Skip to content

Commit

Permalink
Merge branch 'master' into users/sourabhjain/nullchecks
Browse files Browse the repository at this point in the history
  • Loading branch information
@sourabh1007
sourabh1007 authored Aug 31, 2023
2 parents b1c4872 + c72ebc8 commit 11ecfb9
Show file tree
Hide file tree
Showing 24 changed files with 508 additions and 280 deletions.
15 changes: 8 additions & 7 deletions Microsoft.Azure.Cosmos.Samples/Tools/Benchmark/AzureVmBenchmark/azuredeploy.json
View file
Original file line number Diff line number Diff line change
Expand Up @@ -209,10 +209,11 @@
"customScriptUrl": "[concat('https://raw.githubusercontent.com/Azure/azure-cosmos-dotnet-v3/',parameters('benchmarkingToolsBranchName'),'/Microsoft.Azure.Cosmos.Samples/Tools/Benchmark/AzureVmBenchmark/scripts/custom-script.sh')]",
"vmScriptExtensionScriptName": "execute.sh",
"convertedDatetime": "[dateTimeFromEpoch(parameters('startDate'))]",
"metricsPrefix" : "[if(startsWith(parameters('workloadType'), 'Read'), 'Read', if(startsWith(parameters('workloadType'), 'Query'), 'Query', if(startsWith(parameters('workloadType'), 'Insert'), 'Insert', '')))]",
"appInsightsResourceIds": "[concat('/subscriptions/', subscription().subscriptionId, '/resourceGroups/', resourceGroup().name, '/providers/microsoft.insights/components/', parameters('applicationInsightsName'))]",
"chart0Expression": "[concat('customMetrics\n| where name == \"ReadOperationLatencyInMs\" and timestamp > ago(1d)\n| summarize\n percentile(value, 50),\n percentile(value, 75),\n percentile(value, 90),\n percentile(value, 95)\n by ts = bin(timestamp, ', parameters('metricsReportingIntervalInSec'), 's)\n| render timechart \n\n')]",
"chart1Expression": "[concat('customMetrics\n| where name == \"ReadOperationLatencyInMs\" and timestamp > ago(1d)\n| summarize\n percentile(value, 99),\n percentile(value, 99.9),\n percentile(value, 99.99)\n by ts = bin(timestamp, ', parameters('metricsReportingIntervalInSec'), 's)\n| render timechart \n\n')]",
"chart2Expression": "[concat('customMetrics\n| where name == \"ReadOperationRps\" and timestamp > ago(1d)\n| summarize\n avg(value)\n by ts = bin(timestamp, ', parameters('metricsReportingIntervalInSec'), 's)\n| render timechart \n\n')]"
"chart0Expression": "[concat('customMetrics\n| where name == \"', variables('metricsPrefix'), 'OperationLatencyInMs\" and timestamp > ago(1d)\n| summarize\n percentile(value, 50),\n percentile(value, 75),\n percentile(value, 90),\n percentile(value, 95)\n by ts = bin(timestamp, ', parameters('metricsReportingIntervalInSec'), 's)\n| render timechart \n\n')]",
"chart1Expression": "[concat('customMetrics\n| where name == \"', variables('metricsPrefix'), 'OperationLatencyInMs\" and timestamp > ago(1d)\n| summarize\n percentile(value, 99),\n percentile(value, 99.9),\n percentile(value, 99.99)\n by ts = bin(timestamp, ', parameters('metricsReportingIntervalInSec'), 's)\n| render timechart \n\n')]",
"chart2Expression": "[concat('customMetrics\n| where name == \"', variables('metricsPrefix'), 'OperationRps\" and timestamp > ago(1d)\n| summarize\n avg(value)\n by ts = bin(timestamp, ', parameters('metricsReportingIntervalInSec'), 's)\n| render timechart \n\n')]"

},
"resources": [
Expand Down Expand Up @@ -776,7 +777,7 @@
},
{
"name": "Query",
"value": "customMetrics\n| where name == \"ReadOperationFailure\" and timestamp > ago(1d)\n| summarize\n sum(value)\n by ts = bin(timestamp, 1s)\n | render timechart\n",
"value": "[concat('customMetrics\n| where name == \"', variables('metricsPrefix'), 'OperationFailure\" and timestamp > ago(1d)\n| summarize\n sum(value)\n by ts = bin(timestamp, 1s)\n | render timechart\n')]",
"isOptional": true
},
{
Expand Down Expand Up @@ -834,7 +835,7 @@
"type": "Extension/Microsoft_OperationsManagementSuite_Workspace/PartType/LogsDashboardPart",
"settings": {
"content": {
"Query": "customMetrics\n| where name == \"ReadOperationFailure\" and timestamp > ago(1d)\n| summarize\n sum(value)\n by ts = bin(timestamp, 1s)\n | render timechart\n\n",
"Query": "[concat('customMetrics\n| where name == \"', variables('metricsPrefix'), 'OperationFailure\" and timestamp > ago(1d)\n| summarize\n sum(value)\n by ts = bin(timestamp, 1s)\n | render timechart\n\n')]",
"Dimensions": {
"xAxis": {
"name": "ts",
Expand Down Expand Up @@ -903,7 +904,7 @@
},
{
"name": "Query",
"value": "customMetrics\n| where name == \"ReadOperationSuccess\" and timestamp > ago(1d)\n| summarize\n sum(value)\n by ts = bin(timestamp, 1s)\n | render timechart\n",
"value": "[concat('customMetrics\n| where name == \"', variables('metricsPrefix'), 'OperationSuccess\" and timestamp > ago(1d)\n| summarize\n sum(value)\n by ts = bin(timestamp, 1s)\n | render timechart\n')]",
"isOptional": true
},
{
Expand Down Expand Up @@ -961,7 +962,7 @@
"type": "Extension/Microsoft_OperationsManagementSuite_Workspace/PartType/LogsDashboardPart",
"settings": {
"content": {
"Query": "customMetrics\n| where name == \"ReadOperationSuccess\" and timestamp > ago(1d)\n| summarize\n sum(value)\n by ts = bin(timestamp, 1s)\n | render timechart\n\n",
"Query": "[concat('customMetrics\n| where name == \"', variables('metricsPrefix'), 'OperationSuccess\" and timestamp > ago(1d)\n| summarize\n sum(value)\n by ts = bin(timestamp, 1s)\n | render timechart\n\n')]",
"Dimensions": {
"xAxis": {
"name": "ts",
Expand Down
40 changes: 12 additions & 28 deletions Microsoft.Azure.Cosmos.Samples/Tools/Benchmark/BenchmarkConfig.cs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,7 @@ namespace CosmosBenchmark
using System.Linq;
using System.Runtime;
using CommandLine;
using Microsoft.Azure.Cosmos.Telemetry;
using Microsoft.Azure.Documents.Client;
using Newtonsoft.Json;

Expand Down Expand Up @@ -102,18 +103,12 @@ public class BenchmarkConfig
[Option(Required = false, HelpText = "Disable core SDK logging")]
public bool DisableCoreSdkLogging { get; set; }

[Option(Required = false, HelpText = "Enable Client Telemetry")]
public bool EnableTelemetry { get; set; }

[Option(Required = false, HelpText = "Enable Distributed Tracing")]
public bool EnableDistributedTracing { get; set; }

[Option(Required = false, HelpText = "Client Telemetry Schedule in Seconds")]
public int TelemetryScheduleInSec { get; set; }

[Option(Required = false, HelpText = "Client Telemetry Endpoint")]
public string TelemetryEndpoint { get; set; }

[Option(Required = false, HelpText = "Endpoint to publish results to")]
public string ResultsEndpoint { get; set; }

Expand Down Expand Up @@ -143,6 +138,9 @@ public class BenchmarkConfig
[Option(Required = false, HelpText = "Application Insights connection string")]
public string AppInsightsConnectionString { get; set; }

[Option(Required = false, HelpText = "Enable Client Telemetry Feature in SDK. Make sure you enable it from the portal also.")]
public bool EnableClientTelemetry { get; set; } = true;

internal int GetTaskCount(int containerThroughput)
{
int taskCount = this.DegreeOfParallelism;
Expand Down Expand Up @@ -210,35 +208,21 @@ private string GetUserAgentPrefix()

internal Microsoft.Azure.Cosmos.CosmosClient CreateCosmosClient(string accountKey)
{
// Overwrite the default timespan if configured
if(this.TelemetryScheduleInSec > 0)
{
ClientTelemetryOptions.DefaultIntervalForTelemetryJob = TimeSpan.FromSeconds(this.TelemetryScheduleInSec);
}

Microsoft.Azure.Cosmos.CosmosClientOptions clientOptions = new Microsoft.Azure.Cosmos.CosmosClientOptions()
{
ApplicationName = this.GetUserAgentPrefix(),
MaxRetryAttemptsOnRateLimitedRequests = 0,
MaxRequestsPerTcpConnection = this.MaxRequestsPerTcpConnection,
MaxTcpConnectionsPerEndpoint = this.MaxTcpConnectionsPerEndpoint
MaxTcpConnectionsPerEndpoint = this.MaxTcpConnectionsPerEndpoint,
EnableClientTelemetry = this.EnableClientTelemetry
};

if (this.EnableTelemetry)
{
Environment.SetEnvironmentVariable(
Microsoft.Azure.Cosmos.Telemetry.ClientTelemetryOptions.EnvPropsClientTelemetryEnabled,
"true");

if (this.TelemetryScheduleInSec > 0)
{
Environment.SetEnvironmentVariable(
Microsoft.Azure.Cosmos.Telemetry.ClientTelemetryOptions.EnvPropsClientTelemetrySchedulingInSeconds,
Convert.ToString(this.TelemetryScheduleInSec));
}

if (!string.IsNullOrEmpty(this.TelemetryEndpoint))
{
Environment.SetEnvironmentVariable(
Microsoft.Azure.Cosmos.Telemetry.ClientTelemetryOptions.EnvPropsClientTelemetryEndpoint,
this.TelemetryEndpoint);
}
}

if (!string.IsNullOrWhiteSpace(this.ConsistencyLevel))
{
clientOptions.ConsistencyLevel = (Microsoft.Azure.Cosmos.ConsistencyLevel)Enum.Parse(typeof(Microsoft.Azure.Cosmos.ConsistencyLevel), this.ConsistencyLevel, ignoreCase: true);
Expand Down
11 changes: 4 additions & 7 deletions Microsoft.Azure.Cosmos.Samples/Tools/CTL/CTLConfig.cs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -97,18 +97,15 @@ public string DiagnosticsThresholdDuration
[Option("ctl_logging_context", Required = false, HelpText = "Defines a custom context to use on metrics")]
public string LogginContext { get; set; } = string.Empty;

[Option("ctl_telemetry_endpoint", Required = false, HelpText = "telemetry juno end point")]
public string TelemetryEndpoint { get; set; }

[Option("ctl_telemetry_schedule_in_sec", Required = false, HelpText = "telemetry task schedule time in sec")]
public string TelemetryScheduleInSeconds { get; set; }

[Option("ctl_reservoir_type", Required = false, HelpText = "Defines the reservoir type. Valid values are: Uniform, SlidingWindow and ExponentialDecay. The default value is SlidingWindow.")]
public ReservoirTypes ReservoirType { get; set; } = ReservoirTypes.SlidingWindow;

[Option("ctl_reservoir_sample_size", Required = false, HelpText = "The reservoir sample size.")]
public int ReservoirSampleSize { get; set; } = 1028;

[Option("ctl_enable_client_telemetry", Required = false, HelpText = "Enable Client Telemetry Feature in SDK. Make sure you enable it from the portal also.")]
public bool EnableClientTelemetry { get; set; } = true;

internal TimeSpan RunningTimeDurationAsTimespan { get; private set; } = TimeSpan.FromHours(10);
internal TimeSpan DiagnosticsThresholdDurationAsTimespan { get; private set; } = TimeSpan.FromSeconds(60);

Expand All @@ -133,7 +130,7 @@ internal CosmosClient CreateCosmosClient()
CosmosClientOptions clientOptions = new CosmosClientOptions()
{
ApplicationName = CTLConfig.UserAgentSuffix,
EnableClientTelemetry = true
EnableClientTelemetry = this.EnableClientTelemetry
};

if (this.UseGatewayMode)
Expand Down
14 changes: 0 additions & 14 deletions Microsoft.Azure.Cosmos.Samples/Tools/CTL/Program.cs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -29,8 +29,6 @@ public static async Task Main(string[] args)
{
CTLConfig config = CTLConfig.From(args);

SetEnvironmentVariables(config);

if (config.OutputEventTraces)
{
EnableTraceSourcesToConsole();
Expand All @@ -54,12 +52,6 @@ await scenario.InitializeAsync(

logger.LogInformation("Initialization completed.");

if(client.ClientOptions.EnableClientTelemetry.GetValueOrDefault()) {
logger.LogInformation("Telemetry is enabled for CTL.");
} else {
logger.LogInformation("Telemetry is disabled for CTL.");
}

List<Task> tasks = new List<Task>
{
scenario.RunAsync(
Expand Down Expand Up @@ -148,12 +140,6 @@ await scenario.InitializeAsync(
}
}

private static void SetEnvironmentVariables(CTLConfig config)
{
Environment.SetEnvironmentVariable(ClientTelemetryOptions.EnvPropsClientTelemetryEndpoint, config.TelemetryEndpoint);
Environment.SetEnvironmentVariable(ClientTelemetryOptions.EnvPropsClientTelemetrySchedulingInSeconds, config.TelemetryScheduleInSeconds);
}

private static IMetricsRoot ConfigureReporting(
CTLConfig config,
ILogger logger)
Expand Down
5 changes: 4 additions & 1 deletion Microsoft.Azure.Cosmos/src/ConnectionPolicy.cs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -48,7 +48,7 @@ public ConnectionPolicy()
this.MaxConnectionLimit = defaultMaxConcurrentConnectionLimit;
this.RetryOptions = new RetryOptions();
this.EnableReadRequestsFallback = null;
this.EnableClientTelemetry = ClientTelemetryOptions.IsClientTelemetryEnabled();
this.EnableClientTelemetry = false; // by default feature flag is off
this.ServerCertificateCustomValidationCallback = null;
}

Expand Down Expand Up @@ -211,6 +211,9 @@ public bool EnableTcpConnectionEndpointRediscovery
set;
}

/// <summary>
/// Gets or sets the flag to enable client telemetry feature.
/// </summary>
internal bool EnableClientTelemetry
{
get;
Expand Down
4 changes: 3 additions & 1 deletion Microsoft.Azure.Cosmos/src/Handler/DiagnosticsHandler.cs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,7 @@

namespace Microsoft.Azure.Cosmos.Handlers
{
using System;
using System.Threading;
using System.Threading.Tasks;
using Microsoft.Azure.Cosmos.Handler;
Expand All @@ -25,7 +26,8 @@ public override async Task<ResponseMessage> SendAsync(
ResponseMessage responseMessage = await base.SendAsync(request, cancellationToken);

// Record the diagnostics on the response to get the CPU of when the request was executing
SystemUsageHistory systemUsageHistory = DiagnosticsHandlerHelper.Instance.GetDiagnosticsSystemHistory();
SystemUsageHistory systemUsageHistory = DiagnosticsHandlerHelper.GetInstance().GetDiagnosticsSystemHistory();

if (systemUsageHistory != null)
{
request.Trace.AddDatum(
Expand Down
59 changes: 45 additions & 14 deletions Microsoft.Azure.Cosmos/src/Handler/DiagnosticsHandlerHelper.cs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -8,7 +8,6 @@ namespace Microsoft.Azure.Cosmos.Handler
using System.Collections.Generic;
using Documents.Rntbd;
using Microsoft.Azure.Cosmos.Core.Trace;
using Microsoft.Azure.Cosmos.Telemetry;

/// <summary>
/// This is a helper class that creates a single static instance to avoid each
Expand All @@ -22,34 +21,69 @@ internal class DiagnosticsHandlerHelper
private const string Telemetrykey = "telemetry";

public static readonly TimeSpan DiagnosticsRefreshInterval = TimeSpan.FromSeconds(10);
private readonly SystemUsageRecorder diagnosticSystemUsageRecorder = new SystemUsageRecorder(
private static readonly SystemUsageRecorder DiagnosticSystemUsageRecorder = new SystemUsageRecorder(
identifier: Diagnostickey,
historyLength: 6,
refreshInterval: DiagnosticsHandlerHelper.DiagnosticsRefreshInterval);

private static readonly TimeSpan ClientTelemetryRefreshInterval = TimeSpan.FromSeconds(5);
private readonly SystemUsageRecorder telemetrySystemUsageRecorder = new SystemUsageRecorder(
private static readonly SystemUsageRecorder TelemetrySystemUsageRecorder = new SystemUsageRecorder(
identifier: Telemetrykey,
historyLength: 120,
refreshInterval: DiagnosticsHandlerHelper.ClientTelemetryRefreshInterval);

private static bool isDiagnosticsMonitoringEnabled = false;
private static bool isTelemetryMonitoringEnabled = false;

/// <summary>
/// Singleton to make sure only one instance of DiagnosticHandlerHelper is there.
/// The system usage collection is disabled for internal builds so it is set to null to avoid
/// compute for accidentally creating an instance or trying to use it.
/// </summary>
public static readonly DiagnosticsHandlerHelper Instance =
private static DiagnosticsHandlerHelper Instance =
#if INTERNAL
null;
#else
new DiagnosticsHandlerHelper();
#endif

private static bool isDiagnosticsMonitoringEnabled;
private static bool isTelemetryMonitoringEnabled;

private readonly SystemUsageMonitor systemUsageMonitor = null;

public static DiagnosticsHandlerHelper GetInstance()
{
return DiagnosticsHandlerHelper.Instance;
}

/// <summary>
/// Restart the monitor with client telemetry recorder if telemetry is enabled
/// </summary>
/// <param name="isClientTelemetryEnabled"></param>
public static void Refresh(bool isClientTelemetryEnabled)
{
if (isClientTelemetryEnabled != DiagnosticsHandlerHelper.isTelemetryMonitoringEnabled)
{
DiagnosticsHandlerHelper.Instance.StopSystemMonitor();

// Update telemetry flag
DiagnosticsHandlerHelper.isTelemetryMonitoringEnabled = isClientTelemetryEnabled;

// Create new instance, it will start a new system monitor job
DiagnosticsHandlerHelper.Instance = new DiagnosticsHandlerHelper();
}
}

private void StopSystemMonitor()
{
try
{
this.systemUsageMonitor?.Dispose();
}
catch (ObjectDisposedException ex)
{
DefaultTrace.TraceError($"Error while stopping system usage monitor. {0} ", ex);
}
}

/// <summary>
/// Start System Usage Monitor with Diagnostic and Telemetry Recorder if Telemetry is enabled
/// Otherwise Start System Usage Monitor with only Diagnostic Recorder
Expand All @@ -61,16 +95,14 @@ private DiagnosticsHandlerHelper()
// If the CPU monitor fails for some reason don't block the application
try
{
DiagnosticsHandlerHelper.isTelemetryMonitoringEnabled = ClientTelemetryOptions.IsClientTelemetryEnabled();

List<SystemUsageRecorder> recorders = new List<SystemUsageRecorder>()
{
this.diagnosticSystemUsageRecorder,
DiagnosticsHandlerHelper.DiagnosticSystemUsageRecorder,
};

if (DiagnosticsHandlerHelper.isTelemetryMonitoringEnabled)
{
recorders.Add(this.telemetrySystemUsageRecorder);
recorders.Add(DiagnosticsHandlerHelper.TelemetrySystemUsageRecorder);
}

this.systemUsageMonitor = SystemUsageMonitor.CreateAndStart(recorders);
Expand All @@ -82,7 +114,6 @@ private DiagnosticsHandlerHelper()
DefaultTrace.TraceError(ex.Message);

DiagnosticsHandlerHelper.isDiagnosticsMonitoringEnabled = false;
DiagnosticsHandlerHelper.isTelemetryMonitoringEnabled = false;
}
}

Expand All @@ -99,7 +130,7 @@ public SystemUsageHistory GetDiagnosticsSystemHistory()

try
{
return this.diagnosticSystemUsageRecorder.Data;
return DiagnosticsHandlerHelper.DiagnosticSystemUsageRecorder.Data;
}
catch (Exception ex)
{
Expand All @@ -123,7 +154,7 @@ public SystemUsageHistory GetClientTelemetrySystemHistory()

try
{
return this.telemetrySystemUsageRecorder.Data;
return DiagnosticsHandlerHelper.TelemetrySystemUsageRecorder.Data;
}
catch (Exception ex)
{
Expand Down
Loading

0 comments on commit 11ecfb9

Please sign in to comment.