,
statefulReplica.ReplicaStatus);
if (statefulReplica.ReplicaRole != ReplicaRole.Primary)
{
replicasToRestartWithoutPrimary--;
}
if (replicasToRestartWithoutPrimary >= 0 || statefulReplica.ReplicaRole == ReplicaRole.Primary)
{
ActionTraceSource.WriteInfo(TraceType, "Removing replica {0}:{1} to induce data loss", currentReplica.Id, partitionId);
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.FaultManager.RemoveReplicaAsync(
currentReplica.NodeName,
partitionId,
currentReplica.Id,
CompletionMode.DoNotVerify,
false, /*force remove*/
action.RequestTimeout.TotalSeconds,
cancellationToken),
helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
}
}
triedToRemovedBehavior = true;
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.TestManager.RemoveUnreliableTransportBehaviorAsync(
fmPrimaryNodeName,
behaviorName,
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.RemoveUnreliableTransportBehaviorErrors.Value,
helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
// TODO: Wait for some time so that the removal of this unreliable transport behavior can be read from the files.
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successully applied
await Task.Delay(TimeSpan.FromSeconds(5.0), cancellationToken).ConfigureAwait(false);
// retry check for whether data loss number has increased 5 times else do the entire process again
const int maxRetryCount = 5;
int retryCount = 0;
do
{
partitionsResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetPartitionListAsync(
action.PartitionSelector.ServiceName,
null,
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.GetPartitionListFabricErrors.Value,
helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
partitionFound = false;
foreach (StatefulServicePartition partition in partitionsResult)
{
if (partition.PartitionInformation.Id == partitionId)
{
postDataLossNumber = partition.PrimaryEpoch.DataLossNumber;
partitionFound = true;
break;
}
}
if (!partitionFound)
{
throw new FabricException(StringHelper.Format(StringResources.Error_PartitionNotFound), FabricErrorCode.PartitionNotFound);
}
ActionTraceSource.WriteInfo(
TraceType,
"Checking data loss numbers for partition {0} with retryCount {1}. Current numbers {2}:{3}",
partitionId,
retryCount,
preDataLossNumber,
postDataLossNumber);
if (postDataLossNumber != preDataLossNumber)
{
break;
}
await AsyncWaiter.WaitAsync(TimeSpan.FromSeconds(5), cancellationToken);
++retryCount;
} while (retryCount < maxRetryCount);
}
finally
{
if (!triedToRemovedBehavior)
{
ActionTraceSource.WriteWarning(TraceType, "Exception after adding behavior to block messages. Removing behavior synchronously");
FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.TestManager.RemoveUnreliableTransportBehaviorAsync(
fmPrimaryNodeName,
behaviorName,
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.RemoveUnreliableTransportBehaviorErrors.Value,
helper.GetRemainingTime(),
cancellationToken).GetAwaiter().GetResult();
// TODO: Wait for some time so that the removal of this unreliable transport behavior can be read from the files.
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successully applied
Task.Delay(TimeSpan.FromSeconds(5.0), cancellationToken).GetAwaiter().GetResult();
}
}
}while (postDataLossNumber == preDataLossNumber);
ActionTraceSource.WriteInfo(
TraceType,
"InvokeDataLossAction action completed postDataLossNumber:{0}, preDataLossNumber:{1}",
postDataLossNumber, preDataLossNumber);
action.Result = new InvokeDataLossResult(getPartitionStateAction.Result);
this.ResultTraceString = StringHelper.Format("InvokeDataLossAction succeeded for {0} with DatalossMode = {1}", partitionId, action.DataLossMode);
}
internal static bool IsPrimaryOrSecondary(StatefulServiceReplica replica)
{
return(replica.ReplicaRole == ReplicaRole.Primary || replica.ReplicaRole == ReplicaRole.ActiveSecondary);
}
public override async Task <ActionStateBase> RunAsync(CancellationToken cancellationToken, ServiceInternalFaultInfo serviceInternalFaultInfo)
{
RestartPartitionState state = Convert(this.State);
Guid partitionId = state.Info.PartitionId;
bool hasPersistedState = state.Info.HasPersistedState;
string failoverManagerPrimaryNodeName = state.Info.NodeName;
string behaviorName = state.Info.UnreliableTransportInfo.First().Item2;
System.Fabric.Common.UnreliableTransportBehavior behavior = new System.Fabric.Common.UnreliableTransportBehavior("*", "DoReconfiguration");
behavior.AddFilterForPartitionId(partitionId);
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.TestManager.AddUnreliableTransportBehaviorAsync(
failoverManagerPrimaryNodeName,
behaviorName,
behavior,
this.RequestTimeout,
cancellationToken),
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successfully applied
await Task.Delay(TimeSpan.FromSeconds(5.0), cancellationToken).ConfigureAwait(false);
TestabilityTrace.TraceSource.WriteInfo(StepBase.TraceType, "{0} - applied UT on partitionId {1}, node={2}", this.State.OperationId, partitionId, failoverManagerPrimaryNodeName);
ServiceReplicaList replicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.QueryManager.GetReplicaListAsync(
partitionId,
0,
this.RequestTimeout,
cancellationToken),
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
var stableReplicasToRestart = replicasResult.Where(r => r.ReplicaStatus == ServiceReplicaStatus.Ready).ToArray();
foreach (var replica in stableReplicasToRestart)
{
if (this.restartPartitionMode == RestartPartitionMode.OnlyActiveSecondaries)
{
StatefulServiceReplica statefulReplica = replica as StatefulServiceReplica;
ReleaseAssert.AssertIf(statefulReplica == null, "Stateful service replica is not StatefulServiceReplica");
if (statefulReplica.ReplicaRole == ReplicaRole.Primary)
{
continue;
}
}
TestabilityTrace.TraceSource.WriteInfo(StepBase.TraceType, "{0} - restarting replica partition={1}, node={2}, replica id={3}", this.State.OperationId, partitionId, replica.NodeName, replica.Id);
if (hasPersistedState)
{
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.ServiceManager.RestartReplicaAsync(
replica.NodeName,
partitionId,
replica.Id,
this.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.RestartReplicaErrors.Value,
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
}
else
{
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.ServiceManager.RemoveReplicaAsync(
replica.NodeName,
partitionId,
replica.Id,
this.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.RemoveReplicaErrors.Value,
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
}
}
ActionTest.PerformInternalServiceFaultIfRequested(this.State.OperationId, serviceInternalFaultInfo, this.State, cancellationToken, true);
await RemoveUnreliableTransportAsync(this.State, this.FabricClient, this.RequestTimeout, this.OperationTimeout, cancellationToken);
state.StateProgress.Push(StepStateNames.CompletedSuccessfully);
return(state);
}
public void ServiceReplicaSerializationTest()
{
StatefulServiceReplica svcReplica = new StatefulServiceReplica(ServiceReplicaStatus.Ready, HealthState.Ok, ReplicaRole.ActiveSecondary, "fabric:/testsvc/testreplica", "nodeA", 890, TimeSpan.FromSeconds(5345));
TestUsingSerializer(this.Serializer, svcReplica);
}
protected override async Task ExecuteActionAsync(FabricTestContext testContext, MovePrimaryAction action, CancellationToken cancellationToken)
{
ThrowIf.Null(action.PartitionSelector, "PartitionSelector");
this.helper = new TimeoutHelper(action.ActionTimeout);
string newPrimaryNodeName = action.NodeName;
var getPartitionStateAction = new GetSelectedPartitionStateAction(action.PartitionSelector)
{
RequestTimeout = action.RequestTimeout,
ActionTimeout = this.helper.GetRemainingTime()
};
await testContext.ActionExecutor.RunAsync(getPartitionStateAction, cancellationToken);
Guid partitionId = getPartitionStateAction.Result.PartitionId;
if (!action.IgnoreConstraints)
{
// select random node where replica's primary not present
var nodesInfo = await testContext.FabricCluster.GetLatestNodeInfoAsync(action.RequestTimeout, this.helper.GetRemainingTime(), cancellationToken);
if ((nodesInfo == null || nodesInfo.Count() == 0))
{
throw new InvalidOperationException(StringHelper.Format(StringResources.Error_NotEnoughNodesForTestabilityAction, "MovePrimary"));
}
ServiceReplicaList replicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetReplicaListAsync(
partitionId,
0,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
NodeInfo currentPrimaryNodeInfo = null;
string currentPrimaryNodeName = string.Empty;
foreach (var replica in replicasResult)
{
StatefulServiceReplica statefulReplica = replica as StatefulServiceReplica;
if (statefulReplica == null)
{
throw new InvalidOperationException(StringHelper.Format(StringResources.Error_InvalidServiceTypeTestability, "MovePrimary", "Stateful", action.PartitionSelector.ServiceName, "Stateless"));
}
if (statefulReplica.ReplicaRole == ReplicaRole.Primary)
{
currentPrimaryNodeInfo = nodesInfo.FirstOrDefault(n => n.NodeName == statefulReplica.NodeName);
if (!string.IsNullOrEmpty(newPrimaryNodeName) && newPrimaryNodeName == statefulReplica.NodeName)
{
throw new FabricException(
StringHelper.Format(StringResources.Error_InvalidNodeNameProvided, newPrimaryNodeName, "MovePrimary", "Primary already exists on node"),
FabricErrorCode.AlreadyPrimaryReplica);
}
break;
}
}
if (currentPrimaryNodeInfo == null)
{
throw new FabricException(StringHelper.Format(StringResources.Error_PartitionPrimaryNotReady, action.PartitionSelector + ":" + partitionId), FabricErrorCode.NotReady);
}
currentPrimaryNodeName = currentPrimaryNodeInfo.NodeName;
if (newPrimaryNodeName == currentPrimaryNodeName)
{
throw new FabricException(
StringHelper.Format(StringResources.Error_InvalidNodeNameProvided, newPrimaryNodeName, "MovePrimary", "Primary already exists on node"),
FabricErrorCode.AlreadyPrimaryReplica);
}
}
ActionTraceSource.WriteInfo(TraceSource, "Calling move primary with node {0}, partition {1}", string.IsNullOrEmpty(newPrimaryNodeName) ? "Random" : newPrimaryNodeName, partitionId);
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.FaultManager.MovePrimaryUsingNodeNameAsync(
newPrimaryNodeName,
getPartitionStateAction.Result.ServiceName,
partitionId,
action.IgnoreConstraints,
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.MovePrimaryFabricErrors.Value,
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
action.Result = new MovePrimaryResult(newPrimaryNodeName, getPartitionStateAction.Result);
ResultTraceString = StringHelper.Format("MovePrimaryAction succeeded for moving Primary for {0} to node {1}.", partitionId, newPrimaryNodeName);
}
internal static bool IsReplicaUp(StatefulServiceReplica replica)
{
return(replica.ReplicaStatus == ServiceReplicaStatus.InBuild ||
replica.ReplicaStatus == ServiceReplicaStatus.Ready ||
replica.ReplicaStatus == ServiceReplicaStatus.Standby);
}
protected override async Task ExecuteActionAsync(FabricTestContext testContext, InvokeQuorumLossAction action, CancellationToken cancellationToken)
{
ThrowIf.Null(action.PartitionSelector, "PartitionSelector");
var helper = new TimeoutHelper(action.ActionTimeout);
// get info about the service so we can check type and trss
ServiceDescription result = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.ServiceManager.GetServiceDescriptionAsync(
action.PartitionSelector.ServiceName,
action.RequestTimeout,
cancellationToken),
helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
if (result.Kind != ServiceDescriptionKind.Stateful)
{
throw new InvalidOperationException(StringHelper.Format(StringResources.Error_InvalidServiceTypeTestability, "QuorumLoss", "Stateful", action.PartitionSelector.ServiceName, "Stateless"));
}
StatefulServiceDescription statefulServiceDescription = result as StatefulServiceDescription;
ReleaseAssert.AssertIf(statefulServiceDescription == null, "Service is not a stateful service");
if (!statefulServiceDescription.HasPersistedState)
{
throw new InvalidOperationException(StringHelper.Format(StringResources.Error_InvalidServiceTypeTestability, "QuorumLoss", "Stateful Persistent", action.PartitionSelector.ServiceName, "Stateful In-Memory Only"));
}
// figure out /which/ partition to select
var getPartitionStateAction = new GetSelectedPartitionStateAction(action.PartitionSelector)
{
RequestTimeout = action.RequestTimeout,
ActionTimeout = helper.GetRemainingTime()
};
await testContext.ActionExecutor.RunAsync(getPartitionStateAction, cancellationToken);
Guid partitionId = getPartitionStateAction.Result.PartitionId;
// get data about replicas in that partition
ServiceReplicaList replicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetReplicaListAsync(
partitionId,
0,
action.RequestTimeout,
cancellationToken),
helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
var removeUTRequestList = new List <Tuple <string, string> >();
Dictionary <Tuple <string, string>, Task> removeUTTaskDictionary = new Dictionary <Tuple <string, string>, Task>();
try
{
var stableReplicas = replicasResult.Where(r => r.ReplicaStatus == ServiceReplicaStatus.Ready).ToArray();
var stableReplicasToRemove = new List <StatefulServiceReplica>();
long replicasToRestartWithoutPrimary =
action.QuorumLossMode == QuorumLossMode.AllReplicas
? stableReplicas.Length - 1
: FabricCluster.GetWriteQuorumSize(replicasResult.Count);
foreach (var replica in stableReplicas)
{
StatefulServiceReplica statefulReplica = replica as StatefulServiceReplica;
ReleaseAssert.AssertIf(statefulReplica == null, "Service Replica is not of stateful type even though service is stateful");
if (statefulReplica.ReplicaRole != ReplicaRole.Primary)
{
replicasToRestartWithoutPrimary--;
}
if (replicasToRestartWithoutPrimary >= 0 || statefulReplica.ReplicaRole == ReplicaRole.Primary)
{
stableReplicasToRemove.Add(statefulReplica);
}
}
// for selected replicas, block reopen so that when we restart the replica (NOT remove the replica) it doesn't come up
var utTaskList = new List <Task>();
foreach (var statefulReplica in stableReplicasToRemove)
{
string nodeName = statefulReplica.NodeName;
UnreliableTransportBehavior behavior = new UnreliableTransportBehavior("*", "StatefulServiceReopen");
behavior.AddFilterForPartitionId(partitionId);
string behaviorName = "BlockStatefulServiceReopen_" + nodeName;
removeUTRequestList.Add(new Tuple <string, string>(nodeName, behaviorName));
utTaskList.Add(
FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() =>
testContext.FabricClient.TestManager.AddUnreliableTransportBehaviorAsync(
nodeName,
behaviorName,
behavior,
action.RequestTimeout,
cancellationToken),
helper.GetRemainingTime(),
cancellationToken));
}
await Task.WhenAll(utTaskList).ConfigureAwait(false);
// TODO: Wait for some time so that the unreliable transport behavior can be read from the files.
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successfully applied
await Task.Delay(TimeSpan.FromSeconds(5.0), cancellationToken);
var restartReplicaTaskList = new List <Task>();
foreach (var statefulReplica in stableReplicasToRemove)
{
ReplicaSelector replicaSelector = ReplicaSelector.ReplicaIdOf(PartitionSelector.PartitionIdOf(action.PartitionSelector.ServiceName, partitionId), statefulReplica.Id);
var restartReplicaAction = new RestartReplicaAction(replicaSelector)
{
CompletionMode = CompletionMode.DoNotVerify,
RequestTimeout = action.RequestTimeout,
ActionTimeout = helper.GetRemainingTime()
};
restartReplicaTaskList.Add(testContext.ActionExecutor.RunAsync(restartReplicaAction, cancellationToken));
}
await Task.WhenAll(restartReplicaTaskList).ConfigureAwait(false);
await AsyncWaiter.WaitAsync(action.QuorumLossDuration, cancellationToken).ConfigureAwait(false);
// validate
ServicePartitionList partitionsResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetPartitionListAsync(
action.PartitionSelector.ServiceName,
null,
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.GetPartitionListFabricErrors.Value,
helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
foreach (StatefulServicePartition partition in partitionsResult)
{
if (partition.PartitionInformation.Id == partitionId)
{
ReleaseAssert.AssertIf(partition.PartitionStatus != ServicePartitionStatus.InQuorumLoss, "Partition failed to be in Quorum Loss.");
break;
}
}
foreach (var removeUTParams in removeUTRequestList)
{
var currentParams = removeUTParams;
Task task = FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.TestManager.RemoveUnreliableTransportBehaviorAsync(
currentParams.Item1, /*nodeName*/
currentParams.Item2, /*behaviorName*/
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.RemoveUnreliableTransportBehaviorErrors.Value,
helper.GetRemainingTime(),
cancellationToken);
removeUTTaskDictionary[currentParams] = task;
}
await Task.WhenAll(removeUTTaskDictionary.Values).ConfigureAwait(false);
// TODO: Wait for some time so that the removal of this unreliable transport behavior can be read from the files.
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successully applied
await Task.Delay(TimeSpan.FromSeconds(5.0), cancellationToken);
}
finally
{
var removeUTTaskList = new List <Task>();
foreach (var removeUTRequest in removeUTTaskDictionary)
{
var currentRemoveUTRequest = removeUTRequest;
if (currentRemoveUTRequest.Value == null || currentRemoveUTRequest.Value.IsFaulted)
{
removeUTTaskList.Add(
FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.TestManager.RemoveUnreliableTransportBehaviorAsync(
currentRemoveUTRequest.Key.Item1, /*nodeName*/
currentRemoveUTRequest.Key.Item2, /*behaviorName*/
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.RemoveUnreliableTransportBehaviorErrors.Value,
helper.GetRemainingTime(),
cancellationToken));
}
}
Task.WhenAll(removeUTTaskList).Wait(cancellationToken);
// TODO: Wait for some time so that the removal of this unreliable transport behavior can be read from the files.
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successully applied
Task.Delay(TimeSpan.FromSeconds(5.0), cancellationToken).GetAwaiter().GetResult();
}
action.Result = new InvokeQuorumLossResult(getPartitionStateAction.Result);
this.ResultTraceString = StringHelper.Format("InvokeQuorumLossAction succeeded for {0} with QuorumLossMode = {1}", partitionId, action.QuorumLossMode);
}
public override async Task <ActionStateBase> RunAsync(CancellationToken cancellationToken, ServiceInternalFaultInfo serviceInternalFaultInfo)
{
InvokeQuorumLossState state = Convert(this.State);
// get info about the service so we can check type and trss
ServiceDescription result = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.ServiceManager.GetServiceDescriptionAsync(
this.partitionSelector.ServiceName,
this.RequestTimeout,
cancellationToken),
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
if (result.Kind != ServiceDescriptionKind.Stateful)
{
// The message in the first arg is only for debugging, it is not returned to the user.
throw new FabricInvalidForStatelessServicesException("FabricInvalidForStatelessServicesException", FabricErrorCode.InvalidForStatelessServices);
}
StatefulServiceDescription statefulServiceDescription = result as StatefulServiceDescription;
ReleaseAssert.AssertIf(statefulServiceDescription == null, string.Format(CultureInfo.InvariantCulture, "{0} - Service is not a stateful service", this.State.OperationId));
if (!statefulServiceDescription.HasPersistedState)
{
// The message in the first arg is only for debugging, it is not returned to the user.
throw new FabricOnlyValidForStatefulPersistentServicesException("This is only valid for stateful persistent services", FabricErrorCode.OnlyValidForStatefulPersistentServices);
}
SelectedPartition targetPartition = await FaultAnalysisServiceUtility.GetSelectedPartitionStateAsync(
this.FabricClient,
this.partitionSelector,
this.RequestTimeout,
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
Guid partitionId = targetPartition.PartitionId;
// get data about replicas in that partition
ServiceReplicaList replicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.QueryManager.GetReplicaListAsync(
partitionId,
0,
this.RequestTimeout,
cancellationToken),
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
List <StatefulServiceReplica> tempReplicas = new List <StatefulServiceReplica>();
foreach (var replica in replicasResult)
{
StatefulServiceReplica statefulReplica = replica as StatefulServiceReplica;
ReleaseAssert.AssertIf(statefulReplica == null, "Expected stateful replica");
tempReplicas.Add(statefulReplica);
}
List <StatefulServiceReplica> targetReplicas = null;
if (this.quorumLossMode == QuorumLossMode.AllReplicas)
{
targetReplicas = tempReplicas.Where(r => r.ReplicaRole == ReplicaRole.Primary || r.ReplicaRole == ReplicaRole.ActiveSecondary).ToList();
}
else if (this.quorumLossMode == QuorumLossMode.QuorumReplicas)
{
targetReplicas = FaultAnalysisServiceUtility.GetReplicasForPartialLoss(state.OperationId, tempReplicas);
}
else
{
throw FaultAnalysisServiceUtility.CreateException(StepBase.TraceType, Interop.NativeTypes.FABRIC_ERROR_CODE.E_INVALIDARG, Strings.StringResources.Error_UnsupportedQuorumLossMode);
}
if (targetReplicas == null)
{
// This will cause the command to rollback and retry
throw new FabricTransientException("The operation could not be performed, please retry", FabricErrorCode.NotReady);
}
List <string> targetNodes = new List <string>();
foreach (var replica in targetReplicas)
{
targetNodes.Add(replica.NodeName);
}
List <Tuple <string, string> > unreliableTransportInfoList = new List <Tuple <string, string> >();
foreach (string nodeName in targetNodes)
{
UnreliableTransportBehavior behavior = new UnreliableTransportBehavior("*", "StatefulServiceReopen");
behavior.AddFilterForPartitionId(partitionId);
// ApplyingUnreliableTransport.BehaviorNamePrefix + nodeName;
string behaviorName = this.CreateBehaviorName(nodeName);
unreliableTransportInfoList.Add(new Tuple <string, string>(nodeName, behaviorName));
}
state.StateProgress.Push(StepStateNames.PerformingActions);
state.Info.PartitionId = partitionId;
state.Info.ReplicaIds = targetReplicas.Select(r => r.Id).ToList();
state.Info.UnreliableTransportInfo = unreliableTransportInfoList;
return(state);
}
public override async Task <ActionStateBase> RunAsync(CancellationToken cancellationToken, ServiceInternalFaultInfo serviceInternalFaultInfo)
{
TestabilityTrace.TraceSource.WriteInfo(StepBase.TraceType, "Inside CollectingState, service={0}", this.partitionSelector.ServiceName);
RestartPartitionState state = Convert(this.State);
// Get service info and validate if the parameters are valid
ServiceDescription result = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.ServiceManager.GetServiceDescriptionAsync(
this.partitionSelector.ServiceName,
this.RequestTimeout,
cancellationToken),
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
if (result.Kind != ServiceDescriptionKind.Stateful && this.restartPartitionMode == RestartPartitionMode.OnlyActiveSecondaries)
{
// The message in the first arg is only for debugging, it is not returned to the user.
string debugText = string.Format(CultureInfo.InvariantCulture, "RestartPartition: for stateless services only RestartPartitionMode.AllReplicasOrInstances is valid");
TestabilityTrace.TraceSource.WriteWarning(StepBase.TraceType, debugText);
throw FaultAnalysisServiceUtility.CreateException(StepBase.TraceType, NativeTypes.FABRIC_ERROR_CODE.E_INVALIDARG, debugText);
}
bool hasPersistedState = false;
if (result.Kind == ServiceDescriptionKind.Stateful)
{
StatefulServiceDescription statefulDescription = result as StatefulServiceDescription;
ReleaseAssert.AssertIf(statefulDescription == null, "Stateful service description is not WinFabricStatefulServiceDescription");
hasPersistedState = statefulDescription.HasPersistedState;
}
SelectedPartition targetPartition = await FaultAnalysisServiceUtility.GetSelectedPartitionStateAsync(
this.FabricClient,
this.partitionSelector,
this.RequestTimeout,
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
Guid partitionId = targetPartition.PartitionId;
// get replicas for target
ServiceReplicaList replicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.QueryManager.GetReplicaListAsync(
partitionId,
0,
this.RequestTimeout,
cancellationToken),
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
// get replicas for fm in order to get the primary
ServiceReplicaList failoverManagersReplicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.QueryManager.GetReplicaListAsync(
FASConstants.FmPartitionId,
0,
this.RequestTimeout,
cancellationToken),
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
string failoverManagerPrimaryNodeName = string.Empty;
var readyFMReplicas = failoverManagersReplicasResult.Where(r => r.ReplicaStatus == ServiceReplicaStatus.Ready).ToArray();
foreach (var replica in readyFMReplicas)
{
StatefulServiceReplica statefulReplica = replica as StatefulServiceReplica;
ReleaseAssert.AssertIf(statefulReplica == null, "FM Replica is not a stateful replica");
if (statefulReplica.ReplicaRole == ReplicaRole.Primary)
{
failoverManagerPrimaryNodeName = replica.NodeName;
}
}
if (string.IsNullOrEmpty(failoverManagerPrimaryNodeName))
{
throw new FabricException(StringHelper.Format(StringResources.Error_PartitionPrimaryNotReady, "FailoverManager"), FabricErrorCode.NotReady);
}
TestabilityTrace.TraceSource.WriteInfo(StepBase.TraceType, "{0} - FM primary is at node={1}", this.State.OperationId, failoverManagerPrimaryNodeName);
string behaviorName = RestartingSelectedReplicas.UTBehaviorPrefixName + "_" + this.State.OperationId;
List <Tuple <string, string> > unreliableTransportInfo = new List <Tuple <string, string> >();
unreliableTransportInfo.Add(new Tuple <string, string>(failoverManagerPrimaryNodeName, behaviorName));
state.StateProgress.Push(StepStateNames.PerformingActions);
state.Info.PartitionId = partitionId;
state.Info.NodeName = failoverManagerPrimaryNodeName;
state.Info.HasPersistedState = hasPersistedState;
state.Info.UnreliableTransportInfo = unreliableTransportInfo;
return(state);
}
