/// <summary>
/// Prepares the claimed repair tasks belonging to POS as per the TaskApprovalPolicy
/// </summary>
/// <returns>Task for the asynchronous operation</returns>
internal async Task PrepareRepairTasks(CancellationToken cancellationToken)
{
NodeList nodeList = await this.fabricClient.QueryManager.GetNodeListAsync(null, null, this.DefaultTimeoutForOperation, cancellationToken);
IList <RepairTask> claimedTaskList = await this.GetClaimedRepairTasks(nodeList, cancellationToken);
switch (RmPolicy)
{
case TaskApprovalPolicy.NodeWise:
{
RepairTaskList processingTaskList = await this.GetRepairTasksUnderProcessing(cancellationToken);
if (!processingTaskList.Any())
{
if (claimedTaskList.Any())
{
RepairTask oldestClaimedTask = claimedTaskList.Aggregate(
(curMin, task) => (task.CreatedTimestamp < curMin.CreatedTimestamp ? task : curMin));
ServiceEventSource.Current.VerboseMessage(
"Out of {0} claimed tasks, Oldest repair task = {0} with node = {1} will be prepared",
claimedTaskList.Count, oldestClaimedTask.TaskId, oldestClaimedTask.Target);
this.StartPreparingRepairTask(oldestClaimedTask);
}
}
break;
}
case TaskApprovalPolicy.UpgradeDomainWise:
{
string currentUpgradeDomain = await this.GetCurrentUpgradeDomainUnderProcessing(nodeList, cancellationToken);
ServiceEventSource.Current.VerboseMessage(String.Format("{0} repair tasks were found in claimed state", claimedTaskList.Count));
// Below line can be enabled for debugging
// rmHelper.PrintRepairTasks(claimedTaskList);
foreach (var claimedTask in claimedTaskList)
{
string udName = this.GetUpgradeDomainOfRepairTask(claimedTask, nodeList);
if (string.IsNullOrEmpty(currentUpgradeDomain))
{
currentUpgradeDomain = udName;
}
if (udName == currentUpgradeDomain)
{
this.StartPreparingRepairTask(claimedTask);
}
}
break;
}
default:
{
string errorMessage = String.Format("Illegal RmPolicy found: {0}", RmPolicy);
ServiceEventSource.Current.ErrorMessage(errorMessage);
throw new InvalidOperationException(errorMessage);
}
}
}
public async Task <IList <IRepairTask> > GetRepairTaskListAsync(
Guid activityId,
string taskIdFilter = null,
RepairTaskStateFilter stateFilter = RepairTaskStateFilter.Default,
string executorFilter = null)
{
//var startTime = DateTimeOffset.UtcNow;
try
{
// TODO, using the overload without timeout and cancellation token for now since there is some max timeout limit
// being exercised somewhere. if timeout provided is more than that, repair task creation fails
RepairTaskList repairTaskList = await repairManager.GetRepairTaskListAsync(taskIdFilter, stateFilter, executorFilter).ConfigureAwait(false);
var repairTasks = new List <IRepairTask>(repairTaskList.Count);
foreach (var repairTask in repairTaskList)
{
repairTasks.Add(new ServiceFabricRepairTask(repairTask));
}
//activityLogger.LogOperation(activityId, startTime);
return(repairTasks);
}
catch (Exception ex)
{
traceType.WriteWarning("Unable to get repair task list. Errors: {0}", ex.GetMessage());
//activityLogger.LogOperation(activityId, startTime, OperationResult.Failure, ex);
throw;
}
}
/// <summary>
/// Gets the upgrade domain from current repair tasks under processing, ideally all the repair tasks under processing should've the same upgrade domain.
/// However if repair tasks belonging to multiple UpgradeDomains are found, then we consider the UD of first repair task in the list of repair tasks.
/// </summary>
/// <param name="nodeList">List of Nodes currently in cluster, used to get nodename and upgradedomain mapping</param>
/// <returns>Upgrade domain of the first repair task among the list of repair tasks under processing</returns>
private async Task <string> GetCurrentUpgradeDomainUnderProcessing(NodeList nodeList, CancellationToken cancellationToken)
{
string currentUpgradeDomain = null;
RepairTaskList processingTaskList = await this.GetRepairTasksUnderProcessing(cancellationToken);
ServiceEventSource.Current.VerboseMessage(String.Format("{0} repair tasks were found under processing",
processingTaskList.Count));
// All the tasks under processing should ideally be from the same UpgradeDomain.
// However in case the cluster topology has changed or repair tasks were manually created from some other entity resulting in multiple repair tasks with target nodes belonging to different UD's.
// In that case we'll consider the upgrade domain of the first repair task among the list of repair tasks which are already under processing.
foreach (var task in processingTaskList)
{
string udName = this.GetUpgradeDomainOfRepairTask(task, nodeList);
if (string.IsNullOrEmpty(currentUpgradeDomain))
{
currentUpgradeDomain = udName;
}
else if (currentUpgradeDomain != udName)
{
ServiceEventSource.Current.ErrorMessage(
String.Format(
"Found repair task {0} under processing belonging to UpgradeDomains {1}, Expected only repair tasks from {2}. cluster topology might've changed",
task.TaskId, currentUpgradeDomain, udName));
}
}
return(currentUpgradeDomain);
}
public void Onebox_Successful_SelfRepairAfterTaskIsApproved()
{
const string property = "Onebox_Successful_SelfRepairAfterTaskIsCreated";
MockRepairExecutor executor = Executors.First();
LogHelper.Log("Starting test {0}", property);
Node node = GetHealthyNode(_fabricClient);
// Inject error
MockWatchdog watchdog = new MockWatchdog(_fabricClient.HealthManager, node.NodeName);
watchdog.ReportError(property);
RepairTask createdTask = PollingForCreatedTask(executor, node);
Assert.IsNotNull(createdTask, "There must be a repair task created.");
executor.Approve(ref createdTask);
ClearHealthError(watchdog, property, executor);
// Make sure the task is cancelled
Thread.Sleep(_repairManagerLatency);
RepairTaskList repairTaskList = _fabricClient.RepairManager.GetRepairTaskListAsync(createdTask.TaskId, RepairTaskStateFilter.All, null).Result;
Assert.IsTrue((repairTaskList.Any() && repairTaskList.Count == 1), "There must be a unique repair task with task id = {0}", createdTask.TaskId);
RepairTask cancelledTask = repairTaskList.First();
Assert.IsTrue(cancelledTask.Flags == RepairTaskFlags.CancelRequested, "PE should request cancellation of the task");
}
/// <summary>
/// This function returns the list of repair tasks which are undergoing work.
/// At any point of time there will be only one UD which will have POS repair tasks in these states.
/// </summary>
/// <returns>List of repair tasks in Preparing, Approved, Executing or Restoring state</returns>
internal async Task <RepairTaskList> GetRepairTasksUnderProcessing(CancellationToken cancellationToken)
{
RepairTaskList repairTasks = await this.fabricClient.RepairManager.GetRepairTaskListAsync(TaskIdPrefix,
RepairTaskStateFilter.Preparing | RepairTaskStateFilter.Approved | RepairTaskStateFilter.Executing |
RepairTaskStateFilter.Restoring,
ExecutorName, this.DefaultTimeoutForOperation, cancellationToken);
return(repairTasks);
}
/// <summary>
/// Prints the repair task for ease of debugging
/// </summary>
internal async Task PrintRepairTasks(CancellationToken cancellationToken)
{
RepairTaskList repairTasks = await this.fabricClient.RepairManager.GetRepairTaskListAsync(TaskIdPrefix,
RepairTaskStateFilter.All,
ExecutorName, this.DefaultTimeoutForOperation, cancellationToken);
ServiceEventSource.Current.VerboseMessage("Total {0} repair tasks were found for POS", repairTasks.Count);
foreach (var task in repairTasks)
{
ServiceEventSource.Current.PrintRepairTasks(task.TaskId, task.State.ToString(), task.Action, task.Executor,
task.Description, task.ExecutorData, task.Target.ToString());
}
}
/// <summary>
/// Posts the cluster patching status by finding the nodes on which patching is going on.
/// </summary>
private async Task PostRMTaskNodeUpdate(CancellationToken cancellationToken)
{
NodeList nodeList = await this.fabricClient.QueryManager.GetNodeListAsync(null, null, this.DefaultTimeoutForOperation, cancellationToken);
HashSet <string> processingNodes = new HashSet <string>();
HashSet <string> pendingNodes = new HashSet <string>();
IList <RepairTask> claimedTaskList = await this.GetClaimedRepairTasks(nodeList, cancellationToken);
foreach (var task in claimedTaskList)
{
pendingNodes.Add(task.Target.ToString());
}
cancellationToken.ThrowIfCancellationRequested();
RepairTaskList processingTaskListFinal = await this.GetRepairTasksUnderProcessing(cancellationToken);
foreach (var task in processingTaskListFinal)
{
processingNodes.Add(task.Target.ToString());
}
string pendingNodesString = string.Join(", ", pendingNodes);
string processingNodesString = string.Join(", ", processingNodes);
if (String.IsNullOrEmpty(pendingNodesString))
{
pendingNodesString = "None";
}
if (String.IsNullOrEmpty(processingNodesString))
{
processingNodesString = "None";
}
string description = string.Format(" Node currently being patched: {0} \nNodes waiting to be patched: {1}", processingNodesString, pendingNodesString);
HealthManagerHelper.PostNodeHealthReport(fabricClient, this.context.ServiceName, ClusterPatchingStatusProperty, description, HealthState.Ok);
}
/// <summary>
/// Fetches all the repair tasks which are under execution and checks
/// if any of them has exceeded the pre-specified execution timeout limit
/// </summary>
/// <param name="cancellationToken"></param>
/// <returns></returns>
internal async Task TimeoutRepairTasks(CancellationToken cancellationToken)
{
if (!this.ManageRepairTasksOnTimeout)
{
return;
}
// Get repair tasks which have been approved and are still under execution by POA
RepairTaskList repairTasks = await this.fabricClient.RepairManager.GetRepairTaskListAsync(TaskIdPrefix,
RepairTaskStateFilter.Approved | RepairTaskStateFilter.Executing, ExecutorName, this.DefaultTimeoutForOperation, cancellationToken);
foreach (var task in repairTasks)
{
ExecutorDataForRmTask executorData =
SerializationUtility.Deserialize <ExecutorDataForRmTask>(task.ExecutorData);
Debug.Assert(task.ApprovedTimestamp != null, "ApprovedTimestamp of an approved repair task can never be null");
TimeSpan elapsedTime = DateTime.UtcNow.Subtract(task.ApprovedTimestamp.Value);
if (elapsedTime > (TimeSpan.FromMinutes(executorData.ExecutorTimeoutInMinutes) + GraceTimeForNtService))
{
// Check if the node exists or not. If node does not exists, then don't break;
bool nodeExists = false;
string nodeName = this.GetNodeNameFromRepairTask(task);
NodeList nodeList = await this.fabricClient.QueryManager.GetNodeListAsync(nodeName, null, this.DefaultTimeoutForOperation, cancellationToken);
foreach (var node in nodeList)
{
if (node.NodeName.Equals(nodeName))
{
// Node Exists.
nodeExists = true;
break;
}
}
if (!nodeExists)
{
// If node does not exist now, there is no point in waiting on the task.
ServiceEventSource.Current.VerboseMessage("Cancelling repair task {0} which is in {1} state as the node {2} does not exist anymore.", task.TaskId, task.State, nodeName);
await this.CancelRepairTask(task);
continue;
}
switch (executorData.ExecutorSubState)
{
// These are special states where its best if NodeAgentNtService should move the repair task, just post warning in this case
case NodeAgentSfUtilityExitCodes.RestartRequested:
case NodeAgentSfUtilityExitCodes.RestartCompleted:
case NodeAgentSfUtilityExitCodes.InstallationCompleted:
{
string healthproperty = string.Format(
NodeTimeoutStatusFormat,
nodeName);
string healthDescription =
string.Format(
"Installation timeout {0} minutes alloted to repair task {1}, node {2} is over, however since node is in post-installation phase, wait for few more minutes for operation to complete"
+ "In case problem persists, please check if recent installations of updates has caused any problem on the node",
executorData.ExecutorTimeoutInMinutes,
task.TaskId,
nodeName);
ServiceEventSource.Current.ErrorMessage("Title = {0}, Description = {1}", healthproperty, healthDescription);
HealthManagerHelper.PostNodeHealthReport(this.fabricClient,
this.context.ServiceName,
healthproperty,
healthDescription,
HealthState.Warning,
60);
break;
}
default:
{
await UpdateRepairTaskState(task, nodeName, RepairTaskState.Restoring, executorData.ExecutorTimeoutInMinutes, cancellationToken);
break;
}
}
}
}
}
/// <summary>
/// Post the cluster patching status as events on CoordinatorService
/// </summary>
public async Task PostClusterPatchingStatus(CancellationToken cancellationToken)
{
try
{
NodeList nodeList = await this.fabricClient.QueryManager.GetNodeListAsync(null, null, this.DefaultTimeoutForOperation, cancellationToken);
IList <RepairTask> claimedTaskList = await this.GetClaimedRepairTasks(nodeList, cancellationToken);
RepairTaskList processingTaskList = await this.GetRepairTasksUnderProcessing(cancellationToken);
cancellationToken.ThrowIfCancellationRequested();
if (claimedTaskList.Any())
{
if (!processingTaskList.Any())
{
// This means that repair tasks are not getting approved.
ClusterHealth clusterHealth = await this.fabricClient.HealthManager.GetClusterHealthAsync();
if (clusterHealth.AggregatedHealthState == HealthState.Error)
{
// Reset Count
postUpdateCount = 0;
string warningDescription = " Cluster is currently unhealthy. Nodes are currently not getting patched by Patch Orchestration Application. Please ensure the cluster becomes healthy for patching to continue.";
await PostWarningOnCoordinatorService(warningDescription, 1);
}
else
{
postUpdateCount++;
if (postUpdateCount > 60)
{
// Reset Count and throw a warning on the service saying we dont know the reason. But POA not is not approving tasks.
postUpdateCount = 0;
string warningDescription = "Patch Orchestration Application is currently not patching nodes. This could be possible if there is some node which is stuck in disabling state for long time.";
await PostWarningOnCoordinatorService(warningDescription, 61);
}
}
}
else
{
// Reset Count
postUpdateCount = 0;
await PostRMTaskNodeUpdate(cancellationToken);
}
}
else
{
// Reset Count
postUpdateCount = 0;
if (processingTaskList.Any())
{
await PostRMTaskNodeUpdate(cancellationToken);
}
else
{
// Post the health event saying that there is no repair task and things are working fine.
string description = "No claimed tasks and no processing tasks are found.";
HealthManagerHelper.PostNodeHealthReport(this.fabricClient, this.context.ServiceName, ClusterPatchingStatusProperty, description, HealthState.Ok, -1);
}
}
}
catch (Exception ex)
{
ServiceEventSource.Current.ErrorMessage("PostClusterPatchingStatus failed with exception {0}", ex.ToString());
}
}
/// <summary>
/// Fetches all the repair tasks which are under execution and checks
/// if any of them has exceeded the pre-specified execution timeout limit
/// </summary>
/// <param name="cancellationToken"></param>
/// <returns></returns>
internal async Task TimeoutRepairTasks(CancellationToken cancellationToken)
{
if (!this.ManageRepairTasksOnTimeout)
{
return;
}
// Get repair tasks which have been approved and are still under execution by POA
RepairTaskList repairTasks = await this.fabricClient.RepairManager.GetRepairTaskListAsync(TaskIdPrefix,
RepairTaskStateFilter.Approved | RepairTaskStateFilter.Executing, ExecutorName, this.DefaultTimeoutForOperation, cancellationToken);
foreach (var task in repairTasks)
{
ExecutorDataForRmTask executorData =
SerializationUtility.Deserialize <ExecutorDataForRmTask>(task.ExecutorData);
Debug.Assert(task.ApprovedTimestamp != null, "ApprovedTimestamp of an approved repair task can never be null");
TimeSpan elapsedTime = DateTime.UtcNow.Subtract(task.ApprovedTimestamp.Value);
if (elapsedTime > (TimeSpan.FromMinutes(executorData.ExecutorTimeoutInMinutes) + GraceTimeForNtService))
{
// Check if the node exists or not. If node does not exists, then don't break;
bool nodeExists = false;
string nodeName = this.GetNodeNameFromRepairTask(task);
NodeList nodeList = await this.fabricClient.QueryManager.GetNodeListAsync(nodeName, null, this.DefaultTimeoutForOperation, cancellationToken);
foreach (var node in nodeList)
{
if (node.NodeName.Equals(nodeName))
{
// Node Exists.
nodeExists = true;
break;
}
}
if (!nodeExists)
{
// If node does not exist now, there is no point in waiting on the task.
ServiceEventSource.Current.VerboseMessage("Cancelling repair task {0} which is in {1} state as the node {2} does not exist anymore.", task.TaskId, task.State, nodeName);
await this.CancelRepairTask(task);
continue;
}
switch (executorData.ExecutorSubState)
{
// These are special states where its best if NodeAgentNtService should move the repair task, just post warning in this case
case NodeAgentSfUtilityExitCodes.RestartRequested:
case NodeAgentSfUtilityExitCodes.RestartCompleted:
case NodeAgentSfUtilityExitCodes.InstallationCompleted:
{
string message =
string.Format(
"Repair Task {0} did not complete within the Timeout period for node {1}. Since Installation was already started, updating Repair Task state to further proceed with Node enabling",
task.TaskId,
nodeName);
ServiceEventSource.Current.InfoMessage(message);
await UpdateRepairTaskState(task, nodeName, RepairTaskState.Restoring, executorData.ExecutorTimeoutInMinutes, cancellationToken);
break;
}
default:
{
string message =
string.Format(
"Repair Task {0} completed within the Timeout period for node {1}. Updating Repair Task state to further proceed with Node enabling",
task.TaskId,
nodeName);
ServiceEventSource.Current.InfoMessage(message);
await UpdateRepairTaskState(task, nodeName, RepairTaskState.Restoring, executorData.ExecutorTimeoutInMinutes, cancellationToken);
break;
}
}
}
}
}
private RepairTaskList GetRepairTaskListAsyncEndWrapper(
NativeCommon.IFabricAsyncOperationContext context)
{
return(RepairTaskList.CreateFromNativeListResult(
this.nativeRepairClient.EndGetRepairTaskList(context)));
}
