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, 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, 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, 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);
}
