/// <inheritdoc/>
public override Task ReportAsync(CancellationToken token)
{
this.Token.ThrowIfCancellationRequested();
var timeToLiveWarning = this.SetTimeToLiveWarning(this.runtime.Seconds);
var portInformationReport = new Utilities.HealthReport
{
Observer = this.ObserverName,
NodeName = this.NodeName,
HealthMessage = $"Number of ports in use by Fabric services: {this.TotalActivePortCount}\n" +
$"Number of ephemeral ports in use by Fabric services: {this.TotalActiveEphemeralPortCount}",
State = HealthState.Ok,
HealthReportTimeToLive = timeToLiveWarning,
};
// TODO: Report on port count based on thresholds PortCountWarning/Error...
this.HealthReporter.ReportHealthToServiceFabric(portInformationReport);
// Reset ports counters...
this.TotalActivePortCount = 0;
this.TotalActiveEphemeralPortCount = 0;
// CPU
this.ProcessResourceDataList(
this.allCpuData,
this.CpuErrorUsageThresholdPct,
this.CpuWarnUsageThresholdPct);
// Memory
this.ProcessResourceDataList(
this.allMemData,
this.MemErrorUsageThresholdMB,
this.MemWarnUsageThresholdMB);
// Disk IO - Reads
this.ProcessResourceDataList(
this.allAppDiskReadsData,
this.DiskErrorIOReadsThresholdMS,
this.DiskWarnIOReadsThresholdMS);
// Disk IO - Writes
this.ProcessResourceDataList(
this.allAppDiskWritesData,
this.DiskErrorIOWritesThresholdMS,
this.DiskWarnIOWritesThresholdMS);
// Windows Event Log
if (this.monitorWinEventLog)
{
// SF Eventlog Errors?
// Write this out to a new file, for use by the web front end log viewer...
// Format = HTML...
int count = this.evtRecordList.Count();
var logPath = Path.Combine(this.ObserverLogger.LogFolderBasePath, "EventVwrErrors.txt");
// Remove existing file...
if (File.Exists(logPath))
{
try
{
File.Delete(logPath);
}
catch (IOException)
{
}
catch (UnauthorizedAccessException)
{
}
}
if (count >= 10)
{
var sb = new StringBuilder();
sb.AppendLine("<br/><div><strong>" +
"<a href='javascript:toggle(\"evtContainer\")'>" +
"<div id=\"plus\" style=\"display: inline; font-size: 25px;\">+</div> " + count +
" Error Events in ServiceFabric and System</a> " +
"Event logs</strong>.<br/></div>");
sb.AppendLine("<div id='evtContainer' style=\"display: none;\">");
foreach (var evt in this.evtRecordList.Distinct())
{
token.ThrowIfCancellationRequested();
try
{
// Access event properties:
sb.AppendLine("<div>" + evt.LogName + "</div>");
sb.AppendLine("<div>" + evt.LevelDisplayName + "</div>");
if (evt.TimeCreated.HasValue)
{
sb.AppendLine("<div>" + evt.TimeCreated.Value.ToShortDateString() + "</div>");
}
foreach (var prop in evt.Properties)
{
if (prop.Value != null && Convert.ToString(prop.Value).Length > 0)
{
sb.AppendLine("<div>" + prop.Value + "</div>");
}
}
}
catch (EventLogException)
{
}
}
sb.AppendLine("</div>");
this.ObserverLogger.TryWriteLogFile(logPath, sb.ToString());
sb.Clear();
}
// Clean up...
if (count > 0)
{
this.evtRecordList.Clear();
}
}
return(Task.CompletedTask);
}
/// <inheritdoc/>
public override Task ReportAsync(CancellationToken token)
{
try
{
token.ThrowIfCancellationRequested();
// OS Health...
if (this.osStatus != null &&
this.osStatus.ToUpper() != "OK")
{
string healthMessage = $"OS reporting unhealthy: {this.osStatus}";
var healthReport = new Utilities.HealthReport
{
Observer = this.ObserverName,
NodeName = this.NodeName,
HealthMessage = healthMessage,
State = HealthState.Error,
HealthReportTimeToLive = this.SetTimeToLiveWarning(),
};
this.HealthReporter.ReportHealthToServiceFabric(healthReport);
// This means this observer created a Warning or Error SF Health Report
this.HasActiveFabricErrorOrWarning = true;
// Send Health Report as Telemetry (perhaps it signals an Alert from App Insights, for example...)...
if (this.IsTelemetryEnabled)
{
_ = this.ObserverTelemetryClient?.ReportHealthAsync(
FabricRuntime.GetActivationContext().ApplicationName,
this.FabricServiceContext.ServiceName.OriginalString,
"FabricObserver",
this.ObserverName,
$"{this.NodeName}/OS reporting unhealthy: {this.osStatus}",
HealthState.Error,
token);
}
}
else if (this.HasActiveFabricErrorOrWarning &&
this.osStatus != null &&
this.osStatus.ToUpper() == "OK")
{
// Clear Error or Warning with an OK Health Report...
string healthMessage = $"OS reporting healthy: {this.osStatus}";
var healthReport = new Utilities.HealthReport
{
Observer = this.ObserverName,
NodeName = this.NodeName,
HealthMessage = healthMessage,
State = HealthState.Ok,
HealthReportTimeToLive = default(TimeSpan),
};
this.HealthReporter.ReportHealthToServiceFabric(healthReport);
// Reset internal health state...
this.HasActiveFabricErrorOrWarning = false;
}
var logPath = Path.Combine(this.ObserverLogger.LogFolderBasePath, "SysInfo.txt");
// This file is used by the web application (log reader...)...
if (!this.ObserverLogger.TryWriteLogFile(logPath, $"Last updated on {DateTime.UtcNow.ToString("M/d/yyyy HH:mm:ss")} UTC<br/>{this.osReport}"))
{
this.HealthReporter.ReportFabricObserverServiceHealth(
this.FabricServiceContext.ServiceName.OriginalString,
this.ObserverName,
HealthState.Warning,
"Unable to create SysInfo.txt file...");
}
var osReport = new Utilities.HealthReport
{
Observer = this.ObserverName,
HealthMessage = this.osReport,
State = HealthState.Ok,
NodeName = this.NodeName,
HealthReportTimeToLive = this.SetTimeToLiveWarning(),
};
this.HealthReporter.ReportHealthToServiceFabric(osReport);
return Task.CompletedTask;
}
catch (Exception e)
{
this.HealthReporter.ReportFabricObserverServiceHealth(
this.FabricServiceContext.ServiceName.OriginalString,
this.ObserverName,
HealthState.Error,
$"Unhandled exception processing OS information: {e.Message}: \n {e.StackTrace}");
throw;
}
}
/// <inheritdoc/>
public override async Task ReportAsync(CancellationToken token)
{
string app;
var timeToLiveWarning = this.SetTimeToLiveWarning();
// Report on connection state.
for (int j = 0; j < this.userEndpoints.Count; j++)
{
token.ThrowIfCancellationRequested();
var deployedApps = await this.FabricClientInstance.QueryManager
.GetDeployedApplicationListAsync(
this.NodeName,
new Uri(this.userEndpoints[j].AppTarget)).ConfigureAwait(true);
// We only care about deployed apps.
if (deployedApps == null || deployedApps.Count < 1)
{
continue;
}
app = this.userEndpoints[j].AppTarget.Replace("fabric:/", string.Empty);
for (int i = 0; i < this.connectionStatus.Count; i++)
{
token.ThrowIfCancellationRequested();
var connStatus = this.connectionStatus[i];
if (!connStatus.Connected)
{
this.healthState = HealthState.Warning;
var healthMessage = "Outbound Internet connection failure detected for endpoint " + connStatus.HostName + "\n";
Utilities.HealthReport report = new Utilities.HealthReport
{
AppName = new Uri(this.userEndpoints[j].AppTarget),
Code = FOErrorWarningCodes.AppWarningNetworkEndpointUnreachable,
EmitLogEvent = true,
HealthMessage = healthMessage,
HealthReportTimeToLive = timeToLiveWarning,
State = this.healthState,
NodeName = this.NodeName,
Observer = this.ObserverName,
ReportType = HealthReportType.Application,
ResourceUsageDataProperty = $"{ErrorWarningProperty.InternetConnectionFailure}: connStatus.HostName",
};
// Send health report Warning and log event locally.
this.HealthReporter.ReportHealthToServiceFabric(report);
// This means this observer created a Warning or Error SF Health Report
this.HasActiveFabricErrorOrWarning = true;
// Send Health Report as Telemetry (perhaps it signals an Alert from App Insights, for example.).
if (this.IsTelemetryEnabled)
{
_ = this.ObserverTelemetryClient?.ReportHealthAsync(
Utilities.Telemetry.HealthScope.Application,
this.userEndpoints[j].AppTarget,
HealthState.Warning,
$"{this.NodeName}/{FOErrorWarningCodes.AppWarningNetworkEndpointUnreachable}: {healthMessage}",
this.ObserverName,
this.Token);
}
}
else
{
if (connStatus.Health == HealthState.Warning)
{
this.healthState = HealthState.Ok;
var healthMessage = "Outbound Internet connection test successful.";
// Clear existing Health Warning.
Utilities.HealthReport report = new Utilities.HealthReport
{
AppName = new Uri(this.userEndpoints[j].AppTarget),
EmitLogEvent = true,
HealthMessage = healthMessage,
HealthReportTimeToLive = default(TimeSpan),
State = this.healthState,
NodeName = this.NodeName,
Observer = this.ObserverName,
ReportType = HealthReportType.Application,
};
this.HealthReporter.ReportHealthToServiceFabric(report);
// Reset health state.
this.HasActiveFabricErrorOrWarning = false;
}
}
}
}
// Clear
this.connectionStatus.RemoveAll(conn => conn.Connected == true);
this.connectionStatus.TrimExcess();
}
public void ProcessResourceDataReportHealth <T>(
FabricResourceUsageData <T> data,
T thresholdError,
T thresholdWarning,
TimeSpan healthReportTtl,
HealthReportType healthReportType = HealthReportType.Node,
string app = null,
ReplicaMonitoringInfo replicaOrInstance = null,
bool dumpOnError = false)
{
if (data == null)
{
throw new ArgumentException("Supply all required parameters with non-null value...");
}
var thresholdName = "Minimum";
bool warningOrError = false;
string repPartitionId = null, repOrInstanceId = null, name = null, id = null, procName = null;
T threshold = thresholdWarning;
var healthState = HealthState.Ok;
Uri appName = null;
if (replicaOrInstance != null)
{
repPartitionId = $"Partition: {replicaOrInstance.Partitionid}";
repOrInstanceId = $"Replica: {replicaOrInstance.ReplicaOrInstanceId}";
procName = Process.GetProcessById((int)replicaOrInstance.ReplicaHostProcessId)?.ProcessName;
}
// Create a unique node id which may be used in the case of warnings or OK clears...
if (app != null)
{
appName = new Uri(app);
name = app.Replace("fabric:/", string.Empty);
id = name + "_" + data.Property.Replace(" ", string.Empty);
}
// Telemetry...
if (this.IsTelemetryEnabled)
{
_ = this.ObserverTelemetryClient?.ReportMetricAsync($"{this.NodeName}-{app}-{data.Id}-{data.Property}", data.AverageDataValue, this.Token);
}
// ETW...
if (this.IsEtwEnabled)
{
Logger.EtwLogger?.Write(
$"FabricObserverDataEvent",
new
{
Level = 0, // Info
Node = this.NodeName,
Observer = this.ObserverName,
Property = data.Property,
Id = data.Id,
Value = $"{Math.Round(data.AverageDataValue)}",
Unit = data.Units,
});
}
// Health Error
if (data.IsUnhealthy(thresholdError))
{
thresholdName = "Maximum";
threshold = thresholdError;
warningOrError = true;
healthState = HealthState.Error;
// This is primarily useful for AppObserver, but makes sense to be
// part of the base class for future use, like for FSO...
if (replicaOrInstance != null && procName != null && dumpOnError)
{
try
{
int procId = (int)replicaOrInstance.ReplicaHostProcessId;
if (!this.serviceDumpCountDictionary.ContainsKey(procName))
{
this.serviceDumpCountDictionary.Add(procName, 0);
}
if (this.serviceDumpCountDictionary[procName] < this.maxDumps)
{
// DumpServiceProcess defaults to a Full dump with
// process memory, handles and thread data...
bool success = this.DumpServiceProcess(procId);
if (success)
{
this.serviceDumpCountDictionary[procName]++;
}
}
}
// Ignore these, it just means no dmp will be created.This is not
// critical to FO... Log as info, not warning...
catch (ArgumentException ae)
{
this.ObserverLogger.LogInfo($"Unable to generate dmp file:\n{ae.ToString()}");
}
catch (InvalidOperationException ioe)
{
this.ObserverLogger.LogInfo($"Unable to generate dmp file:\n{ioe.ToString()}");
}
}
}
// Health Warning
if (!warningOrError && data.IsUnhealthy(thresholdWarning))
{
warningOrError = true;
healthState = HealthState.Warning;
}
if (warningOrError)
{
string errorWarningKind = null;
if (data.Property.ToLower().Contains("cpu"))
{
errorWarningKind = (healthState == HealthState.Error) ? ErrorWarningCode.ErrorCpuTime : ErrorWarningCode.WarningCpuTime;
}
else if (data.Property.ToLower().Contains("disk space"))
{
errorWarningKind = (healthState == HealthState.Error) ? ErrorWarningCode.ErrorDiskSpace : ErrorWarningCode.WarningDiskSpace;
}
else if (data.Property == "Memory Consumption MB")
{
errorWarningKind = (healthState == HealthState.Error) ? ErrorWarningCode.ErrorMemoryCommitted : ErrorWarningCode.WarningMemoryCommitted;
}
else if (data.Property == "Memory Consumption %")
{
errorWarningKind = (healthState == HealthState.Error) ? ErrorWarningCode.ErrorMemoryPercentUsed : ErrorWarningCode.WarningMemoryPercentUsed;
}
else if (data.Property.Contains("Read"))
{
errorWarningKind = (healthState == HealthState.Error) ? ErrorWarningCode.ErrorDiskIoReads : ErrorWarningCode.WarningDiskIoReads;
}
else if (data.Property.Contains("Write"))
{
errorWarningKind = (healthState == HealthState.Error) ? ErrorWarningCode.ErrorDiskIoWrites : ErrorWarningCode.WarningDiskIoWrites;
}
else if (data.Property.Contains("Queue"))
{
errorWarningKind = (healthState == HealthState.Error) ? ErrorWarningCode.ErrorDiskAverageQueueLength : ErrorWarningCode.WarningDiskAverageQueueLength;
}
else if (data.Property.Contains("Firewall"))
{
errorWarningKind = (healthState == HealthState.Error) ? ErrorWarningCode.ErrorTooManyFirewallRules : ErrorWarningCode.WarningTooManyFirewallRules;
}
else if (data.Property.Contains("Ports"))
{
errorWarningKind = (healthState == HealthState.Error) ? ErrorWarningCode.ErrorTooManyActivePorts : ErrorWarningCode.WarningTooManyActivePorts;
}
var healthMessage = new StringBuilder();
if (name != null)
{
healthMessage.Append($"{name} (Service Process: {procName}, {repPartitionId}, {repOrInstanceId}): ");
}
healthMessage.Append($"{data.Property} is at or above the specified {thresholdName} limit ({threshold}{data.Units})");
healthMessage.AppendLine($" - Average {data.Property}: {Math.Round(data.AverageDataValue)}{data.Units}");
// Set internal fabric health states...
data.ActiveErrorOrWarning = true;
// This means this observer created a Warning or Error SF Health Report
this.HasActiveFabricErrorOrWarning = true;
var healthReport = new Utilities.HealthReport
{
AppName = appName,
Code = errorWarningKind,
EmitLogEvent = true,
HealthMessage = healthMessage.ToString(),
HealthReportTimeToLive = healthReportTtl,
ReportType = healthReportType,
State = healthState,
NodeName = this.NodeName,
Observer = this.ObserverName,
};
// Emit a Fabric Health Report and optionally a local log write...
this.HealthReporter.ReportHealthToServiceFabric(healthReport);
// Send Health Report as Telemetry event (perhaps it signals an Alert from App Insights, for example...)...
if (this.IsTelemetryEnabled)
{
_ = this.ObserverTelemetryClient?.ReportHealthAsync(
id,
this.FabricServiceContext.ServiceName.OriginalString,
"FabricObserver",
this.ObserverName,
$"{this.NodeName}/{errorWarningKind}/{data.Property}/{Math.Round(data.AverageDataValue)}",
healthState,
this.Token);
}
// ETW...
if (this.IsEtwEnabled)
{
Logger.EtwLogger?.Write(
$"FabricObserverDataEvent",
new
{
Level = (healthState == HealthState.Warning) ? 1 : 2,
Node = this.NodeName,
Observer = this.ObserverName,
HealthEventErrorCode = errorWarningKind,
HealthEventDescription = healthMessage.ToString(),
Property = data.Property,
Id = data.Id,
Value = $"{Math.Round(data.AverageDataValue)}",
Unit = data.Units,
});
}
// Clean up sb...
healthMessage.Clear();
}
else
{
if (data.ActiveErrorOrWarning)
{
Utilities.HealthReport report = new Utilities.HealthReport
{
AppName = appName,
EmitLogEvent = true,
HealthMessage = $"{data.Id}: {data.Property} is now within normal/expected range.",
HealthReportTimeToLive = default(TimeSpan),
ReportType = healthReportType,
State = HealthState.Ok,
NodeName = this.NodeName,
Observer = this.ObserverName,
};
// Emit an Ok Health Report to clear Fabric Health warning...
this.HealthReporter.ReportHealthToServiceFabric(report);
// Reset health states...
data.ActiveErrorOrWarning = false;
this.HasActiveFabricErrorOrWarning = false;
}
}
// No need to keep data in memory...
data.Data.Clear();
data.Data.TrimExcess();
}
/// <inheritdoc/>
public override Task ReportAsync(CancellationToken token)
{
if (token.IsCancellationRequested)
{
return(Task.CompletedTask);
}
// Someone calling without observing first, must be run after a new run of ObserveAsync
if (this.ExpiringWarnings == null ||
this.ExpiredWarnings == null ||
this.NotFoundWarnings == null)
{
return(Task.CompletedTask);
}
HealthReport healthReport;
if (this.ExpiringWarnings.Count == 0 &&
this.ExpiredWarnings.Count == 0 &&
this.NotFoundWarnings.Count == 0)
{
healthReport = new HealthReport
{
Observer = this.ObserverName,
ReportType = HealthReportType.Node,
EmitLogEvent = true,
NodeName = this.NodeName,
HealthMessage = $"All cluster and monitored app certificates are healthy.",
State = HealthState.Ok,
HealthReportTimeToLive = this.RunInterval > TimeSpan.MinValue ? this.RunInterval : this.HealthReportTimeToLive,
};
this.HasActiveFabricErrorOrWarning = false;
}
else
{
string healthMessage = (this.ExpiredWarnings.Count == 0 ? string.Empty : (this.ExpiredWarnings.Aggregate(string.Empty, (i, j) => i + "\n" + j) + "\n")) +
(this.NotFoundWarnings.Count == 0 ? string.Empty : (this.NotFoundWarnings.Aggregate(string.Empty, (i, j) => i + "\n" + j) + "\n")) +
(this.ExpiringWarnings.Count == 0 ? string.Empty : this.ExpiringWarnings.Aggregate(string.Empty, (i, j) => i + "\n" + j));
healthReport = new HealthReport
{
Code = FOErrorWarningCodes.WarningCertificateExpiration,
Observer = this.ObserverName,
ReportType = HealthReportType.Node,
EmitLogEvent = true,
NodeName = this.NodeName,
HealthMessage = healthMessage,
State = HealthState.Warning,
HealthReportTimeToLive = this.RunInterval > TimeSpan.MinValue ? this.RunInterval : this.HealthReportTimeToLive,
};
this.HasActiveFabricErrorOrWarning = true;
}
this.HealthReporter.ReportHealthToServiceFabric(healthReport);
this.ExpiredWarnings = null;
this.ExpiringWarnings = null;
this.NotFoundWarnings = null;
this.LastRunDateTime = DateTime.Now;
return(Task.CompletedTask);
}
public void ReportHealthToServiceFabric(HealthReport healthReport)
{
if (healthReport == null)
{
return;
}
// There is no real need to change Immediate to true here for errors/warnings. This only adds unecessary stress to the
// Health subsystem.
var sendOptions = new HealthReportSendOptions {
Immediate = false
};
// Quickly send OK (clears warning/errors states).
if (healthReport.State == HealthState.Ok)
{
sendOptions.Immediate = true;
}
var timeToLive = TimeSpan.FromMinutes(5);
if (healthReport.HealthReportTimeToLive != default)
{
timeToLive = healthReport.HealthReportTimeToLive;
}
// In order for multiple Error/Warning/Ok events to show up in SFX Details view from observer instances,
// Event Source Ids must be unique, thus the seemingly strange conditionals inside the cases below:
// The apparent duplicity in OR checks is for the case when the incoming report is an OK report, where there is
// no error code, but the specific ErrorWarningProperty is known.
string property;
switch (healthReport.Observer)
{
case ObserverConstants.AppObserverName:
property = "AppHealth";
break;
case ObserverConstants.CertificateObserverName:
property = "SecurityHealth";
break;
case ObserverConstants.DiskObserverName:
property = "DiskHealth";
break;
case ObserverConstants.FabricSystemObserverName:
property = "FabricSystemServiceHealth";
break;
case ObserverConstants.NetworkObserverName:
property = "NetworkingHealth";
break;
case ObserverConstants.OSObserverName:
property = "MachineInformation";
break;
case ObserverConstants.NodeObserverName:
property = "MachineResourceHealth";
break;
default:
property = "FOGenericHealth";
break;
}
string sourceId = healthReport.Observer;
if (!string.IsNullOrEmpty(healthReport.Code))
{
// Only use FOErrorWarningCode for source
sourceId = $"{healthReport.Code}";
}
var healthInformation = new HealthInformation(sourceId, property, healthReport.State)
{
Description = healthReport.HealthMessage,
TimeToLive = timeToLive,
RemoveWhenExpired = true,
};
// Log event only if ObserverWebApi (REST Log reader.) app is deployed.
if (ObserverManager.ObserverWebAppDeployed &&
healthReport.EmitLogEvent)
{
if (healthReport.State == HealthState.Error)
{
this.logger.LogError(healthReport.NodeName + ": {0}", healthInformation.Description);
}
else if (healthReport.State == HealthState.Warning)
{
this.logger.LogWarning(healthReport.NodeName + ": {0}", healthInformation.Description);
}
else
{
this.logger.LogInfo(healthReport.NodeName + ": {0}", healthInformation.Description);
}
}
// To SFX and Telemetry provider.
if (healthReport.ReportType == HealthReportType.Application && healthReport.AppName != null)
{
var appHealthReport = new ApplicationHealthReport(healthReport.AppName, healthInformation);
this.fabricClient.HealthManager.ReportHealth(appHealthReport, sendOptions);
}
else
{
var nodeHealthReport = new NodeHealthReport(healthReport.NodeName, healthInformation);
this.fabricClient.HealthManager.ReportHealth(nodeHealthReport, sendOptions);
}
}
