protected override async Task ExecuteActionAsync(FabricTestContext testContext, StartNodeAction action, CancellationToken cancellationToken)
{
ThrowIf.Null(action.NodeName, "NodeName");
this.helper = new TimeoutHelper(action.ActionTimeout);
string nodeName = action.NodeName;
BigInteger nodeInstance = action.NodeInstance;
if (nodeInstance == BigInteger.MinusOne)
{
var nodeInfo = await GetCurrentNodeInfoAsync(testContext, action, cancellationToken);
if (nodeInfo == null)
{
throw new FabricException(StringResources.Error_NodeNotFound, FabricErrorCode.NodeNotFound);
}
nodeInstance = nodeInfo.NodeInstanceId;
}
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.FaultManager.StartNodeUsingNodeNameAsync(
nodeName,
nodeInstance,
action.IPAddressOrFQDN,
action.ClusterConnectionPort,
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, action, cancellationToken).ConfigureAwait(false);
if (nodeInfo != null && nodeInfo.NodeInstanceId > nodeInstance && nodeInfo.IsNodeUp)
{
success = true;
break;
}
ActionTraceSource.WriteInfo(TraceSource, "NodeName = {0} not yet Started. Retrying...", action.NodeName);
await AsyncWaiter.WaitAsync(TimeSpan.FromSeconds(5), cancellationToken).ConfigureAwait(false);
}
if (!success)
{
throw new TimeoutException(StringHelper.Format(StringResources.Error_TestabilityActionTimeout,
"StartNode",
action.NodeName + ":" + action.NodeInstance));
}
}
action.Result = new StartNodeResult(action.NodeName, nodeInstance);
this.ResultTraceString = StringHelper.Format("StartNodeAction succeeded for {0}:{1} with CompletionMode = {2}", action.NodeName, nodeInstance, action.CompletionMode);
}
protected override async Task ExecuteActionAsync(
FabricTestContext testContext,
CleanTestStateAction action,
CancellationToken cancellationToken)
{
this.helper = new TimeoutHelper(action.ActionTimeout);
var nodes = await testContext.FabricCluster.GetLatestNodeInfoAsync(action.RequestTimeout, this.helper.GetRemainingTime(), cancellationToken).ConfigureAwait(false);
foreach (var nodeInfo in nodes)
{
if (nodeInfo.IsNodeUp)
{
var info = nodeInfo;
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.TestManager.RemoveUnreliableTransportBehaviorAsync(
info.NodeName,
"*",
action.RequestTimeout,
cancellationToken),
this.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);
ActionTraceSource.WriteInfo(TraceType, "Test state cleaned for node:{0}", nodeInfo.NodeName);
}
else
{
ActionTraceSource.WriteInfo(TraceType, "Test clean failed to start node {0}", nodeInfo.NodeName);
}
}
ResultTraceString = StringHelper.Format("CleanTestStateAction succeeded");
}
protected override async Task ExecuteActionAsync(FabricTestContext testContext, RestartPartitionAction action, CancellationToken cancellationToken)
{
ThrowIf.Null(action.PartitionSelector, "partitionSelector");
this.helper = new TimeoutHelper(action.ActionTimeout);
// get service info so we can validate if the operation is valid
ServiceDescription result = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.ServiceManager.GetServiceDescriptionAsync(
action.PartitionSelector.ServiceName,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
if (result.Kind != ServiceDescriptionKind.Stateful && action.RestartPartitionMode == RestartPartitionMode.OnlyActiveSecondaries)
{
throw new InvalidOperationException(StringHelper.Format(StringResources.Error_InvalidServiceTypeTestability, "RestartPartitionMode.OnlyActiveSecondaries", "Stateful", action.PartitionSelector.ServiceName, "Stateless"));
}
bool hasPersistedState = false;
if (result.Kind == ServiceDescriptionKind.Stateful)
{
StatefulServiceDescription statefulDescription = result as StatefulServiceDescription;
ReleaseAssert.AssertIf(statefulDescription == null, "Stateful service description is not WinFabricStatefulServiceDescription");
hasPersistedState = statefulDescription.HasPersistedState;
}
// now actually select a partition
var getPartitionStateAction = new GetSelectedPartitionStateAction(action.PartitionSelector)
{
RequestTimeout = action.RequestTimeout,
ActionTimeout = helper.GetRemainingTime()
};
await testContext.ActionExecutor.RunAsync(getPartitionStateAction, cancellationToken);
Guid partitionId = getPartitionStateAction.Result.PartitionId;
// get replicas for target
ServiceReplicaList replicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetReplicaListAsync(
partitionId,
0,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
// get replicas for fm in order to get the primary
ServiceReplicaList fmReplicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetReplicaListAsync(
Constants.FmPartitionId,
0,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
string fmPrimaryNodeName = string.Empty;
var readyFMReplicas = fmReplicasResult.Where(r => r.ReplicaStatus == ServiceReplicaStatus.Ready).ToArray();
foreach (var replica in readyFMReplicas)
{
StatefulServiceReplica statefulReplica = replica as StatefulServiceReplica;
ReleaseAssert.AssertIf(statefulReplica == null, "FM Replica is not a stateful replica");
if (statefulReplica.ReplicaRole == ReplicaRole.Primary)
{
fmPrimaryNodeName = replica.NodeName;
}
}
if (string.IsNullOrEmpty(fmPrimaryNodeName))
{
throw new FabricException(StringHelper.Format(StringResources.Error_PartitionPrimaryNotReady, "FailoverManager"), FabricErrorCode.NotReady);
}
////------------------------------------------------------
// target ut at the fm primary only
UnreliableTransportBehavior behavior = new UnreliableTransportBehavior("*", "DoReconfiguration");
behavior.AddFilterForPartitionId(partitionId);
string behaviorName = "BlockDoReconfiguration";
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.TestManager.AddUnreliableTransportBehaviorAsync(
fmPrimaryNodeName,
behaviorName,
behavior,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
// TODO: Wait for some time so that the unreliable transport behavior can be read from the files.
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successfully applied
await Task.Delay(TimeSpan.FromSeconds(5.0), cancellationToken).ConfigureAwait(false);
bool triedToRemovedBehavior = false;
// inspect the actual replicas to restart, only operate on stable ones
try
{
var stableReplicasToRestart = replicasResult.Where(r => r.ReplicaStatus == ServiceReplicaStatus.Ready).ToArray();
foreach (var replica in stableReplicasToRestart)
{
var currentReplica = replica;
if (action.RestartPartitionMode == RestartPartitionMode.OnlyActiveSecondaries)
{
StatefulServiceReplica statefulReplica = currentReplica as StatefulServiceReplica;
ReleaseAssert.AssertIf(statefulReplica == null, "Stateful service replica is not StatefulServiceReplica");
if (statefulReplica.ReplicaRole == ReplicaRole.Primary)
{
continue;
}
}
if (hasPersistedState)
{
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.FaultManager.RestartReplicaAsync(
currentReplica.NodeName,
partitionId,
currentReplica.Id,
CompletionMode.DoNotVerify,
action.RequestTimeout.TotalSeconds,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
}
else
{
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.FaultManager.RemoveReplicaAsync(
currentReplica.NodeName,
partitionId,
currentReplica.Id,
CompletionMode.DoNotVerify,
false, /*force remove*/
action.RequestTimeout.TotalSeconds,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
}
}
triedToRemovedBehavior = true;
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.TestManager.RemoveUnreliableTransportBehaviorAsync(
fmPrimaryNodeName,
behaviorName,
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.RemoveUnreliableTransportBehaviorErrors.Value,
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
// TODO: Wait for some time so that the unreliable transport behavior can be read from the files.
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successfully applied
await Task.Delay(TimeSpan.FromSeconds(5.0)).ConfigureAwait(false);
}
finally
{
// TODO: Provide a way to clear all behaviors just in case.
if (!triedToRemovedBehavior)
{
ActionTraceSource.WriteWarning(TraceType, "Exception after adding behavior to block messages. Removing behavior synchronously");
FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.TestManager.RemoveUnreliableTransportBehaviorAsync(
fmPrimaryNodeName,
behaviorName,
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.RemoveUnreliableTransportBehaviorErrors.Value,
this.helper.GetRemainingTime(),
cancellationToken).GetAwaiter().GetResult();
// TODO: Wait for some time so that the unreliable transport behavior can be read from the files.
// Bug#2271465 - Unreliable transport through API should return only once the behavior has been successfully applied
Task.Delay(TimeSpan.FromSeconds(5.0)).GetAwaiter().GetResult();
}
}
// -- note there's no explict validation
// action result
action.Result = new RestartPartitionResult(getPartitionStateAction.Result);
ResultTraceString = StringHelper.Format("RestartPartitionAction succeeded for {0} with RestartPartitionMode = {1}", partitionId, action.RestartPartitionMode);
}
protected override async Task ExecuteActionAsync(FabricTestContext testContext, MoveSecondaryAction action, CancellationToken cancellationToken)
{
ThrowIf.Null(action.PartitionSelector, "PartitionSelector");
this.helper = new TimeoutHelper(action.ActionTimeout);
string newSecondaryNode = action.NewSecondaryNodeName;
string currentSecondaryNode = action.CurrentSecondaryNodeName;
var getPartitionStateAction = new GetSelectedPartitionStateAction(action.PartitionSelector)
{
RequestTimeout = action.RequestTimeout,
ActionTimeout = this.helper.GetRemainingTime()
};
await testContext.ActionExecutor.RunAsync(getPartitionStateAction, cancellationToken).ConfigureAwait(false);
Guid partitionId = getPartitionStateAction.Result.PartitionId;
if (!action.IgnoreConstraints)
{
// get current primary replica node name.
ServiceReplicaList replicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetReplicaListAsync(
partitionId,
0,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
string currentPrimaryNodeInfo = string.Empty;
List <string> currentSecReplicaNodes = new List <string>();
foreach (var replica in replicasResult)
{
StatefulServiceReplica statefulReplica = replica as StatefulServiceReplica;
if (statefulReplica == null)
{
throw new InvalidOperationException(StringHelper.Format(StringResources.Error_InvalidServiceTypeTestability, "MoveSecondary", "Stateful", action.PartitionSelector.ServiceName, "Stateless"));
}
if (statefulReplica.ReplicaRole == ReplicaRole.Primary)
{
currentPrimaryNodeInfo = statefulReplica.NodeName;
if (!string.IsNullOrEmpty(newSecondaryNode) && newSecondaryNode == statefulReplica.NodeName)
{
throw new FabricException(
StringHelper.Format(StringResources.Error_InvalidNodeNameProvided, newSecondaryNode, "MoveSecondary", "Primary exists on node"),
FabricErrorCode.AlreadyPrimaryReplica);
}
}
else if (statefulReplica.ReplicaRole == ReplicaRole.ActiveSecondary)
{
currentSecReplicaNodes.Add(statefulReplica.NodeName);
if (!string.IsNullOrEmpty(newSecondaryNode) && newSecondaryNode == statefulReplica.NodeName)
{
throw new FabricException(
StringHelper.Format(StringResources.Error_InvalidNodeNameProvided, newSecondaryNode, "MoveSecondary", "Secondary exists on node"),
FabricErrorCode.AlreadySecondaryReplica);
}
}
}
if (currentSecReplicaNodes.Count == 0)
{
throw new InvalidOperationException(StringResources.Error_NoSecondariesInReplicaSet);
}
if (string.IsNullOrEmpty(currentSecondaryNode))
{
int num = testContext.Random.Next(currentSecReplicaNodes.Count);
currentSecondaryNode = currentSecReplicaNodes.ElementAt(num);
}
if (!currentSecReplicaNodes.Contains(currentSecondaryNode))
{
throw new FabricException(
StringHelper.Format(StringResources.Error_InvalidNodeNameProvided, newSecondaryNode, "MoveSecondary", "Current node does not have a secondary replica"),
FabricErrorCode.InvalidReplicaStateForReplicaOperation);
}
}
ReleaseAssert.AssertIf(string.IsNullOrEmpty(currentSecondaryNode), "Current node name cannot be null or empty.");
ReleaseAssert.AssertIf(newSecondaryNode == currentSecondaryNode, "Current and New node names are same.");
ActionTraceSource.WriteInfo(TraceSource, "Calling move secondary with current node {0}, new node {1}, partition {2}", currentSecondaryNode, string.IsNullOrEmpty(newSecondaryNode) ? "Random" : newSecondaryNode, partitionId);
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.FaultManager.MoveSecondaryUsingNodeNameAsync(
currentSecondaryNode,
newSecondaryNode,
getPartitionStateAction.Result.ServiceName,
partitionId,
action.IgnoreConstraints,
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.MoveSecondaryFabricErrors.Value,
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
action.Result = new MoveSecondaryResult(currentSecondaryNode, newSecondaryNode, getPartitionStateAction.Result);
this.ResultTraceString = StringHelper.Format(
"MoveSecondaryAction succeeded for moving Primary for {0} from {1} to {2}.",
partitionId,
currentSecondaryNode,
newSecondaryNode);
}
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);
}
protected override async Task ExecuteActionAsync(FabricTestContext testContext, RestartDeployedCodePackageAction action, CancellationToken cancellationToken)
{
this.helper = new TimeoutHelper(action.ActionTimeout);
string nodeName = action.NodeName;
Uri applicationName = action.ApplicationName;
string serviceManifestName = action.ServiceManifestName;
string servicePackageActivationId = action.ServicePackageActivationId;
string codePackageName = action.CodePackageName;
SelectedReplica replicaSelectorResult = SelectedReplica.None;
ThrowIf.Null(applicationName, "ApplicationName");
if (string.IsNullOrEmpty(nodeName) ||
string.IsNullOrEmpty(serviceManifestName) ||
string.IsNullOrEmpty(codePackageName))
{
ThrowIf.Null(action.ReplicaSelector, "ReplicaSelector");
var getReplicaStateAction = new GetSelectedReplicaStateAction(action.ReplicaSelector)
{
RequestTimeout = action.RequestTimeout,
ActionTimeout = this.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;
ReleaseAssert.AssertIf(replicaSelectorResult == null || replicaSelectorResult.SelectedPartition == null,
"replicaSelectorResult cannot be null or for a non-null replicaSelectorResult, the selected partition must be non-null");
Guid partitionId = replicaStateActionResult.Item1.SelectedPartition.PartitionId;
Replica replicaStateResult = replicaStateActionResult.Item2;
ReleaseAssert.AssertIf(replicaStateResult == null, "replicaStateResult cannot be null");
nodeName = replicaStateResult.NodeName;
var deployedReplicaListResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync <DeployedServiceReplicaList>(
() => testContext.FabricClient.QueryManager.GetDeployedReplicaListAsync(
nodeName,
applicationName,
null,
partitionId,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
DeployedServiceReplica selectedReplica = deployedReplicaListResult.FirstOrDefault(r => r.Partitionid == partitionId);
if (selectedReplica == null)
{
throw new FabricException(
StringHelper.Format(StringResources.Error_DidNotFindDeployedReplicaOnNode, partitionId, nodeName),
FabricErrorCode.ReplicaDoesNotExist);
}
serviceManifestName = selectedReplica.ServiceManifestName;
servicePackageActivationId = selectedReplica.ServicePackageActivationId;
codePackageName = selectedReplica.CodePackageName;
}
ActionTraceSource.WriteInfo(TraceSource, "SelectedReplica: serviceManifestName: {0}, servicePackageActivationId: {1}, codePackageName: {2}", serviceManifestName, servicePackageActivationId, codePackageName);
DeployedCodePackage deployedCodePackageListResult = await this.GetCodePackageInfoAsync(testContext, nodeName, applicationName, serviceManifestName, servicePackageActivationId, codePackageName, action, cancellationToken).ConfigureAwait(false);
var codepackageEntrypointToRestart = GetCodepackageEntrypointToRestart(action, deployedCodePackageListResult);
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.FaultManager.RestartDeployedCodePackageUsingNodeNameAsync(
nodeName,
applicationName,
serviceManifestName,
servicePackageActivationId,
codePackageName,
codepackageEntrypointToRestart.EntryPoint.CodePackageInstanceId,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
if (action.CompletionMode == CompletionMode.Verify)
{
bool success = false;
while (this.helper.GetRemainingTime() > TimeSpan.Zero)
{
var deployedCodePackageListResultAfterRestart = await this.GetCodePackageInfoAsync(testContext, nodeName, applicationName, serviceManifestName, servicePackageActivationId, codePackageName, action, cancellationToken).ConfigureAwait(false);
if (deployedCodePackageListResultAfterRestart != null)
{
var entryPointAfterRestart = codepackageEntrypointToRestart.EntryPointType == EntryPointType.Main ? deployedCodePackageListResultAfterRestart.EntryPoint : deployedCodePackageListResultAfterRestart.SetupEntryPoint;
if (entryPointAfterRestart != null && entryPointAfterRestart.CodePackageInstanceId > codepackageEntrypointToRestart.EntryPoint.CodePackageInstanceId && entryPointAfterRestart.EntryPointStatus == EntryPointStatus.Started)
{
success = true;
break;
}
}
ActionTraceSource.WriteInfo(TraceSource, "CodePackage = {0}:{1}:{2} not yet restarted. Retrying...", nodeName, applicationName, codePackageName);
await AsyncWaiter.WaitAsync(TimeSpan.FromSeconds(5), cancellationToken).ConfigureAwait(false);
}
if (!success)
{
throw new TimeoutException(StringHelper.Format(StringResources.Error_TestabilityActionTimeout,
"RestartDeployedCodePackage",
applicationName));
}
}
action.Result = new RestartDeployedCodePackageResult(
nodeName,
applicationName,
serviceManifestName,
servicePackageActivationId,
codePackageName,
codepackageEntrypointToRestart.EntryPoint.CodePackageInstanceId,
replicaSelectorResult);
ResultTraceString = StringHelper.Format("RestartCodePackageAction succeeded for {0}:{1}:{2} with CompletionMode = {3}", nodeName, applicationName, codePackageName, 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);
}
protected override async Task ExecuteActionAsync(FabricTestContext testContext, MovePrimaryAction action, CancellationToken cancellationToken)
{
ThrowIf.Null(action.PartitionSelector, "PartitionSelector");
this.helper = new TimeoutHelper(action.ActionTimeout);
string newPrimaryNodeName = action.NodeName;
var getPartitionStateAction = new GetSelectedPartitionStateAction(action.PartitionSelector)
{
RequestTimeout = action.RequestTimeout,
ActionTimeout = this.helper.GetRemainingTime()
};
await testContext.ActionExecutor.RunAsync(getPartitionStateAction, cancellationToken);
Guid partitionId = getPartitionStateAction.Result.PartitionId;
if (!action.IgnoreConstraints)
{
// select random node where replica's primary not present
var nodesInfo = await testContext.FabricCluster.GetLatestNodeInfoAsync(action.RequestTimeout, this.helper.GetRemainingTime(), cancellationToken);
if ((nodesInfo == null || nodesInfo.Count() == 0))
{
throw new InvalidOperationException(StringHelper.Format(StringResources.Error_NotEnoughNodesForTestabilityAction, "MovePrimary"));
}
ServiceReplicaList replicasResult = await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.QueryManager.GetReplicaListAsync(
partitionId,
0,
action.RequestTimeout,
cancellationToken),
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
NodeInfo currentPrimaryNodeInfo = null;
string currentPrimaryNodeName = string.Empty;
foreach (var replica in replicasResult)
{
StatefulServiceReplica statefulReplica = replica as StatefulServiceReplica;
if (statefulReplica == null)
{
throw new InvalidOperationException(StringHelper.Format(StringResources.Error_InvalidServiceTypeTestability, "MovePrimary", "Stateful", action.PartitionSelector.ServiceName, "Stateless"));
}
if (statefulReplica.ReplicaRole == ReplicaRole.Primary)
{
currentPrimaryNodeInfo = nodesInfo.FirstOrDefault(n => n.NodeName == statefulReplica.NodeName);
if (!string.IsNullOrEmpty(newPrimaryNodeName) && newPrimaryNodeName == statefulReplica.NodeName)
{
throw new FabricException(
StringHelper.Format(StringResources.Error_InvalidNodeNameProvided, newPrimaryNodeName, "MovePrimary", "Primary already exists on node"),
FabricErrorCode.AlreadyPrimaryReplica);
}
break;
}
}
if (currentPrimaryNodeInfo == null)
{
throw new FabricException(StringHelper.Format(StringResources.Error_PartitionPrimaryNotReady, action.PartitionSelector + ":" + partitionId), FabricErrorCode.NotReady);
}
currentPrimaryNodeName = currentPrimaryNodeInfo.NodeName;
if (newPrimaryNodeName == currentPrimaryNodeName)
{
throw new FabricException(
StringHelper.Format(StringResources.Error_InvalidNodeNameProvided, newPrimaryNodeName, "MovePrimary", "Primary already exists on node"),
FabricErrorCode.AlreadyPrimaryReplica);
}
}
ActionTraceSource.WriteInfo(TraceSource, "Calling move primary with node {0}, partition {1}", string.IsNullOrEmpty(newPrimaryNodeName) ? "Random" : newPrimaryNodeName, partitionId);
await FabricClientRetryHelper.ExecuteFabricActionWithRetryAsync(
() => testContext.FabricClient.FaultManager.MovePrimaryUsingNodeNameAsync(
newPrimaryNodeName,
getPartitionStateAction.Result.ServiceName,
partitionId,
action.IgnoreConstraints,
action.RequestTimeout,
cancellationToken),
FabricClientRetryErrors.MovePrimaryFabricErrors.Value,
this.helper.GetRemainingTime(),
cancellationToken).ConfigureAwait(false);
action.Result = new MovePrimaryResult(newPrimaryNodeName, getPartitionStateAction.Result);
ResultTraceString = StringHelper.Format("MovePrimaryAction succeeded for moving Primary for {0} to node {1}.", partitionId, newPrimaryNodeName);
}
