Client Telemetry And Diagnostics : Adds VMId or Unique Id in payload

Microsoft.Azure.Cosmos/src/DocumentClient.cs

@@ -22,6 +22,7 @@ namespace Microsoft.Azure.Cosmos
     using Microsoft.Azure.Cosmos.Query;
     using Microsoft.Azure.Cosmos.Query.Core.QueryPlan;
     using Microsoft.Azure.Cosmos.Routing;
+    using Microsoft.Azure.Cosmos.Telemetry;
     using Microsoft.Azure.Cosmos.Tracing;
     using Microsoft.Azure.Cosmos.Tracing.TraceData;
     using Microsoft.Azure.Documents;
@@ -897,6 +898,9 @@ internal virtual void Initialize(Uri serviceEndpoint,
                 this.sendingRequest,
                 this.receivedResponse);
 
+            // Loading VM Information (non blocking call and initialization won't fail if this call fails)
+            VmMetadataApiHandler.TryInitialize(this.httpClient);
+
             if (sessionContainer != null)
             {
                 this.sessionContainer = sessionContainer;

Microsoft.Azure.Cosmos/src/HttpClient/HttpTimeoutPolicyNoRetry.cs

@@ -0,0 +1,44 @@
+//------------------------------------------------------------
+// Copyright (c) Microsoft Corporation.  All rights reserved.
+//------------------------------------------------------------
+namespace Microsoft.Azure.Cosmos
+{
+    using System;
+    using System.Collections.Generic;
+    using System.Net.Http;
+    using System.Text;
+
+    internal sealed class HttpTimeoutPolicyNoRetry : HttpTimeoutPolicy
+    {
+        public static readonly HttpTimeoutPolicy Instance = new HttpTimeoutPolicyNoRetry();
+        private static readonly string Name = nameof(HttpTimeoutPolicyNoRetry);
+
+        private HttpTimeoutPolicyNoRetry()
+        {
+        }
+
+        private readonly IReadOnlyList<(TimeSpan requestTimeout, TimeSpan delayForNextRequest)> TimeoutsAndDelays = new List<(TimeSpan requestTimeout, TimeSpan delayForNextRequest)>();
+
+        public override string TimeoutPolicyName => HttpTimeoutPolicyNoRetry.Name;
+
+        public override TimeSpan MaximumRetryTimeLimit => TimeSpan.Zero;
+
+        public override int TotalRetryCount => 0;
+
+        public override IEnumerator<(TimeSpan requestTimeout, TimeSpan delayForNextRequest)> GetTimeoutEnumerator()
+        {
+            return this.TimeoutsAndDelays.GetEnumerator();
+        }
+
+        // Always Unsafe to retry
+        public override bool IsSafeToRetry(HttpMethod httpMethod)
+        {
+            return false;
+        }
+
+        public override bool ShouldRetryBasedOnResponse(HttpMethod requestHttpMethod, HttpResponseMessage responseMessage)
+        {
+            return false;
+        }
+    }
+}

Microsoft.Azure.Cosmos/src/Telemetry/ClientTelemetry.cs

@@ -32,7 +32,6 @@ internal class ClientTelemetry : IDisposable
         private static readonly TimeSpan observingWindow = ClientTelemetryOptions.GetScheduledTimeSpan();
 
         private readonly ClientTelemetryProperties clientTelemetryInfo;
-
         private readonly DocumentClient documentClient;
         private readonly CosmosHttpClient httpClient;
         private readonly AuthorizationTokenProvider tokenProvider;
@@ -135,17 +134,18 @@ private async Task EnrichAndSendAsync()
                         this.clientTelemetryInfo.GlobalDatabaseAccountName = accountProperties?.Id;
                     }
 
-                    // Load host information if not available (it caches the information)
-                    AzureVMMetadata azMetadata = await ClientTelemetryHelper.LoadAzureVmMetaDataAsync(this.httpClient);
-
-                    Compute vmInformation = azMetadata?.Compute;
+                    // Load host information from cache
+                    Compute vmInformation = VmMetadataApiHandler.GetMachineInfo();
                     if (vmInformation != null)
                     {
                         this.clientTelemetryInfo.ApplicationRegion = vmInformation.Location;
                         this.clientTelemetryInfo.HostEnvInfo = ClientTelemetryOptions.GetHostInformation(vmInformation);
+
                         //TODO: Set AcceleratingNetwork flag from instance metadata once it is available.
                     }
 
+                    this.clientTelemetryInfo.MachineId = VmMetadataApiHandler.GetMachineId();
+
                     await Task.Delay(observingWindow, this.cancellationTokenSource.Token);
 
                     // If cancellation is requested after the delay then return from here.
@@ -329,7 +329,7 @@ await this.tokenProvider.AddAuthorizationHeaderAsync(
 
                 using HttpResponseMessage response = await this.httpClient.SendHttpAsync(CreateRequestMessage,
                                                     ResourceType.Telemetry,
-                                                    HttpTimeoutPolicyDefault.Instance,
+                                                    HttpTimeoutPolicyNoRetry.Instance,
                                                     null,
                                                     this.cancellationTokenSource.Token);
 
@@ -369,7 +369,7 @@ public void Dispose()
         {
             this.cancellationTokenSource.Cancel();
             this.cancellationTokenSource.Dispose();
-
+            
             this.telemetryTask = null;
         }
     }

Microsoft.Azure.Cosmos/src/Telemetry/ClientTelemetryHelper.cs

@@ -16,10 +16,6 @@ namespace Microsoft.Azure.Cosmos.Telemetry
 
     internal static class ClientTelemetryHelper
     {
-        internal static AzureVMMetadata azMetadata = null;
-
-        private static readonly Uri vmMetadataEndpointUrl = ClientTelemetryOptions.GetVmMetadataUrl();
-
         /// <summary>
         /// Task to get Account Properties from cache if available otherwise make a network call.
         /// </summary>
@@ -42,50 +38,6 @@ internal static async Task<AccountProperties> SetAccountNameAsync(DocumentClient
             return null;
         }
 
-        /// <summary>
-        /// Task to collect virtual machine metadata information. using instance metedata service API.
-        /// ref: https://docs.microsoft.com/en-us/azure/virtual-machines/windows/instance-metadata-service?tabs=windows
-        /// Collects only application region and environment information
-        /// </summary>
-        /// <returns>Async Task</returns>
-        internal static async Task<AzureVMMetadata> LoadAzureVmMetaDataAsync(CosmosHttpClient httpClient)
-        {
-            if (azMetadata == null)
-            {
-                DefaultTrace.TraceVerbose("Getting VM Metadata Information for Telemetry.");
-                try
-                {
-                    static ValueTask<HttpRequestMessage> CreateRequestMessage()
-                    {
-                        HttpRequestMessage request = new HttpRequestMessage()
-                        {
-                            RequestUri = vmMetadataEndpointUrl,
-                            Method = HttpMethod.Get,
-                        };
-                        request.Headers.Add("Metadata", "true");
-
-                        return new ValueTask<HttpRequestMessage>(request);
-                    }
-
-                    using HttpResponseMessage httpResponseMessage = await httpClient
-                        .SendHttpAsync(createRequestMessageAsync: CreateRequestMessage,
-                        resourceType: ResourceType.Telemetry,
-                        timeoutPolicy: HttpTimeoutPolicyDefault.Instance,
-                        clientSideRequestStatistics: null,
-                        cancellationToken: new CancellationToken()); // Do not want to cancel the whole process if this call fails
-
-                    azMetadata = await ClientTelemetryOptions.ProcessResponseAsync(httpResponseMessage);
-
-                }
-                catch (Exception ex)
-                {
-                    DefaultTrace.TraceError("Exception in LoadAzureVmMetaDataAsync() {0}", ex.Message);
-                }
-            }
-
-            return azMetadata;
-        }
-
         /// <summary>
         /// Record System Usage and update passed system Info collection. Right now, it collects following metrics
         /// 1) CPU Usage
@@ -118,7 +70,7 @@ internal static void RecordSystemUsage(
             {
                 systemInfoCollection.Add(TelemetrySystemUsage.GetTcpConnectionCount(systemUsageHistory.Values));
             }
-            
+
         }
 
         /// <summary>
@@ -186,5 +138,6 @@ internal static string GetContactedRegions(CosmosDiagnostics cosmosDiagnostics)
 
             return regionsContacted.ToString();
         }
+
     }
 }

Microsoft.Azure.Cosmos/src/Telemetry/ClientTelemetryOptions.cs

@@ -4,12 +4,9 @@
 namespace Microsoft.Azure.Cosmos.Telemetry
 {
     using System;
-    using System.Net.Http;
-    using System.Threading.Tasks;
     using Microsoft.Azure.Cosmos.Core.Trace;
     using Microsoft.Azure.Documents;
     using Newtonsoft.Json;
-    using Newtonsoft.Json.Linq;
 
     internal static class ClientTelemetryOptions
     {
@@ -73,7 +70,6 @@ internal static class ClientTelemetryOptions
         internal const string IsThreadStarvingName = "SystemPool_IsThreadStarving_True";
         internal const string IsThreadStarvingUnit = "Count";
 
-        internal const string DefaultVmMetadataUrL = "http://169.254.169.254/metadata/instance?api-version=2020-06-01";
         internal const double DefaultTimeStampInSeconds = 600;
         internal const double Percentile50 = 50.0;
         internal const double Percentile90 = 90.0;
@@ -92,7 +88,6 @@ internal static class ClientTelemetryOptions
 
         internal static readonly JsonSerializerSettings JsonSerializerSettings = new JsonSerializerSettings { NullValueHandling = NullValueHandling.Ignore };
 
-        private static Uri vmMetadataUrl;
         private static Uri clientTelemetryEndpoint;
         private static string environmentName;
         private static TimeSpan scheduledTimeSpan = TimeSpan.Zero;
@@ -108,22 +103,6 @@ internal static bool IsClientTelemetryEnabled()
             return isTelemetryEnabled;
         }
 
-        internal static Uri GetVmMetadataUrl()
-        {
-            if (vmMetadataUrl == null)
-            {
-                string vmMetadataUrlProp = ConfigurationManager.GetEnvironmentVariable<string>(
-                   EnvPropsClientTelemetryVmMetadataUrl, DefaultVmMetadataUrL);
-                if (!String.IsNullOrEmpty(vmMetadataUrlProp))
-                {
-                    vmMetadataUrl = new Uri(vmMetadataUrlProp);
-                }
-
-                DefaultTrace.TraceInformation($"VM metadata URL for telemetry {vmMetadataUrlProp}");
-            }
-            return vmMetadataUrl;
-        }
-
         internal static TimeSpan GetScheduledTimeSpan()
         {
             if (scheduledTimeSpan.Equals(TimeSpan.Zero))
@@ -154,16 +133,6 @@ internal static TimeSpan GetScheduledTimeSpan()
             return scheduledTimeSpan;
         }
 
-        internal static async Task<AzureVMMetadata> ProcessResponseAsync(HttpResponseMessage httpResponseMessage)
-        {
-            if (httpResponseMessage.Content == null)
-            {
-                return null;
-            }
-            string jsonVmInfo = await httpResponseMessage.Content.ReadAsStringAsync();
-            return JObject.Parse(jsonVmInfo).ToObject<AzureVMMetadata>();
-        }
-
         internal static string GetHostInformation(Compute vmInformation)
         {
             return String.Concat(vmInformation?.OSType, "|",

Microsoft.Azure.Cosmos/src/Telemetry/ClientTelemetryProperties.cs

@@ -19,6 +19,9 @@ internal sealed class ClientTelemetryProperties
         [JsonProperty(PropertyName = "clientId")]
         internal string ClientId { get; }
 
+        [JsonProperty(PropertyName = "machineId")]
+        internal string MachineId { get; set; }
+
         [JsonProperty(PropertyName = "processId")]
         internal string ProcessId { get; }
 
@@ -97,7 +100,8 @@ public ClientTelemetryProperties(string dateTimeUtc,
             IReadOnlyList<string> preferredRegions,
             List<SystemInfo> systemInfo,
             List<OperationInfo> cacheRefreshInfo,
-            List<OperationInfo> operationInfo)
+            List<OperationInfo> operationInfo,
+            string machineId)
         {
             this.DateTimeUtc = dateTimeUtc;
             this.ClientId = clientId;
@@ -112,6 +116,7 @@ public ClientTelemetryProperties(string dateTimeUtc,
             this.CacheRefreshInfo = cacheRefreshInfo;
             this.OperationInfo = operationInfo;
             this.PreferredRegions = preferredRegions;
+            this.MachineId = machineId;
         }
     }
 }

Microsoft.Azure.Cosmos/src/Telemetry/Compute.cs

@@ -10,13 +10,21 @@ namespace Microsoft.Azure.Cosmos.Telemetry
     [Serializable]
     internal sealed class Compute
     {
-        public Compute(string location, string sKU, string azEnvironment, string oSType, string vMSize)
+        [JsonConstructor]
+        public Compute(
+            string vMId, 
+            string location, 
+            string sKU, 
+            string azEnvironment, 
+            string oSType, 
+            string vMSize)
         {
             this.Location = location;
             this.SKU = sKU;
             this.AzEnvironment = azEnvironment;
             this.OSType = oSType;
             this.VMSize = vMSize;
+            this.VMId = "vmId:" + vMId;
         }
 
         [JsonProperty(PropertyName = "location")]
@@ -33,6 +41,9 @@ public Compute(string location, string sKU, string azEnvironment, string oSType,
 
         [JsonProperty(PropertyName = "vmSize")]
         internal string VMSize { get; }
+
+        [JsonProperty(PropertyName = "vmId")]
+        internal string VMId { get; }
     }
 
 }