public static async Task <Tuple <SelectedReplica, Replica> > GetSelectedReplicaAsync(
FabricClient fabricClient,
ReplicaSelector replicaSelector,
TimeSpan requestTimeout,
TimeSpan operationTimeout,
CancellationToken cancellationToken)
{
ThrowIf.Null(replicaSelector, "ReplicaSelector");
SelectedPartition selectedPartition = await FaultAnalysisServiceUtility.GetSelectedPartitionStateAsync(
fabricClient,
replicaSelector.PartitionSelector,
requestTimeout,
operationTimeout,
cancellationToken).ConfigureAwait(false);
Guid partitionId = selectedPartition.PartitionId;
ServiceReplicaList replicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => fabricClient.QueryManager.GetReplicaListAsync(
partitionId,
0,
requestTimeout,
cancellationToken),
operationTimeout,
cancellationToken).ConfigureAwait(false);
Replica replicaResult = replicaSelector.GetSelectedReplica(replicasResult.ToArray(), new Random(), true /*skip invalid replicas*/);
var replicaSelectorResult = new SelectedReplica(replicaResult.Id, selectedPartition);
return(new Tuple <SelectedReplica, Replica>(replicaSelectorResult, replicaResult));
}
/// <summary>
/// Restarts a random node.
/// </summary>
/// <param name="serviceName">Uri of the service in the format fabric:/[application]/[service name]</param>
/// <returns>Task instance.</returns>
public async Task RestartNodeAsync(Uri serviceName)
{
PartitionSelector randomPartitionSelector = PartitionSelector.RandomOf(serviceName);
ReplicaSelector primaryofReplicaSelector = ReplicaSelector.PrimaryOf(randomPartitionSelector);
// Create FabricClient with connection and security information here
await _client.FaultManager.RestartNodeAsync(primaryofReplicaSelector, CompletionMode.Verify).ConfigureAwait(false);
}
public RestartDeployedCodePackageAction(
Uri applicationName,
ReplicaSelector replicaSelector)
{
this.ApplicationName = applicationName;
this.ReplicaSelector = replicaSelector;
this.CompletionMode = CompletionMode.Verify;
}
public static async Task <RestartReplicaResult> RestartReplicaAsync(
FabricClient fabricClient,
ReplicaSelector replicaSelector,
CompletionMode completionMode,
TimeSpan requestTimeout,
TimeSpan operationTimeout,
CancellationToken cancellationToken)
{
System.Fabric.Common.TimeoutHelper helper = new System.Fabric.Common.TimeoutHelper(operationTimeout);
string nodeName = null;
Guid partitionId = Guid.Empty;
long replicaId = 0;
SelectedReplica replicaSelectorResult = SelectedReplica.None;
System.Fabric.Common.ThrowIf.Null(replicaSelector, "ReplicaSelector");
Tuple <SelectedReplica, Replica> replicaStateActionResult = await FaultAnalysisServiceUtility.GetSelectedReplicaAsync(
fabricClient,
replicaSelector,
requestTimeout,
operationTimeout,
cancellationToken).ConfigureAwait(false);
replicaSelectorResult = replicaStateActionResult.Item1;
if (replicaSelectorResult == null)
{
throw new InvalidOperationException("replicaStateActionResult cannot be null");
}
partitionId = replicaStateActionResult.Item1.SelectedPartition.PartitionId;
Replica replicaStateResult = replicaStateActionResult.Item2;
if (replicaStateResult == null)
{
throw new InvalidOperationException("replicaStateResult cannot be null");
}
nodeName = replicaStateResult.NodeName;
replicaId = replicaStateResult.Id;
ThrowIf.IsTrue(partitionId == Guid.Empty, "PartitionID");
ThrowIf.IsTrue(replicaId == 0, "ReplicaID");
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => fabricClient.ServiceManager.RestartReplicaAsync(
nodeName,
partitionId,
replicaId,
requestTimeout,
cancellationToken),
FabricClientRetryErrors.RestartReplicaErrors.Value,
operationTimeout,
cancellationToken).ConfigureAwait(false);
return(new RestartReplicaResult(replicaSelectorResult));
}
public async Task <bool> RestartGetSnowConditionsService()
{
var replicaSelector = ReplicaSelector.PrimaryOf(PartitionSelector.RandomOf(GlobalContext.ServiceName));
//PartitionSelector namedPartitionSelector = PartitionSelector.PartitionKeyOf(new Uri(GlobalContext.appName), "Partition1");
//await fabricClient.ClusterManager.(new ServicePar); ; //statefull
return(true);
}
internal override Task <RestartDeployedCodePackageResult> InvokeCommandAsync(
IClusterConnection clusterConnection,
Uri applicationName,
ReplicaSelector replicaSelector)
{
return(clusterConnection.RestartDeployedCodePackageAsync(
replicaSelector,
applicationName,
this.TimeoutSec,
this.CommandCompletionMode ?? CompletionMode.Verify,
this.GetCancellationToken()));
}
private FabricTestAction GetRemoveReplicaAction(ReplicaStateTransitionAction ragAction)
{
Uri serviceUri = ragAction.ServiceUri;
Guid guid = ragAction.PartitionId;
long replicaId = ragAction.ReplicaId;
string report = StringHelper.Format("Generating Action: {0}\n\t\tService: {1}\n\t\tPartition: {2}\n\t\tReplicaId: {3}", ragAction.ActionType, ragAction.ServiceUri, ragAction.PartitionId, ragAction.ReplicaId);
// Select specific replica using ReplicaSelector.
ReplicaSelector replicaSelector = ReplicaSelector.ReplicaIdOf(PartitionSelector.PartitionIdOf(serviceUri, guid), replicaId);
RemoveReplicaAction restartReplicaAction = new RemoveReplicaAction(replicaSelector);
return(restartReplicaAction);
}
/// <summary>
/// The invoke command async.
/// </summary>
/// <param name="clusterConnection">
/// The cluster connection.
/// </param>
/// <param name="replicaSelector">
/// The replica selector.
/// </param>
/// <param name="completionMode">
/// The completion mode.
/// </param>
/// <param name="cancellationToken">
/// The cancellation token.
/// </param>
/// <returns>
/// The <see cref="Task"/>.
/// </returns>
internal override async Task <ReplicaResult> InvokeCommandAsync(
IClusterConnection clusterConnection,
ReplicaSelector replicaSelector,
CompletionMode completionMode,
CancellationToken cancellationToken)
{
return
(await
Task.FromResult(
clusterConnection.RestartReplicaAsync(
replicaSelector,
this.TimeoutSec,
completionMode,
cancellationToken).Result));
}
private async Task TestReplicaFaultsAsync(
ReplicaSelector replicaSelector,
string replicaRole,
bool hasPersistedState,
CancellationToken token)
{
await this.InvokeAndValidateFaultAsync(
StringHelper.Format("Removing replica state for {0}", replicaRole),
() =>
{
return(this.FabricClient.FaultManager.RemoveReplicaAsync(
replicaSelector,
CompletionMode.Verify,
false,
this.failoverTestScenarioParameters.OperationTimeout,
token));
}, token);
if (hasPersistedState)
{
await this.InvokeAndValidateFaultAsync(
StringHelper.Format("Restarting replica state for {0}", replicaRole),
() =>
{
return(this.FabricClient.FaultManager.RestartReplicaAsync(
replicaSelector,
CompletionMode.Verify,
this.failoverTestScenarioParameters.OperationTimeout,
token));
}, token);
}
await this.InvokeAndValidateFaultAsync(
StringHelper.Format("Restarting code package for {0} replica", replicaRole),
() =>
{
return(this.FabricClient.FaultManager.RestartDeployedCodePackageAsync(
this.serviceDescription.ApplicationName,
replicaSelector,
CompletionMode.Verify,
this.failoverTestScenarioParameters.OperationTimeout,
token));
}, token);
}
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 static ReplicaSelector GetExpectedReplicaSelector(ParitionSelectorTestHelper.PartitionCase partitionCase, ReplicaCase replicaCase)
{
ReplicaSelector result = null;
PartitionSelector partitionSelector = ParitionSelectorTestHelper.GetExpectedPartitionSelector(partitionCase);
switch (replicaCase)
{
case ReplicaCase.ReplicaPrimary:
{
result = ReplicaSelector.PrimaryOf(partitionSelector);
break;
}
case ReplicaCase.ReplicaRandomSecondary:
{
result = ReplicaSelector.RandomSecondaryOf(partitionSelector);
break;
}
case ReplicaCase.ReplicaId:
{
result = ReplicaSelector.ReplicaIdOf(partitionSelector, replicaInstance.Value);
break;
}
case ReplicaCase.ReplicaId_NoValue:
{
result = ReplicaSelector.ReplicaIdOf(partitionSelector, 0);
break;
}
case ReplicaCase.ReplicaRandom:
{
result = ReplicaSelector.RandomOf(partitionSelector);
break;
}
}
return(result);
}
/// <summary>
/// This API supports the Service Fabric platform and is not meant to be called from your code
/// </summary>
/// <param name="token">This API supports the Service Fabric platform and is not meant to be called from your code</param>
/// <returns></returns>
protected override async Task OnExecuteAsync(CancellationToken token)
{
this.serviceDescription = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.FabricClient.ServiceManager.GetServiceDescriptionAsync(
this.failoverTestScenarioParameters.PartitionSelector.ServiceName,
this.failoverTestScenarioParameters.RequestTimeout,
token),
this.failoverTestScenarioParameters.OperationTimeout,
token).ConfigureAwait(false);
bool hasPersistedState = false;
if (this.serviceDescription.IsStateful())
{
StatefulServiceDescription statefulDescription = this.serviceDescription as StatefulServiceDescription;
ReleaseAssert.AssertIf(statefulDescription == null, "Stateful service description is not WinFabricStatefulServiceDescription");
hasPersistedState = statefulDescription.HasPersistedState;
}
Log.WriteInfo(TraceType, "Validating Service health and availability");
await this.FabricClient.TestManager.ValidateServiceAsync(
this.failoverTestScenarioParameters.PartitionSelector.ServiceName,
this.failoverTestScenarioParameters.MaxServiceStabilizationTimeout,
token);
Log.WriteInfo(TraceType, "Getting Selected Partition");
var getPartitionStateAction = new GetSelectedPartitionStateAction(this.failoverTestScenarioParameters.PartitionSelector)
{
RequestTimeout = this.failoverTestScenarioParameters.RequestTimeout,
ActionTimeout = this.failoverTestScenarioParameters.OperationTimeout
};
await this.TestContext.ActionExecutor.RunAsync(getPartitionStateAction, token);
Guid selectedPartitionId = getPartitionStateAction.Result.PartitionId;
Log.WriteInfo(TraceType, "Running test for partition {0}", selectedPartitionId);
this.ReportProgress("Selected partition {0} for testing failover", selectedPartitionId);
PartitionSelector selectedPartition = PartitionSelector.PartitionIdOf(this.failoverTestScenarioParameters.PartitionSelector.ServiceName, selectedPartitionId);
while (this.failoverTestScenarioParameters.TimeToRun - this.GetElapsedTime() > TimeSpan.Zero && !token.IsCancellationRequested)
{
if (this.serviceDescription.IsStateful())
{
ReplicaSelector primaryReplicaSelector = ReplicaSelector.PrimaryOf(selectedPartition);
ReplicaSelector secondaryReplicaSelector = ReplicaSelector.RandomSecondaryOf(selectedPartition);
// Make Primary go through RemoveReplica, RestartReplica and RestartCodePackage
await this.TestReplicaFaultsAsync(primaryReplicaSelector, "Primary", hasPersistedState, token);
// Make Secondary go through RemoveReplica, RestartReplica and RestartCodePackage
await this.TestReplicaFaultsAsync(secondaryReplicaSelector, "Secondary", hasPersistedState, token);
}
else
{
ReplicaSelector randomInstanceSelector = ReplicaSelector.RandomOf(selectedPartition);
// Make Stateless Instance go through RemoveReplica, RestartReplica and RestartCodePackage
await this.TestReplicaFaultsAsync(randomInstanceSelector, "Stateless Instance", hasPersistedState, token);
}
if (this.serviceDescription.IsStateful())
{
// Restart all secondary replicas and make sure the replica set recovers
await this.InvokeAndValidateFaultAsync(
"Restarting all the secondary replicas",
() =>
{
#pragma warning disable 618
return(this.FabricClient.TestManager.RestartPartitionAsync(
selectedPartition,
RestartPartitionMode.OnlyActiveSecondaries,
this.failoverTestScenarioParameters.OperationTimeout,
token));
#pragma warning restore 618
}, token);
// Restart all replicas if service is persisted
if (hasPersistedState)
{
await this.InvokeAndValidateFaultAsync(
"Restarting all replicas including Primary",
() =>
{
#pragma warning disable 618
return(this.FabricClient.TestManager.RestartPartitionAsync(
selectedPartition,
RestartPartitionMode.AllReplicasOrInstances,
this.failoverTestScenarioParameters.OperationTimeout,
token));
#pragma warning restore 618
}, token);
}
// Induce move and swap primary a few times
await this.InvokeAndValidateFaultAsync(
"Move Primary to a different node",
() =>
{
return(this.FabricClient.FaultManager.MovePrimaryAsync(
string.Empty,
selectedPartition,
true,
this.failoverTestScenarioParameters.OperationTimeout,
token));
}, token);
// Induce move secondary a few times
await this.InvokeAndValidateFaultAsync(
"Move Secondary to a different node",
() =>
{
return(this.FabricClient.FaultManager.MoveSecondaryAsync(
string.Empty,
string.Empty,
selectedPartition,
true,
this.failoverTestScenarioParameters.OperationTimeout,
token));
}, token);
}
else
{
// Restart all stateless instances
await this.InvokeAndValidateFaultAsync(
"Restarting all stateless instances for partition",
() =>
{
#pragma warning disable 618
return(this.FabricClient.TestManager.RestartPartitionAsync(
selectedPartition,
RestartPartitionMode.AllReplicasOrInstances,
this.failoverTestScenarioParameters.OperationTimeout,
token));
#pragma warning restore 618
}, token);
}
}
}
internal static ReplicaSelector GetReplicaSelector(string partitionSetName, Guid partitionId, Uri serviceName, string partitionKey, long?replicaOrInstanceId)
{
ReplicaSelector replicaSelector = null;
PartitionSelector partitionSelector = null;
if (partitionSetName.Contains("PartitionId"))
{
partitionSelector = PartitionSelector.PartitionIdOf(serviceName, partitionId);
}
else
{
if (partitionSetName.Contains("PartitionSingleton"))
{
partitionSelector = PartitionSelector.SingletonOf(serviceName);
}
else if (partitionSetName.Contains("PartitionNamed"))
{
partitionSelector = PartitionSelector.PartitionKeyOf(serviceName, partitionKey);
}
else if (partitionSetName.Contains("PartitionUniformedInt"))
{
long partitionKeyLong;
if (!long.TryParse(partitionKey, out partitionKeyLong))
{
throw new ArgumentException(StringResources.Error_InvalidPartitionKey);
}
partitionSelector = PartitionSelector.PartitionKeyOf(serviceName, partitionKeyLong);
}
else if (!partitionSetName.Contains("Partition"))
{
partitionSelector = PartitionSelector.RandomOf(serviceName);
}
}
if (partitionSelector == null)
{
throw new ArgumentException(StringResources.Error_CouldNotParsePartitionSelector);
}
if (partitionSetName.Contains("ReplicaPrimary"))
{
replicaSelector = ReplicaSelector.PrimaryOf(partitionSelector);
}
else if (partitionSetName.Contains("ReplicaRandomSecondary"))
{
replicaSelector = ReplicaSelector.RandomSecondaryOf(partitionSelector);
}
else if (partitionSetName.Contains("ReplicaId"))
{
replicaSelector = ReplicaSelector.ReplicaIdOf(partitionSelector, replicaOrInstanceId ?? 0);
}
else if (!partitionSetName.Contains("Replica"))
{
replicaSelector = ReplicaSelector.RandomOf(partitionSelector);
}
if (replicaSelector == null)
{
throw new ArgumentException(StringResources.Error_CouldNotParseReplicaSelector);
}
return(replicaSelector);
}
internal abstract Task <RestartDeployedCodePackageResult> InvokeCommandAsync(
IClusterConnection clusterConnection,
Uri uri,
ReplicaSelector replicaSelector);
protected NodeControlRequest(IFabricClient fabricClient, ReplicaSelector replicaSelector, CompletionMode completionMode, TimeSpan timeout)
: base(fabricClient, timeout)
{
this.ReplicaSelector = replicaSelector;
this.CompletionMode = completionMode;
}
protected NodeControlRequest(IFabricClient fabricClient, ReplicaSelector replicaSelector, TimeSpan timeout)
: this(fabricClient, replicaSelector, CompletionMode.Invalid, timeout)
{
this.ReplicaSelector = replicaSelector;
}
public static string FormatOutput(string output, ReplicaSelector replicaSelector)
{
return(output.Replace("###ReplicaID###", replicaInstance.ToString()).Replace("###ReplicaSelector###", replicaSelector.ToString()));
}
public RestartNodeAction(ReplicaSelector replicaSelector, bool createFabricDump)
{
this.ReplicaSelector = replicaSelector;
this.CreateFabricDump = createFabricDump;
this.CompletionMode = CompletionMode.Verify;
}
public GetSelectedReplicaStateAction(ReplicaSelector replicaSelector)
{
this.ReplicaSelector = replicaSelector;
}
public RestartNodeRequest(IFabricClient fabricClient, ReplicaSelector replicaSelector, bool createFabricDump, CompletionMode completionMode, TimeSpan timeout)
: base(fabricClient, replicaSelector, completionMode, timeout)
{
this.createFabricDump = createFabricDump;
}
public RemoveReplicaAction(ReplicaSelector replicaSelector)
{
this.ReplicaSelector = replicaSelector;
this.CompletionMode = CompletionMode.Verify;
this.ForceRemove = false;
}
/// <summary>
/// The invoke command async.
/// </summary>
/// <param name="clusterConnection">
/// The cluster connection.
/// </param>
/// <param name="replicaSelector">
/// The replica selector.
/// </param>
/// <param name="completionMode">
/// The completion mode.
/// </param>
/// <param name="cancellationToken">
/// The cancellation token.
/// </param>
/// <returns>
/// The <see cref="Task"/>.
/// </returns>
internal abstract Task <ReplicaResult> InvokeCommandAsync(
IClusterConnection clusterConnection,
ReplicaSelector replicaSelector,
CompletionMode completionMode,
CancellationToken cancellationToken);
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 RestartReplicaAction(ReplicaSelector replicaSelector)
{
this.ReplicaSelector = replicaSelector;
this.CompletionMode = CompletionMode.Verify;
}
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 static async Task <Replica> GetReplicaAsync(this IFabricTestabilityClient client, ReplicaSelector replicaSelector)
{
var systemFabricCient = GetSystemFabricClient(client);
GetSelectedReplicaStateAction replicaAction = new GetSelectedReplicaStateAction(replicaSelector);
await systemFabricCient.FabricClient.TestManager.TestContext.ActionExecutor.RunAsync(replicaAction);
return(replicaAction.Result.Item2);
}
