internal System.Fabric.Common.UnreliableTransportBehavior Convert()
{
var utb = new System.Fabric.Common.UnreliableTransportBehavior(
this.Destination,
this.ActionName,
this.ProbabilityToApply,
this.Delay,
this.DelaySpan,
this.Priority,
this.ApplyCount);
utb.Filters = this.Filters;
return(utb);
}
public override async Task <ActionStateBase> RunAsync(CancellationToken cancellationToken, ServiceInternalFaultInfo serviceInternalFaultInfo)
{
InvokeQuorumLossState state = Convert(this.State);
Guid partitionId = state.Info.PartitionId;
List <Tuple <string, string> > unreliableTransportInfo = state.Info.UnreliableTransportInfo;
List <long> targetReplicas = state.Info.ReplicaIds;
var unreliableTransportTaskList = new List <Task>();
List <Tuple <string, string> > unreliableTransportInfoList = new List <Tuple <string, string> >();
foreach (Tuple <string, string> ut in unreliableTransportInfo)
{
string nodeName = ut.Item1;
string behaviorName = ut.Item2;
System.Fabric.Common.UnreliableTransportBehavior behavior = new System.Fabric.Common.UnreliableTransportBehavior("*", "StatefulServiceReopen");
behavior.AddFilterForPartitionId(partitionId);
TestabilityTrace.TraceSource.WriteInfo(StepBase.TraceType, "{0} - applying '{1}'", this.State.OperationId, behaviorName);
unreliableTransportTaskList.Add(FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.TestManager.AddUnreliableTransportBehaviorAsync(
nodeName,
behaviorName,
behavior,
this.RequestTimeout,
cancellationToken),
this.OperationTimeout,
cancellationToken));
}
await Task.WhenAll(unreliableTransportTaskList).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);
List <Task> tasks = new List <Task>();
foreach (long replicaId in targetReplicas)
{
ReplicaSelector replicaSelector = ReplicaSelector.ReplicaIdOf(PartitionSelector.PartitionIdOf(this.partitionSelector.ServiceName, partitionId), replicaId);
TestabilityTrace.TraceSource.WriteInfo(StepBase.TraceType, "{0} - faulting replica with id={1}", this.State.OperationId, replicaId);
Task task = FaultAnalysisServiceUtility.RestartReplicaAsync(this.FabricClient, replicaSelector, CompletionMode.DoNotVerify, this.RequestTimeout, this.OperationTimeout, cancellationToken);
tasks.Add(task);
}
await Task.WhenAll(tasks).ConfigureAwait(false);
ActionTest.PerformInternalServiceFaultIfRequested(this.State.OperationId, serviceInternalFaultInfo, this.State, cancellationToken, true);
TestabilityTrace.TraceSource.WriteInfo(StepBase.TraceType, "{0} - keeping partition in quorum loss for '{1}'", this.State.OperationId, state.Info.QuorumLossDuration);
await Task.Delay(state.Info.QuorumLossDuration, cancellationToken).ConfigureAwait(false);
TimeoutHelper timeoutHelper = new TimeoutHelper(this.OperationTimeout);
bool conditionSatisfied = false;
int quorumLossCheckRetries = FASConstants.QuorumLossCheckRetryCount;
do
{
TestabilityTrace.TraceSource.WriteInfo(StepBase.TraceType, "{0} - checking PartitionStatus", this.State.OperationId);
ServicePartitionList partitionsResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.QueryManager.GetPartitionListAsync(
this.partitionSelector.ServiceName,
null,
this.RequestTimeout,
cancellationToken),
this.OperationTimeout,
cancellationToken).ConfigureAwait(false);
foreach (StatefulServicePartition partition in partitionsResult)
{
if (partition.PartitionInformation.Id == partitionId)
{
if (partition.PartitionStatus == ServicePartitionStatus.InQuorumLoss)
{
conditionSatisfied = true;
break;
}
}
}
await AsyncWaiter.WaitAsync(TimeSpan.FromSeconds(5), cancellationToken).ConfigureAwait(false);
}while (!conditionSatisfied && quorumLossCheckRetries-- > 0);
if (!conditionSatisfied)
{
string error = string.Format(CultureInfo.InvariantCulture, "{0} - Service could not induce quorum loss for service '{1}', partition '{2}'. Please retry", this.State.OperationId, this.partitionSelector.ServiceName, partitionId);
TestabilityTrace.TraceSource.WriteWarning(StepBase.TraceType, error);
throw new FabricTransientException("The operation could not be performed, please retry", FabricErrorCode.NotReady);
}
await QuorumLossStepsFactory.RemoveUTAsync(this.FabricClient, this.State, this.RequestTimeout, this.OperationTimeout, cancellationToken);
state.StateProgress.Push(StepStateNames.CompletedSuccessfully);
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);
}