protected override async Task ExecuteActionAsync(FabricTestContext testContext, ValidateApplicationServicesAction action, CancellationToken cancellationToken)
{
ThrowIf.Null(action.ApplicationName, "ApplicationName");
TimeoutHelper helper = new TimeoutHelper(action.MaximumStabilizationTimeout);
// TODO: make these actions which store state locally as well.
var serviceListResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetServiceListAsync(
action.ApplicationName,
null,
action.RequestTimeout,
cancellationToken),
helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
List <Task> serviceValidationTasks = new List <Task>();
foreach (Service serviceResult in serviceListResult)
{
var validateService = new ValidateServiceAction(serviceResult.ServiceName, helper.GetRemainingTime())
{
ActionTimeout = action.ActionTimeout,
RequestTimeout = action.RequestTimeout,
CheckFlag = action.CheckFlag
};
serviceValidationTasks.Add(testContext.ActionExecutor.RunAsync(validateService, cancellationToken));
}
await Task.WhenAll(serviceValidationTasks.ToArray()).ConfigureAwait(false);
ResultTraceString = StringHelper.Format("ValidateApplicationServicesAction succeeded for {0}", action.ApplicationName);
}
internal static async Task <Node> GetNodeWithFASSecondary()
{
NodeList nodeList = ActionTest.GetNodeListAsync().Result;
ServiceReplicaList list = null;
FabricClient fc = new FabricClient();
System.Fabric.Common.TimeoutHelper timeoutHelper = new System.Fabric.Common.TimeoutHelper(TimeSpan.FromMinutes(2));
do
{
try
{
list = await fc.QueryManager.GetReplicaListAsync(new Guid("00000000-0000-0000-0000-000000005000"));
}
catch (Exception)
{
Task.Delay(TimeSpan.FromSeconds(1)).Wait();
}
}while (list == null && timeoutHelper.GetRemainingTime() > TimeSpan.Zero);
if (list == null)
{
throw new InvalidOperationException("Could not resolve FAS primary");
}
Replica replica = list.Where(r => ((StatefulServiceReplica)r).ReplicaRole == ReplicaRole.ActiveSecondary).FirstOrDefault();
return(nodeList.Where(n => n.NodeName == replica.NodeName).FirstOrDefault());
}
private void WaitForActionCount(long targetCount)
{
long count = 0;
System.Fabric.Common.TimeoutHelper timeoutHelper = new System.Fabric.Common.TimeoutHelper(TimeSpan.FromMinutes(3));
do
{
count = this.actionStore.GetActionCountAsync(false).GetAwaiter().GetResult();
TestabilityTrace.TraceSource.WriteInfo(TraceType, "Current action count='{0}', target action count='{1}'", count, targetCount);
if (count == targetCount)
{
break;
}
Task.Delay(TimeSpan.FromSeconds(5)).Wait();
}while (count != targetCount && timeoutHelper.GetRemainingTime() > TimeSpan.Zero);
if (count != targetCount)
{
string error = string.Format(CultureInfo.InvariantCulture, "Did not reach expected target action count='{0}', current action count='{1}'", targetCount, count);
TestabilityTrace.TraceSource.WriteError(TraceType, error);
System.Fabric.Common.ReleaseAssert.Failfast(error);
}
}
// Throws exception if validation was unsuccessful.
protected override async Task ExecuteActionAsync(FabricTestContext testContext, ValidateAllServicesAction action, CancellationToken token)
{
var timer = new TimeoutHelper(action.MaximumStabilizationTimeout);
//// Validate system services first.
var validateSystemServices = new ValidateSystemServicesAction(timer.GetRemainingTime())
{
ActionTimeout = action.ActionTimeout,
RequestTimeout = action.RequestTimeout,
CheckFlag = action.CheckFlag
};
await testContext.ActionExecutor.RunAsync(validateSystemServices, token).ConfigureAwait(false);
var appListResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetApplicationListAsync(
null,
string.Empty,
action.RequestTimeout,
token),
timer.GetRemainingTime(),
token).ConfigureAwait(false);
List <Task> tasks = new List <Task>();
foreach (Application appResult in appListResult)
{
var validateAppServices = new ValidateApplicationServicesAction(appResult.ApplicationName, timer.GetRemainingTime())
{
ActionTimeout = action.ActionTimeout,
RequestTimeout = action.RequestTimeout,
CheckFlag = action.CheckFlag
};
var task = testContext.ActionExecutor.RunAsync(validateAppServices, token);
tasks.Add(task);
this.ActionTraceSource.WriteNoise(TraceType, "ValidateAllServicesActionHandler: Validation task added for application: {0}", appResult.ApplicationName.OriginalString);
}
await Task.WhenAll(tasks).ConfigureAwait(false);
this.ResultTraceString = "ValidateAllServicesActionHandler completed for all services";
}
// This doesn't run in automation, but it is being kept here so it can be run as a small test.
// See FaultAnalysisServiceTruncate.test for a test on this code path.
private void TestTruncate()
{
this.StartActionIfItHasNotBeenStarted(Command.StuckAction);
this.StartActionIfItHasNotBeenStarted(Command.FailoverManagerDataLoss);
this.StartActionIfItHasNotBeenStarted(Command.InvokeDataLossMidActionTestFatal);
this.StartActionIfItHasNotBeenStarted(Command.InvokeDataLossMidActionTestTransient);
this.StartActionIfItHasNotBeenStarted(Command.RestartPartitionMidActionTestFatal);
this.StartActionIfItHasNotBeenStarted(Command.RestartPartitionMidActionTestTransient);
this.WaitForActionCount(FASConstants.TestMaxStoredActionCountValue);
// Confirm this action is still stuck - ie that an action not in terminal state is not removed
this.mockClient.WaitForState(MockClient.MockClientCommandInfo[Command.StuckAction], Actions.Steps.StepStateNames.LookingUpState);
this.mockClient.WaitForState(MockClient.MockClientCommandInfo[Command.RestartPartitionMidActionTestTransient], Actions.Steps.StepStateNames.CompletedSuccessfully);
// At this point there should be 1 command in the actionTable, StuckAction, and somewhere between 2 (Constants.TestMaxStoredActionCountValue) and 5 (the total number of possible
// completed commands) commands in the historyTable. In steady state, after truncates have run, the historyTable should have 2 (Constants.TestMaxStoredActionCountValue) commands remaining,
// and they should be the ones that completed last. Since this test only allows 1 action at a time, this will always be the 2 that were started last -
// the RestartPartition ones.
bool conditionSatisfied = false;
var timeoutHelper = new System.Fabric.Common.TimeoutHelper(TimeSpan.FromSeconds(3 * FASConstants.TestStoredActionCleanupIntervalInSeconds));
do
{
TestCommandListDescription queryDescription = new TestCommandListDescription(Query.TestCommandStateFilter.CompletedSuccessfully, Query.TestCommandTypeFilter.PartitionRestart);
TestCommandQueryResult queryResult = this.mockClient.GetTestCommandListAsync(queryDescription).GetAwaiter().GetResult();
List <TestCommandStatus> result = queryResult.Items;
if (result.Count < FASConstants.TestMaxStoredActionCountValue)
{
string error = string.Format(
CultureInfo.InvariantCulture,
"Number of commands in the historyTable {0} is below TestMaxStoredActionCountValue (config 'DefaultMaxStoredActionCount')",
result.Count);
TestabilityTrace.TraceSource.WriteError(TraceType, error);
System.Fabric.Common.ReleaseAssert.Failfast(error);
}
if (result.Where(c => c.TestCommandType == TestCommandType.PartitionRestart).Count() != FASConstants.TestMaxStoredActionCountValue)
{
TestabilityTrace.TraceSource.WriteInfo(TraceType, "Number of PartitionRestart results is {0}, expecting {1}, retrying", result.Count, FASConstants.TestMaxStoredActionCountValue);
continue;
}
conditionSatisfied = true;
}while (!conditionSatisfied && timeoutHelper.GetRemainingTime() > TimeSpan.Zero);
System.Fabric.Common.ReleaseAssert.Failfast(string.Format(CultureInfo.InvariantCulture, "Did not reach expected target action, see traces above filtered by type~ActionTest'"));
TestabilityTrace.TraceSource.WriteInfo(TraceType, "Exiting TestTruncate");
}
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);
}
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);
}
internal async Task ExecuteIterationsWithPauseAsync()
{
TestabilityTrace.TraceSource.WriteNoise("StartTrek", "Enter ExecuteIterationsWithPauseAsync, datetimeutc={0}", DateTime.UtcNow);
this.stopwatch = new Stopwatch();
this.stopwatch.Start();
Exception capturedException = null;
TimeSpan waitTime = this.ChaosParameters.WaitTimeBetweenIterations;
while (!this.cancellationToken.IsCancellationRequested)
{
// If this is not the beginning of a fresh Chaos run, before starting a new iteration,
// consult the NextIterationTimeStampRD to find out if there is some
// residual wait time from the previous iteration, if there is then wait that amount
var nextIterationTimeStampUtc = await this.StateManager.GetUtcTimeStampAsync(
FASConstants.NextItearationTimeStampRDName,
FASConstants.NextItearationTimeStampKey,
this.partition,
this.cancellationToken).ConfigureAwait(false);
var residualWaitTime = nextIterationTimeStampUtc.Subtract(DateTime.UtcNow);
if (residualWaitTime > TimeSpan.Zero)
{
await this.StateManager.RegisterChaosEventAndUpdateChaosStatusAsync(
new WaitingEvent(DateTime.UtcNow, StringHelper.Format(StringResources.ChaosInfo_ResidualWaitingFromPreviousIteration, residualWaitTime)),
ChaosStatus.Running,
this.partition,
this.cancellationToken,
() =>
{
TestabilityTrace.TraceSource.WriteInfo(TraceType, "Registering WaitingEvent for waiting '{0}' left over from the previous iteration.", residualWaitTime);
}).ConfigureAwait(false);
await Task.Delay(residualWaitTime, this.cancellationToken).ConfigureAwait(false);
}
try
{
if (await this.IsClusterReadyForFaultsAsync(this.cancellationToken).ConfigureAwait(false))
{
System.Fabric.Common.TimeoutHelper timer = new System.Fabric.Common.TimeoutHelper(this.ChaosParameters.MaxClusterStabilizationTimeout);
StringBuilder validationReport = new StringBuilder();
var clusterReport =
await
this.validationHelper.ValidateClusterHealthAsync(
timer.GetRemainingTime(),
this.cancellationToken).ConfigureAwait(false);
if (clusterReport.ValidationFailed)
{
// quadratic with an upper bound of DefaultMaximumBackoffForChaosIterations
waitTime += (waitTime >= FASConstants.DefaultMaximumBackoffForChaosIterations) ? TimeSpan.Zero : waitTime;
var serviceReport =
await
this.validationHelper.ValidateAllServicesAsync(
timer.GetRemainingTime(),
this.cancellationToken).ConfigureAwait(false);
if (serviceReport.ValidationFailed)
{
TestabilityTrace.TraceSource.WriteInfo(TraceType, "Even though some services are unhealthy or unstable, going to induce faults, because the cluster is healthy.");
TestabilityTrace.TraceSource.WriteInfo(TraceType, "Failure reason: \n'{0}'", serviceReport.FailureReason);
validationReport.Append(serviceReport.FailureReason);
}
StringBuilder reportBuilder = new StringBuilder();
reportBuilder.Append(StringHelper.Format(StringResources.ChaosInfo_WaitingNotice, waitTime));
reportBuilder.AppendLine();
reportBuilder.AppendLine(clusterReport.FailureReason);
validationReport.Insert(0, reportBuilder.ToString());
TestabilityTrace.TraceSource.WriteInfo(TraceType, "Cluster validation failed for '{0}'.", clusterReport.FailureReason);
var validationFailedEvent = new ValidationFailedEvent(DateTime.UtcNow, validationReport.ToString());
// record validation failed event
await this.StateManager.RegisterChaosEventAndUpdateChaosStatusAsync(
validationFailedEvent,
ChaosStatus.Running,
this.partition,
this.cancellationToken,
() =>
{
FabricEvents.Events.ChaosValidationFailed(
Guid.NewGuid().ToString(),
validationFailedEvent.TimeStampUtc.Ticks,
validationFailedEvent.Reason);
}).ConfigureAwait(false);
TestabilityTrace.TraceSource.WriteInfo(TraceType, "Pausing for '{0}' before performing next check.", waitTime);
}
else
{
waitTime = this.ChaosParameters.WaitTimeBetweenIterations;
await this.StateManager.RegisterCurrentStatusAsync(
ChaosStatus.Running,
this.partition,
this.cancellationToken).ConfigureAwait(false);
var timestampOfNextIteration = DateTime.UtcNow.Add(waitTime);
await this.StateManager.SetUtcTimeStampAsync(
FASConstants.NextItearationTimeStampRDName,
FASConstants.NextItearationTimeStampKey,
timestampOfNextIteration,
this.partition,
this.cancellationToken).ConfigureAwait(false);
await this.ExecuteFaultIterationAsync(this.cancellationToken).ConfigureAwait(false);
TestabilityTrace.TraceSource.WriteInfo(TraceType, "Pausing for '{0}' before executing next iteration.", waitTime);
}
}
}
catch (Exception exception)
{
capturedException = exception;
// quadratic with an upper bound of DefaultMaximumBackoffForChaosIterations
waitTime += (waitTime >= FASConstants.DefaultMaximumBackoffForChaosIterations) ? TimeSpan.Zero : waitTime;
var timestampOfNextIteration = DateTime.UtcNow.Add(waitTime);
await this.StateManager.SetUtcTimeStampAsync(
FASConstants.NextItearationTimeStampRDName,
FASConstants.NextItearationTimeStampKey,
timestampOfNextIteration,
this.partition,
this.cancellationToken).ConfigureAwait(false);
TestabilityTrace.TraceSource.WriteInfo(TraceType, "Exception occurred in the ChaosTestScenario loop: {0}, cancellationreq: {1}", capturedException, this.cancellationToken.IsCancellationRequested);
}
if (capturedException != null)
{
var ae = capturedException as AggregateException;
if (ae != null)
{
capturedException = ae.Flatten().InnerException;
}
if (!ChaosUtil.IsExpectedException(capturedException))
{
string testErrorEventMessage = capturedException.Message;
if (capturedException is FabricChaosEngineException)
{
testErrorEventMessage = StringResources.ChaosError_UnexpectedInChaosEngine;
}
await this.StateManager.RegisterChaosEventAndUpdateChaosStatusAsync(
new TestErrorEvent(DateTime.UtcNow, testErrorEventMessage),
ChaosStatus.Running,
this.partition,
this.cancellationToken).ConfigureAwait(false);
TestabilityTrace.TraceSource.WriteInfo(TraceType, "Unexpected exception '{0}' was turned into TestErrorEvent.", capturedException);
}
// handled the exception, now clear it
capturedException = null;
}
if (this.testMode && (this.ChaosParameters.Context != null && this.ChaosParameters.Context.ContainsKey(ChaosConstants.FailoverAfterChaosFaultsKey)))
{
this.partition.ReportFault(FaultType.Transient);
}
await this.StateManager.RegisterChaosEventAndUpdateChaosStatusAsync(
new WaitingEvent(DateTime.UtcNow, StringHelper.Format(StringResources.ChaosInfo_WaitingBetweenIterations, waitTime)),
ChaosStatus.Running,
this.partition,
this.cancellationToken,
() =>
{
TestabilityTrace.TraceSource.WriteInfo(TraceType, "Registering WaitingEvent for waiting '{0}' between iterations.", waitTime);
}).ConfigureAwait(false);
await Task.Delay(waitTime, this.cancellationToken).ConfigureAwait(false);
this.IterationsCompleted++;
}
TestabilityTrace.TraceSource.WriteInfo(TraceType, "Session has completed. \nTotal iterations: {0}. Total elapsed time: {1}", this.IterationsCompleted, this.GetElapsedTime());
}
private static bool CheckRPCAccess(MachineHealthContainer machineHealthContainer)
{
var retryTimeout = new System.Fabric.Common.TimeoutHelper(DMConstants.BpaRpcRetryTimeout);
SFDeployerTrace.WriteNoise(StringResources.Info_SFRpcInfo);
Parallel.ForEach <string>(
machineHealthContainer.GetHealthyMachineNames(),
(string machine) =>
{
bool result = true;
bool willRetry;
do
{
willRetry = false;
try
{
Utility.GetTempPath(machine);
}
catch (Exception ex)
{
string message;
if (ex is System.IO.IOException)
{
switch (ex.HResult)
{
// If new failures are discovered: https://msdn.microsoft.com/en-us/library/windows/desktop/ms681382(v=vs.85).aspx
case 53: // ERROR_BAD_NETPATH
message = string.Format(StringResources.Error_SFRpcIoNetpath, machine, ex.HResult);
willRetry = true;
break;
case 1723: // RPC_S_SERVER_TOO_BUSY
message = string.Format(StringResources.Error_SFRpcIoTooBusy, machine, ex.HResult);
willRetry = true;
break;
case 1727: // RPC_S_CALL_FAILED_DNE
message = string.Format(StringResources.Error_SFRpcIoFailedDne, machine, ex.HResult);
break;
default:
message = string.Format(StringResources.Error_SFRpcIoGeneric, machine, ex.HResult);
break;
}
}
else if (ex is System.Security.SecurityException)
{
switch (ex.HResult)
{
case -2146233078: // COR_E_SECURITY
message = string.Format(StringResources.Error_SFRpcSecAccess, machine, ex.HResult);
break;
default:
message = string.Format(StringResources.Error_SFRpcSecGeneric, machine, ex.HResult);
break;
}
}
else if (ex is NullReferenceException)
{
switch (ex.HResult)
{
case -2146232828: // COR_E_TARGETINVOCATION
message = string.Format(StringResources.Error_SFRpcNullRegAccess, machine, ex.HResult);
break;
default:
message = string.Format(StringResources.Error_SFRpcNullGeneric, machine, ex.HResult);
break;
}
}
else
{
// This is to catch coding errors.
message = string.Format(StringResources.Error_SFRpcGeneric, machine, ex.HResult);
}
willRetry &= !System.Fabric.Common.TimeoutHelper.HasExpired(retryTimeout);
if (willRetry)
{
SFDeployerTrace.WriteWarning(message);
StandaloneUtility.OpenRemoteRegistryNamedPipe(machine, retryTimeout.GetRemainingTime());
Thread.Sleep(TimeSpan.FromSeconds(5));
}
else
{
SFDeployerTrace.WriteError(message);
result = false;
}
}
}while (willRetry);
if (!result)
{
machineHealthContainer.MarkMachineAsUnhealthy(machine);
}
});
return(machineHealthContainer.EnoughHealthyMachines());
}
public async Task <ValidationReport> EnsureStabilityWithReportAsync(TimeSpan maximumStabilizationTimeout, TimeSpan retryWait, CancellationToken ct)
{
TestabilityTrace.TraceSource.WriteInfo(TraceSource, "Ensuring that '{0}' is online with timeout '{1}'.", this.serviceName, maximumStabilizationTimeout);
bool checkQuorumLoss = (this.checkFlags & ValidationCheckFlag.CheckQuorumLoss) != 0;
// Load basic information about this service.
TestabilityTrace.TraceSource.WriteNoise(TraceSource, "Querying basic information for {0}.", this.serviceName);
await this.LoadPartitionAndReplicaCountAsync(ct);
DateTime startTime = DateTime.Now;
TimeoutHelper timer = new TimeoutHelper(maximumStabilizationTimeout);
bool success = false;
List <Guid> partitionsInQuorumLoss = new List <Guid>();
StringBuilder errorString = new StringBuilder();
int retryCount = 1;
while (!success && timer.GetRemainingTime() > TimeSpan.Zero)
{
TestabilityTrace.TraceSource.WriteInfo(TraceSource, "EnsureStabilityWithReportAsync(): retryCount='{0}', timer.GetRemainingTime()='{1}'", retryCount, timer.GetRemainingTime());
var nodes = await this.TestContext.FabricCluster.GetLatestNodeInfoAsync(this.requestTimeout, this.operationTimeout, ct);
// Empty error string and list of partitions in quorum loss
partitionsInQuorumLoss.Clear();
errorString.Clear();
success = true;
int totalPartitionsFound = 0;
bool stateful;
ReleaseAssert.AssertIfNot(this.isStateful.TryGetValue(out stateful), "isStateful flag is not available");
bool checkTarget = (this.checkFlags & ValidationCheckFlag.CheckTargetReplicaSetSize) != 0;
bool checkInBuild = (this.checkFlags & ValidationCheckFlag.CheckInBuildReplica) != 0;
if (stateful)
{
var partitionDictionary = await this.QueryPartitionAndReplicaResultAsyncStateful(ct);
totalPartitionsFound = partitionDictionary.Count();
foreach (KeyValuePair <Partition, StatefulServiceReplica[]> partition in partitionDictionary)
{
bool partitionIsReady = partition.Key.PartitionStatus == ServicePartitionStatus.Ready;
if (!partitionIsReady)
{
var message = StringHelper.Format("Partition '{0}' is not Ready", partition.Key.PartitionId());
TestabilityTrace.TraceSource.WriteInfo(TraceSource, "{0}", message);
errorString.AppendLine(message);
}
if (partition.Key.PartitionStatus != ServicePartitionStatus.InQuorumLoss)
{
int validCount = 0;
int inBuildReplicas = 0;
foreach (StatefulServiceReplica replica in partition.Value)
{
if (replica.ReplicaStatus == ServiceReplicaStatus.Ready &&
(replica.ReplicaRole == ReplicaRole.Primary || replica.ReplicaRole == ReplicaRole.ActiveSecondary))
{
++validCount;
}
if (replica.ReplicaStatus == ServiceReplicaStatus.InBuild)
{
++inBuildReplicas;
var message = StringHelper.Format("Replica {0} for partition '{1}' is InBuild", replica.Id, partition.Key.PartitionId());
TestabilityTrace.TraceSource.WriteInfo(TraceSource, "{0}", message);
errorString.AppendLine(message);
}
}
bool targetAchieved = this.CheckReplicaSetSize(partition.Key.PartitionInformation.Id, validCount, startTime, nodes, errorString);
if (!partitionIsReady ||
(checkInBuild && inBuildReplicas > 0) ||
(checkTarget && !targetAchieved))
{
success = false;
}
}
else
{
partitionsInQuorumLoss.Add(partition.Key.PartitionInformation.Id);
}
}
}
else
{
int targetInstanceCount = 0;
ReleaseAssert.AssertIf(!this.targetReplicaSetSize.TryGetValue(out targetInstanceCount), "targetReplicaSetSize for service: {0} should have been populated at this point.", this.serviceName);
bool placementConstraintsDefined = false;
try
{
// Get the service description to find out if there are placement constraints on the service
ServiceDescription result = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.TestContext.FabricClient.ServiceManager.GetServiceDescriptionAsync(
this.serviceName,
this.requestTimeout,
ct),
this.operationTimeout,
ct).ConfigureAwait(false);
ThrowIf.IsTrue(result == null, "A description must be associated with the service: {0}", this.serviceName);
placementConstraintsDefined = !string.IsNullOrEmpty(result.PlacementConstraints);
}
catch (UnauthorizedAccessException)
{
ServiceGroupDescription groupDescription = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => this.TestContext.FabricClient.ServiceGroupManager.GetServiceGroupDescriptionAsync(
this.serviceName,
this.requestTimeout,
ct),
this.operationTimeout,
ct).ConfigureAwait(false);
ThrowIf.IsTrue(groupDescription == null, "A description must be associated with the service group: {0}", this.serviceName);
placementConstraintsDefined = !string.IsNullOrEmpty(groupDescription.ServiceDescription.PlacementConstraints);
}
// If a stateless service has instance count == -1 and it has placement constraints such
// that the possible number of instances cannot match the total number of nodes,
// we need to find out the number of eligible nodes for the service which is tracked by RDBug 8993319.
// Until RDBug 8993319 is fixed, we take the presence of placement constraints into consideration to make the
// validation more accurate.
if (targetInstanceCount == -1 && placementConstraintsDefined)
{
checkTarget = false;
}
var partitionDictionary = await this.QueryPartitionAndReplicaResultAsyncStateless(timer.GetRemainingTime(), ct);
totalPartitionsFound = partitionDictionary.Count();
foreach (KeyValuePair <Partition, StatelessServiceInstance[]> partition in partitionDictionary)
{
bool partitionIsReady = partition.Key.PartitionStatus == ServicePartitionStatus.Ready;
if (!partitionIsReady)
{
var message = StringHelper.Format("Partition '{0}' is not Ready", partition.Key.PartitionId());
TestabilityTrace.TraceSource.WriteInfo(TraceSource, "{0}", message);
errorString.AppendLine(message);
}
int validCount = 0;
foreach (StatelessServiceInstance instance in partition.Value)
{
if (instance.ReplicaStatus == ServiceReplicaStatus.Ready)
{
++validCount;
}
}
bool targetAchieved = this.CheckReplicaSetSize(partition.Key.PartitionInformation.Id, validCount, startTime, nodes, errorString);
if (!partitionIsReady ||
(checkTarget && !targetAchieved))
{
success = false;
}
}
}
if (!this.ValidatePartitionCount(totalPartitionsFound))
{
success = false;
}
if (partitionsInQuorumLoss.Count > 0 && checkQuorumLoss)
{
string paritionIds = string.Join(",", partitionsInQuorumLoss.ToArray());
var message = StringHelper.Format("Partitions '{0}' in quorum loss for service {1}", paritionIds, this.serviceName);
TestabilityTrace.TraceSource.WriteInfo(TraceSource, "{0}", message);
errorString.AppendLine(message);
success = false;
}
if (!success)
{
if (retryCount % 10 == 0)
{
TestabilityTrace.TraceSource.WriteWarning(TraceSource, "Service {0} validation failed due to issues below, will retry: \n{1}", this.serviceName, errorString);
}
// Delay before querying again so we allow some time for state to change - don't spam the node
await AsyncWaiter.WaitAsync(retryWait, ct).ConfigureAwait(false);
}
retryCount++;
}
if (partitionsInQuorumLoss.Count > 0)
{
string partitionIds = string.Join(",", partitionsInQuorumLoss.ToArray());
TestabilityTrace.TraceSource.WriteInfo(TraceSource, "Partitions in quorum loss for service {0} are '{1}'", this.serviceName, partitionIds);
if (checkQuorumLoss)
{
throw new FabricValidationException(StringHelper.Format(StringResources.Error_PartitionsInQuorumLoss, partitionIds, this.serviceName));
}
}
if (!success)
{
return(new ValidationReport(
true,
StringHelper.Format(StringResources.Error_ServiceNotStable, this.serviceName, maximumStabilizationTimeout, errorString)));
}
else
{
return(ValidationReport.Default);
}
}
private async Task <string> GetUnhealthyItemsAsync(ServiceHealth serviceHealth, TimeoutHelper timer, CancellationToken ct)
{
StringBuilder healthinfo = new StringBuilder();
foreach (var serviceHealthEvent in serviceHealth.HealthEvents)
{
if (serviceHealthEvent.HealthInformation.HealthState == HealthState.Ok)
{
continue;
}
healthinfo.AppendLine(StringHelper.Format(
"Service {0} health state is '{1}' with property '{2}', sourceId '{3}' and description '{4}'",
this.serviceName,
serviceHealthEvent.HealthInformation.HealthState,
serviceHealthEvent.HealthInformation.Property,
serviceHealthEvent.HealthInformation.SourceId,
serviceHealthEvent.HealthInformation.Description));
}
foreach (var partitionHealthState in serviceHealth.PartitionHealthStates)
{
if (partitionHealthState.AggregatedHealthState == HealthState.Ok)
{
continue;
}
var partitionHealth = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() =>
this.TestContext.FabricClient.HealthManager.GetPartitionHealthAsync(
partitionHealthState.PartitionId,
this.requestTimeout,
ct),
FabricClientRetryErrors.GetEntityHealthFabricErrors.Value,
timer.GetRemainingTime(),
ct).ConfigureAwait(false);
foreach (var partitionHealthEvent in partitionHealth.HealthEvents)
{
if (partitionHealthEvent.HealthInformation.HealthState == HealthState.Ok)
{
continue;
}
healthinfo.AppendLine(StringHelper.Format(
"Service {0}:{1} health state is '{2}' with property '{3}', sourceId '{4}' and description '{5}'",
this.serviceName,
partitionHealth.PartitionId,
partitionHealthEvent.HealthInformation.HealthState,
partitionHealthEvent.HealthInformation.Property,
partitionHealthEvent.HealthInformation.SourceId,
partitionHealthEvent.HealthInformation.Description));
}
foreach (var replicaHealthState in partitionHealth.ReplicaHealthStates)
{
if (replicaHealthState.AggregatedHealthState == HealthState.Ok)
{
continue;
}
var replicaHealth = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() =>
this.TestContext.FabricClient.HealthManager.GetReplicaHealthAsync(
replicaHealthState.PartitionId,
replicaHealthState.Id,
this.requestTimeout,
ct),
FabricClientRetryErrors.GetEntityHealthFabricErrors.Value,
timer.GetRemainingTime(),
ct).ConfigureAwait(false);
foreach (var replicaHealthEvent in replicaHealth.HealthEvents)
{
if (replicaHealthEvent.HealthInformation.HealthState == HealthState.Ok)
{
continue;
}
healthinfo.AppendLine(StringHelper.Format(
"Service {0}:{1}:{2} health state is '{3}' with property '{4}', sourceId '{5}' and description '{6}'",
this.serviceName,
replicaHealth.PartitionId,
replicaHealth.Id,
replicaHealthEvent.HealthInformation.HealthState,
replicaHealthEvent.HealthInformation.Property,
replicaHealthEvent.HealthInformation.SourceId,
replicaHealthEvent.HealthInformation.Description));
}
}
}
return(healthinfo.ToString());
}
public async Task <ValidationReport> ValidateHealthWithReportAsync(TimeSpan maximumStabilizationTimeout, TimeSpan retryWait, CancellationToken ct)
{
TestabilityTrace.TraceSource.WriteInfo(TraceSource, "Validating that '{0}' is healthy with timeout '{1}'.", this.serviceName, maximumStabilizationTimeout);
TimeoutHelper timer = new TimeoutHelper(maximumStabilizationTimeout);
bool success = false;
string healthinfo = string.Empty;
int retryCount = 1;
while (!success && timer.GetRemainingTime() > TimeSpan.Zero)
{
TestabilityTrace.TraceSource.WriteInfo(TraceSource, "ValidateHealthWithReportAsync(): retryCount='{0}', timer.GetRemainingTime()='{1}'", retryCount, timer.GetRemainingTime());
healthinfo = string.Empty;
if (this.TestContext == null)
{
Console.WriteLine("testcontext is null");
}
ReleaseAssert.AssertIfNull(this.TestContext, "test context");
ReleaseAssert.AssertIfNull(this.serviceName, "serviceName");
ReleaseAssert.AssertIfNull(FabricClientRetryErrors.GetEntityHealthFabricErrors.Value, "health error code");
ApplicationHealthPolicy healthPolicy = new ApplicationHealthPolicy();
var serviceHealthResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() =>
this.TestContext.FabricClient.HealthManager.GetServiceHealthAsync(
this.serviceName,
healthPolicy,
this.requestTimeout,
ct),
FabricClientRetryErrors.GetEntityHealthFabricErrors.Value,
timer.GetRemainingTime(),
ct).ConfigureAwait(false);
bool checkError = (this.checkFlags & ValidationCheckFlag.CheckError) != 0;
bool checkWarning = (this.checkFlags & ValidationCheckFlag.CheckWarning) != 0;
if ((checkError && serviceHealthResult.AggregatedHealthState == HealthState.Error) ||
(checkWarning && serviceHealthResult.AggregatedHealthState == HealthState.Warning) ||
serviceHealthResult.AggregatedHealthState == HealthState.Invalid ||
serviceHealthResult.AggregatedHealthState == HealthState.Unknown)
{
TestabilityTrace.TraceSource.WriteInfo(TraceSource, "{0} is health state is {1}. Will Retry check", this.serviceName, serviceHealthResult.AggregatedHealthState);
healthinfo = await this.GetUnhealthyItemsAsync(serviceHealthResult, timer, ct).ConfigureAwait(false);
TestabilityTrace.TraceSource.WriteInfo(TraceSource, healthinfo);
}
else
{
success = true;
}
if (!success)
{
if (retryCount % 10 == 0)
{
TestabilityTrace.TraceSource.WriteWarning(TraceSource, "Service {0} health validation failed due to issues below, will retry: \n{1}", this.serviceName, healthinfo);
}
// Delay before querying again so we allow some time for state to change - don't spam the node
await AsyncWaiter.WaitAsync(retryWait);
}
retryCount++;
}
if (!success)
{
return(new ValidationReport(true, StringHelper.Format(StringResources.Error_ServiceNotHealthy, serviceName, maximumStabilizationTimeout, healthinfo)));
}
else
{
return(ValidationReport.Default);
}
}
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);
}
protected override async Task ExecuteActionAsync(FabricTestContext testContext, RemoveReplicaAction action, CancellationToken cancellationToken)
{
TimeoutHelper helper = new TimeoutHelper(action.ActionTimeout);
string nodeName = action.NodeName;
Guid? partitionId = action.PartitionId;
long? replicaId = action.ReplicaId;
SelectedReplica replicaSelectorResult = SelectedReplica.None;
if (string.IsNullOrEmpty(nodeName) ||
!partitionId.HasValue ||
!replicaId.HasValue)
{
ThrowIf.Null(action.ReplicaSelector, "ReplicaSelector");
var getReplicaStateAction = new GetSelectedReplicaStateAction(action.ReplicaSelector)
{
RequestTimeout = action.RequestTimeout,
ActionTimeout = helper.GetRemainingTime()
};
await testContext.ActionExecutor.RunAsync(getReplicaStateAction, cancellationToken).ConfigureAwait(false);
var replicaStateActionResult = getReplicaStateAction.Result;
ReleaseAssert.AssertIf(replicaStateActionResult == null, "replicaStateActionResult cannot be null");
replicaSelectorResult = replicaStateActionResult.Item1;
partitionId = replicaStateActionResult.Item1.SelectedPartition.PartitionId;
Replica replicaStateResult = replicaStateActionResult.Item2;
ReleaseAssert.AssertIf(replicaStateResult == null, "replicaStateResult cannot be null");
nodeName = replicaStateResult.NodeName;
replicaId = replicaStateResult.Id;
}
ThrowIf.IsFalse(partitionId.HasValue, "PartitionID");
ThrowIf.IsFalse(replicaId.HasValue, "ReplicaID");
bool forceRemove = action.ForceRemove;
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.ServiceManager.RemoveReplicaAsync(
nodeName,
partitionId.Value,
replicaId.Value,
forceRemove,
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.RemoveReplicaErrors.Value,
helper.GetRemainingTime(),
cancellationToken);
if (action.CompletionMode == CompletionMode.Verify)
{
// Check that replica on selected node has been removed i.e. the replica id does not exist anymore.
bool success = false;
while (helper.GetRemainingTime() > TimeSpan.Zero)
{
var replicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetReplicaListAsync(
partitionId.Value,
replicaId.Value,
action.RequestTimeout,
cancellationToken),
helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
bool dropped = replicasResult.Count == 0;
if (!dropped)
{
// Since we added a replica filter the result should contain the replica or none
ReleaseAssert.AssertIf(replicasResult.Count > 1, "More than 1 replica returned with replica filter {0}:{1}", partitionId.Value, replicaId.Value);
ReleaseAssert.AssertIf(replicasResult[0].Id != replicaId, "Incorrect replica Id {0} returned by query instead of {1}", replicasResult[0].Id, replicaId);
dropped = replicasResult[0].ReplicaStatus == ServiceReplicaStatus.Dropped;
}
if (dropped)
{
success = true;
break;
}
ActionTraceSource.WriteInfo(TraceSource, "Replica = {0}:{1} not yet completely removed. Retrying...", partitionId.Value, replicaId.Value);
await AsyncWaiter.WaitAsync(TimeSpan.FromSeconds(5), cancellationToken);
}
if (!success)
{
throw new TimeoutException(StringHelper.Format(StringResources.Error_TestabilityActionTimeout,
"RemoveReplica",
StringHelper.Format("{0}:{1}", partitionId.Value, replicaId.Value)));
}
}
action.Result = new RemoveReplicaResult(replicaSelectorResult);
ResultTraceString = StringHelper.Format(
"RemoveReplicaOrInstance succeeded by removing replica {0}:{1} on node {2} with CompletionMode {3}",
partitionId.Value,
replicaId.Value,
nodeName,
action.CompletionMode);
}
public static bool HasExpired(TimeoutHelper timeoutHelper)
{
return(timeoutHelper.GetRemainingTime() == TimeSpan.Zero);
}
protected override async Task ExecuteActionAsync(FabricTestContext testContext, RestartReplicaAction action, CancellationToken cancellationToken)
{
TimeoutHelper helper = new TimeoutHelper(action.ActionTimeout);
string nodeName = action.NodeName;
Guid? partitionId = action.PartitionId;
long? replicaId = action.ReplicaId;
SelectedReplica replicaSelectorResult = SelectedReplica.None;
if (string.IsNullOrEmpty(nodeName) ||
!partitionId.HasValue ||
!replicaId.HasValue)
{
ThrowIf.Null(action.ReplicaSelector, "ReplicaSelector");
var getReplicaStateAction = new GetSelectedReplicaStateAction(action.ReplicaSelector)
{
RequestTimeout = action.RequestTimeout,
ActionTimeout = helper.GetRemainingTime()
};
await testContext.ActionExecutor.RunAsync(getReplicaStateAction, cancellationToken).ConfigureAwait(false);
var replicaStateActionResult = getReplicaStateAction.Result;
replicaSelectorResult = replicaStateActionResult.Item1;
ReleaseAssert.AssertIf(replicaSelectorResult == null, "replicaSelectorResult cannot be null");
partitionId = replicaStateActionResult.Item1.SelectedPartition.PartitionId;
Replica replicaStateResult = replicaStateActionResult.Item2;
ReleaseAssert.AssertIf(replicaStateResult == null, "replicaStateResult cannot be null");
nodeName = replicaStateResult.NodeName;
replicaId = replicaStateResult.Id;
}
ThrowIf.IsFalse(partitionId.HasValue, "PartitionID");
ThrowIf.IsFalse(replicaId.HasValue, "ReplicaID");
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.ServiceManager.RestartReplicaAsync(
nodeName,
partitionId.Value,
replicaId.Value,
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.RestartReplicaErrors.Value,
helper.GetRemainingTime(),
cancellationToken);
if (action.CompletionMode == CompletionMode.Verify)
{
// TODO: Check with failover team to see how to confirm that the replica actually restarted. We do not expose instance id for persisted replicas
}
action.Result = new RestartReplicaResult(replicaSelectorResult);
this.ResultTraceString = StringHelper.Format(
"RestartReplicaOrInstance succeeded by restarting replica {0}:{1} node {2} with CompletionMode {3}",
partitionId.Value,
replicaId.Value,
nodeName,
action.CompletionMode);
}
protected override async Task ExecuteActionAsync(FabricTestContext testContext, RestartNodeAction action, CancellationToken cancellationToken)
{
ActionTraceSource.WriteInfo(TraceSource, "Enter RestartNodeAction/ExecuteActionAsync: operationTimeout='{0}', requestTimeout='{1}'", action.ActionTimeout, action.RequestTimeout);
this.helper = new TimeoutHelper(action.ActionTimeout);
SelectedReplica selectedReplica = SelectedReplica.None;
string nodeName = action.NodeName;
BigInteger nodeInstance = action.NodeInstance;
bool createFabricDump = action.CreateFabricDump;
if (string.IsNullOrEmpty(nodeName))
{
ThrowIf.Null(action.ReplicaSelector, "ReplicaSelector");
var getReplicaStateAction = new GetSelectedReplicaStateAction(action.ReplicaSelector)
{
RequestTimeout = action.RequestTimeout,
ActionTimeout = helper.GetRemainingTime()
};
await testContext.ActionExecutor.RunAsync(getReplicaStateAction, cancellationToken).ConfigureAwait(false);
var replicaStateActionResult = getReplicaStateAction.Result;
ReleaseAssert.AssertIf(replicaStateActionResult == null, "replicaStateActionResult cannot be null");
selectedReplica = replicaStateActionResult.Item1;
Replica replicaStateResult = replicaStateActionResult.Item2;
ReleaseAssert.AssertIf(replicaStateResult == null, "replicaStateResult cannot be null");
nodeName = replicaStateResult.NodeName;
nodeInstance = BigInteger.MinusOne;
}
if (nodeInstance == BigInteger.MinusOne)
{
var nodeInfo = await this.GetCurrentNodeInfoAsync(testContext, nodeName, action, cancellationToken).ConfigureAwait(false);
nodeInstance = nodeInfo.NodeInstanceId;
}
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.FaultManager.RestartNodeUsingNodeNameAsync(
nodeName,
nodeInstance,
createFabricDump,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
if (action.CompletionMode == CompletionMode.Verify)
{
bool success = false;
while (this.helper.GetRemainingTime() > TimeSpan.Zero)
{
var nodeInfo = await this.GetCurrentNodeInfoAsync(testContext, nodeName, action, cancellationToken).ConfigureAwait(false);
if (nodeInfo.NodeInstanceId > nodeInstance && nodeInfo.IsNodeUp)
{
success = true;
break;
}
ActionTraceSource.WriteInfo(TraceSource, "NodeName = {0} not yet restarted. '{1}' seconds remain. Retrying...", nodeName, this.helper.GetRemainingTime().TotalSeconds);
await AsyncWaiter.WaitAsync(TimeSpan.FromSeconds(5), cancellationToken);
}
if (!success)
{
throw new TimeoutException(StringHelper.Format(StringResources.Error_TestabilityActionTimeout,
"RestartNode",
nodeName));
}
}
// create result
action.Result = new RestartNodeResult(selectedReplica, new NodeResult(nodeName, nodeInstance));
ResultTraceString = StringHelper.Format("RestartNodeAction succeeded for {0}:{1} with CompletionMode = {2}", nodeName, nodeInstance, action.CompletionMode);
}