Implement Broker Redirects for Session and Messages

src/Runner.Common/BrokerServer.cs

@@ -17,7 +17,10 @@ public interface IBrokerServer : IRunnerService
     {
         Task ConnectAsync(Uri serverUrl, VssCredentials credentials);
 
-        Task<TaskAgentMessage> GetRunnerMessageAsync(CancellationToken token, TaskAgentStatus status, string version, string os, string architecture, bool disableUpdate);
+        Task<TaskAgentSession> CreateSessionAsync(TaskAgentSession session, CancellationToken cancellationToken);
+        Task DeleteSessionAsync(CancellationToken cancellationToken);
+
+        Task<TaskAgentMessage> GetRunnerMessageAsync(Guid? sessionId, TaskAgentStatus status, string version, string os, string architecture, bool disableUpdate, CancellationToken token);
     }
 
     public sealed class BrokerServer : RunnerService, IBrokerServer
@@ -44,13 +47,27 @@ private void CheckConnection()
             }
         }
 
-        public Task<TaskAgentMessage> GetRunnerMessageAsync(CancellationToken cancellationToken, TaskAgentStatus status, string version, string os, string architecture, bool disableUpdate)
+        public async Task<TaskAgentSession> CreateSessionAsync(TaskAgentSession session, CancellationToken cancellationToken)
         {
             CheckConnection();
-            var jobMessage = RetryRequest<TaskAgentMessage>(
-                async () => await _brokerHttpClient.GetRunnerMessageAsync(version, status, os, architecture, disableUpdate, cancellationToken), cancellationToken);
+            var jobMessage = await _brokerHttpClient.CreateSessionAsync(session, cancellationToken);
 
             return jobMessage;
         }
+
+        public Task<TaskAgentMessage> GetRunnerMessageAsync(Guid? sessionId, TaskAgentStatus status, string version, string os, string architecture, bool disableUpdate, CancellationToken cancellationToken)
+        {
+            CheckConnection();
+            var brokerSession = RetryRequest<TaskAgentMessage>(
+                async () => await _brokerHttpClient.GetRunnerMessageAsync(sessionId, version, status, os, architecture, disableUpdate, cancellationToken), cancellationToken);
+
+            return brokerSession;
+        }
+
+        public async Task DeleteSessionAsync(CancellationToken cancellationToken)
+        {
+            CheckConnection();
+            await _brokerHttpClient.DeleteSessionAsync(cancellationToken);
+        }
     }
 }

src/Runner.Listener/BrokerMessageListener.cs

@@ -24,7 +24,15 @@ public sealed class BrokerMessageListener : RunnerService, IMessageListener
         private TimeSpan _getNextMessageRetryInterval;
         private TaskAgentStatus runnerStatus = TaskAgentStatus.Online;
         private CancellationTokenSource _getMessagesTokenSource;
+        private VssCredentials _creds;
+        private TaskAgentSession _session;
         private IBrokerServer _brokerServer;
+        private readonly Dictionary<string, int> _sessionCreationExceptionTracker = new();
+        private bool _accessTokenRevoked = false;
+        private readonly TimeSpan _sessionCreationRetryInterval = TimeSpan.FromSeconds(30);
+        private readonly TimeSpan _sessionConflictRetryLimit = TimeSpan.FromMinutes(4);
+        private readonly TimeSpan _clockSkewRetryLimit = TimeSpan.FromMinutes(30);
+
 
         public override void Initialize(IHostContext hostContext)
         {
@@ -36,13 +44,134 @@ public override void Initialize(IHostContext hostContext)
 
         public async Task<Boolean> CreateSessionAsync(CancellationToken token)
         {
-            await RefreshBrokerConnection();
-            return await Task.FromResult(true);
+            Trace.Entering();
+
+            // Settings
+            var configManager = HostContext.GetService<IConfigurationManager>();
+            _settings = configManager.LoadSettings();
+            var serverUrl = _settings.ServerUrlV2;
+            Trace.Info(_settings);
+
+            if (string.IsNullOrEmpty(_settings.ServerUrlV2))
+            {
+                throw new InvalidOperationException("ServerUrlV2 is not set");
+            }
+
+            // Create connection.
+            Trace.Info("Loading Credentials");
+            var credMgr = HostContext.GetService<ICredentialManager>();
+            _creds = credMgr.LoadCredentials();
+
+            var agent = new TaskAgentReference
+            {
+                Id = _settings.AgentId,
+                Name = _settings.AgentName,
+                Version = BuildConstants.RunnerPackage.Version,
+                OSDescription = RuntimeInformation.OSDescription,
+            };
+            string sessionName = $"{Environment.MachineName ?? "RUNNER"}";
+            var taskAgentSession = new TaskAgentSession(sessionName, agent);
+
+            string errorMessage = string.Empty;
+            bool encounteringError = false;
+
+            while (true)
+            {
+                token.ThrowIfCancellationRequested();
+                Trace.Info($"Attempt to create session.");
+                try
+                {
+                    Trace.Info("Connecting to the Broker Server...");
+                    await _brokerServer.ConnectAsync(new Uri(serverUrl), _creds);
+                    Trace.Info("VssConnection created");
+
+                    _term.WriteLine();
+                    _term.WriteSuccessMessage("Connected to GitHub");
+                    _term.WriteLine();
+
+                    _session = await _brokerServer.CreateSessionAsync(taskAgentSession, token);
+
+                    Trace.Info($"Session created.");
+                    if (encounteringError)
+                    {
+                        _term.WriteLine($"{DateTime.UtcNow:u}: Runner reconnected.");
+                        _sessionCreationExceptionTracker.Clear();
+                        encounteringError = false;
+                    }
+
+                    return true;
+                }
+                catch (OperationCanceledException) when (token.IsCancellationRequested)
+                {
+                    Trace.Info("Session creation has been cancelled.");
+                    throw;
+                }
+                catch (TaskAgentAccessTokenExpiredException)
+                {
+                    Trace.Info("Runner OAuth token has been revoked. Session creation failed.");
+                    _accessTokenRevoked = true;
+                    throw;
+                }
+                catch (Exception ex)
+                {
+                    Trace.Error("Catch exception during create session.");
+                    Trace.Error(ex);
+
+                    if (ex is VssOAuthTokenRequestException vssOAuthEx && _creds.Federated is VssOAuthCredential vssOAuthCred)
+                    {
+                        // "invalid_client" means the runner registration has been deleted from the server.
+                        if (string.Equals(vssOAuthEx.Error, "invalid_client", StringComparison.OrdinalIgnoreCase))
+                        {
+                            _term.WriteError("Failed to create a session. The runner registration has been deleted from the server, please re-configure. Runner registrations are automatically deleted for runners that have not connected to the service recently.");
+                            return false;
+                        }
+
+                        // Check whether we get 401 because the runner registration already removed by the service.
+                        // If the runner registration get deleted, we can't exchange oauth token.
+                        Trace.Error("Test oauth app registration.");
+                        var oauthTokenProvider = new VssOAuthTokenProvider(vssOAuthCred, new Uri(serverUrl));
+                        var authError = await oauthTokenProvider.ValidateCredentialAsync(token);
+                        if (string.Equals(authError, "invalid_client", StringComparison.OrdinalIgnoreCase))
+                        {
+                            _term.WriteError("Failed to create a session. The runner registration has been deleted from the server, please re-configure. Runner registrations are automatically deleted for runners that have not connected to the service recently.");
+                            return false;
+                        }
+                    }
+
+                    if (!IsSessionCreationExceptionRetriable(ex))
+                    {
+                        _term.WriteError($"Failed to create session. {ex.Message}");
+                        return false;
+                    }
+
+                    if (!encounteringError) //print the message only on the first error
+                    {
+                        _term.WriteError($"{DateTime.UtcNow:u}: Runner connect error: {ex.Message}. Retrying until reconnected.");
+                        encounteringError = true;
+                    }
+
+                    Trace.Info("Sleeping for {0} seconds before retrying.", _sessionCreationRetryInterval.TotalSeconds);
+                    await HostContext.Delay(_sessionCreationRetryInterval, token);
+                }
+            }
         }
 
         public async Task DeleteSessionAsync()
         {
-            await Task.CompletedTask;
+            if (_session != null && _session.SessionId != Guid.Empty)
+            {
+                if (!_accessTokenRevoked)
+                {
+                    using (var ts = new CancellationTokenSource(TimeSpan.FromSeconds(30)))
+                    {
+                        await _brokerServer.DeleteSessionAsync(ts.Token);
+                    }
+                }
+                else
+                {
+                    Trace.Warning("Runner OAuth token has been revoked. Skip deleting session.");
+                }
+            }
         }
 
         public void OnJobStatus(object sender, JobStatusEventArgs e)
@@ -73,12 +202,13 @@ public async Task<TaskAgentMessage> GetNextMessageAsync(CancellationToken token)
                 _getMessagesTokenSource = CancellationTokenSource.CreateLinkedTokenSource(token);
                 try
                 {
-                    message = await _brokerServer.GetRunnerMessageAsync(_getMessagesTokenSource.Token,
+                    message = await _brokerServer.GetRunnerMessageAsync(_session.SessionId,
                                                                         runnerStatus,
                                                                         BuildConstants.RunnerPackage.Version,
                                                                         VarUtil.OS,
                                                                         VarUtil.OSArchitecture,
-                                                                        _settings.DisableUpdate);
+                                                                        _settings.DisableUpdate,
+                                                                        _getMessagesTokenSource.Token);
 
                     if (message == null)
                     {
@@ -196,12 +326,84 @@ ex is AccessDeniedException ||
             }
         }
 
+        private bool IsSessionCreationExceptionRetriable(Exception ex)
+        {
+            if (ex is TaskAgentNotFoundException)
+            {
+                Trace.Info("The runner no longer exists on the server. Stopping the runner.");
+                _term.WriteError("The runner no longer exists on the server. Please reconfigure the runner.");
+                return false;
+            }
+            else if (ex is TaskAgentSessionConflictException)
+            {
+                Trace.Info("The session for this runner already exists.");
+                _term.WriteError("A session for this runner already exists.");
+                if (_sessionCreationExceptionTracker.ContainsKey(nameof(TaskAgentSessionConflictException)))
+                {
+                    _sessionCreationExceptionTracker[nameof(TaskAgentSessionConflictException)]++;
+                    if (_sessionCreationExceptionTracker[nameof(TaskAgentSessionConflictException)] * _sessionCreationRetryInterval.TotalSeconds >= _sessionConflictRetryLimit.TotalSeconds)
+                    {
+                        Trace.Info("The session conflict exception have reached retry limit.");
+                        _term.WriteError($"Stop retry on SessionConflictException after retried for {_sessionConflictRetryLimit.TotalSeconds} seconds.");
+                        return false;
+                    }
+                }
+                else
+                {
+                    _sessionCreationExceptionTracker[nameof(TaskAgentSessionConflictException)] = 1;
+                }
+
+                Trace.Info("The session conflict exception haven't reached retry limit.");
+                return true;
+            }
+            else if (ex is VssOAuthTokenRequestException && ex.Message.Contains("Current server time is"))
+            {
+                Trace.Info("Local clock might be skewed.");
+                _term.WriteError("The local machine's clock may be out of sync with the server time by more than five minutes. Please sync your clock with your domain or internet time and try again.");
+                if (_sessionCreationExceptionTracker.ContainsKey(nameof(VssOAuthTokenRequestException)))
+                {
+                    _sessionCreationExceptionTracker[nameof(VssOAuthTokenRequestException)]++;
+                    if (_sessionCreationExceptionTracker[nameof(VssOAuthTokenRequestException)] * _sessionCreationRetryInterval.TotalSeconds >= _clockSkewRetryLimit.TotalSeconds)
+                    {
+                        Trace.Info("The OAuth token request exception have reached retry limit.");
+                        _term.WriteError($"Stopped retrying OAuth token request exception after {_clockSkewRetryLimit.TotalSeconds} seconds.");
+                        return false;
+                    }
+                }
+                else
+                {
+                    _sessionCreationExceptionTracker[nameof(VssOAuthTokenRequestException)] = 1;
+                }
+
+                Trace.Info("The OAuth token request exception haven't reached retry limit.");
+                return true;
+            }
+            else if (ex is TaskAgentPoolNotFoundException ||
+                     ex is AccessDeniedException ||
+                     ex is VssUnauthorizedException)
+            {
+                Trace.Info($"Non-retriable exception: {ex.Message}");
+                return false;
+            }
+
+            else if (ex is InvalidOperationException)
+            {
+                Trace.Info($"Non-retriable exception: {ex.Message}");
+                return false;
+            }
+            else
+            {
+                Trace.Info($"Retriable exception: {ex.Message}");
+                return true;
+            }
+        }
+
         private async Task RefreshBrokerConnection()
         {
             var configManager = HostContext.GetService<IConfigurationManager>();
             _settings = configManager.LoadSettings();
 
-            if (_settings.ServerUrlV2 == null)
+            if (string.IsNullOrEmpty(_settings.ServerUrlV2))
             {
                 throw new InvalidOperationException("ServerUrlV2 is not set");
             }