protected override async Task ExecuteActionAsync(FabricTestContext testContext, RestartPartitionAction action, CancellationToken cancellationToken)
{
ThrowIf.Null(action.PartitionSelector, "partitionSelector");
this.helper = new TimeoutHelper(action.ActionTimeout);
// get service info so we can validate if the operation is valid
ServiceDescription result = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.ServiceManager.GetServiceDescriptionAsync(
action.PartitionSelector.ServiceName,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
if (result.Kind != ServiceDescriptionKind.Stateful && action.RestartPartitionMode == RestartPartitionMode.OnlyActiveSecondaries)
{
throw new InvalidOperationException(StringHelper.Format(StringResources.Error_InvalidServiceTypeTestability, "RestartPartitionMode.OnlyActiveSecondaries", "Stateful", action.PartitionSelector.ServiceName, "Stateless"));
}
bool hasPersistedState = false;
if (result.Kind == ServiceDescriptionKind.Stateful)
{
StatefulServiceDescription statefulDescription = result as StatefulServiceDescription;
ReleaseAssert.AssertIf(statefulDescription == null, "Stateful service description is not WinFabricStatefulServiceDescription");
hasPersistedState = statefulDescription.HasPersistedState;
}
// now actually select a partition
var getPartitionStateAction = new GetSelectedPartitionStateAction(action.PartitionSelector)
{
RequestTimeout = action.RequestTimeout,
ActionTimeout = helper.GetRemainingTime()
};
await testContext.ActionExecutor.RunAsync(getPartitionStateAction, cancellationToken);
Guid partitionId = getPartitionStateAction.Result.PartitionId;
// get replicas for target
ServiceReplicaList replicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetReplicaListAsync(
partitionId,
0,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
// get replicas for fm in order to get the primary
ServiceReplicaList fmReplicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetReplicaListAsync(
Constants.FmPartitionId,
0,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
string fmPrimaryNodeName = string.Empty;
var readyFMReplicas = fmReplicasResult.Where(r => r.ReplicaStatus == ServiceReplicaStatus.Ready).ToArray();
foreach (var replica in readyFMReplicas)
{
StatefulServiceReplica statefulReplica = replica as StatefulServiceReplica;
ReleaseAssert.AssertIf(statefulReplica == null, "FM Replica is not a stateful replica");
if (statefulReplica.ReplicaRole == ReplicaRole.Primary)
{
fmPrimaryNodeName = replica.NodeName;
}
}
if (string.IsNullOrEmpty(fmPrimaryNodeName))
{
throw new FabricException(StringHelper.Format(StringResources.Error_PartitionPrimaryNotReady, "FailoverManager"), FabricErrorCode.NotReady);
}
////------------------------------------------------------
// target ut at the fm primary only
UnreliableTransportBehavior behavior = new UnreliableTransportBehavior("*", "DoReconfiguration");
behavior.AddFilterForPartitionId(partitionId);
string behaviorName = "BlockDoReconfiguration";
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.TestManager.AddUnreliableTransportBehaviorAsync(
fmPrimaryNodeName,
behaviorName,
behavior,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
// TODO: Wait for some time so that the unreliable transport behavior can be read from the files.
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successfully applied
await Task.Delay(TimeSpan.FromSeconds(5.0), cancellationToken).ConfigureAwait(false);
bool triedToRemovedBehavior = false;
// inspect the actual replicas to restart, only operate on stable ones
try
{
var stableReplicasToRestart = replicasResult.Where(r => r.ReplicaStatus == ServiceReplicaStatus.Ready).ToArray();
foreach (var replica in stableReplicasToRestart)
{
var currentReplica = replica;
if (action.RestartPartitionMode == RestartPartitionMode.OnlyActiveSecondaries)
{
StatefulServiceReplica statefulReplica = currentReplica as StatefulServiceReplica;
ReleaseAssert.AssertIf(statefulReplica == null, "Stateful service replica is not StatefulServiceReplica");
if (statefulReplica.ReplicaRole == ReplicaRole.Primary)
{
continue;
}
}
if (hasPersistedState)
{
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.FaultManager.RestartReplicaAsync(
currentReplica.NodeName,
partitionId,
currentReplica.Id,
CompletionMode.DoNotVerify,
action.RequestTimeout.TotalSeconds,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
}
else
{
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.FaultManager.RemoveReplicaAsync(
currentReplica.NodeName,
partitionId,
currentReplica.Id,
CompletionMode.DoNotVerify,
false, /*force remove*/
action.RequestTimeout.TotalSeconds,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
}
}
triedToRemovedBehavior = true;
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.TestManager.RemoveUnreliableTransportBehaviorAsync(
fmPrimaryNodeName,
behaviorName,
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.RemoveUnreliableTransportBehaviorErrors.Value,
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
// TODO: Wait for some time so that the unreliable transport behavior can be read from the files.
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successfully applied
await Task.Delay(TimeSpan.FromSeconds(5.0)).ConfigureAwait(false);
}
finally
{
// TODO: Provide a way to clear all behaviors just in case.
if (!triedToRemovedBehavior)
{
ActionTraceSource.WriteWarning(TraceType, "Exception after adding behavior to block messages. Removing behavior synchronously");
FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.TestManager.RemoveUnreliableTransportBehaviorAsync(
fmPrimaryNodeName,
behaviorName,
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.RemoveUnreliableTransportBehaviorErrors.Value,
this.helper.GetRemainingTime(),
cancellationToken).GetAwaiter().GetResult();
// TODO: Wait for some time so that the unreliable transport behavior can be read from the files.
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successfully applied
Task.Delay(TimeSpan.FromSeconds(5.0)).GetAwaiter().GetResult();
}
}
// -- note there's no explict validation
// action result
action.Result = new RestartPartitionResult(getPartitionStateAction.Result);
ResultTraceString = StringHelper.Format("RestartPartitionAction succeeded for {0} with RestartPartitionMode = {1}", partitionId, action.RestartPartitionMode);
}
protected override async Task ExecuteActionAsync(FabricTestContext testContext, 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);
}
