coder · stirby · May 15, 2025 · May 12, 2025 · May 9, 2025 · May 14, 2025
diff --git a/agent/agent.go b/agent/agent.go
@@ -363,9 +363,11 @@ func (a *agent) runLoop() {
 		if ctx.Err() != nil {
 			// Context canceled errors may come from websocket pings, so we
 			// don't want to use `errors.Is(err, context.Canceled)` here.
+			a.logger.Warn(ctx, "runLoop exited with error", slog.Error(ctx.Err()))
 			return
 		}
 		if a.isClosed() {
+			a.logger.Warn(ctx, "runLoop exited because agent is closed")
 			return
 		}
 		if errors.Is(err, io.EOF) {
@@ -1046,7 +1048,11 @@ func (a *agent) run() (retErr error) {
 		return a.statsReporter.reportLoop(ctx, aAPI)
 	})
 
-	return connMan.wait()
+	err = connMan.wait()
+	if err != nil {
+		a.logger.Info(context.Background(), "connection manager errored", slog.Error(err))
+	}
+	return err
 }
 
 // handleManifest returns a function that fetches and processes the manifest

diff --git a/cli/agent.go b/cli/agent.go
@@ -25,6 +25,8 @@ import (
 	"cdr.dev/slog/sloggers/sloghuman"
 	"cdr.dev/slog/sloggers/slogjson"
 	"cdr.dev/slog/sloggers/slogstackdriver"
+	"github.com/coder/serpent"
+
 	"github.com/coder/coder/v2/agent"
 	"github.com/coder/coder/v2/agent/agentexec"
 	"github.com/coder/coder/v2/agent/agentssh"
@@ -33,7 +35,6 @@ import (
 	"github.com/coder/coder/v2/cli/clilog"
 	"github.com/coder/coder/v2/codersdk"
 	"github.com/coder/coder/v2/codersdk/agentsdk"
-	"github.com/coder/serpent"
 )
 
 func (r *RootCmd) workspaceAgent() *serpent.Command {
@@ -62,8 +63,10 @@ func (r *RootCmd) workspaceAgent() *serpent.Command {
 		// This command isn't useful to manually execute.
 		Hidden: true,
 		Handler: func(inv *serpent.Invocation) error {
-			ctx, cancel := context.WithCancel(inv.Context())
-			defer cancel()
+			ctx, cancel := context.WithCancelCause(inv.Context())
+			defer func() {
+				cancel(xerrors.New("agent exited"))
+			}()
 
 			var (
 				ignorePorts = map[int]string{}
@@ -280,7 +283,6 @@ func (r *RootCmd) workspaceAgent() *serpent.Command {
 				return xerrors.Errorf("add executable to $PATH: %w", err)
 			}
 
-			prometheusRegistry := prometheus.NewRegistry()
 			subsystemsRaw := inv.Environ.Get(agent.EnvAgentSubsystem)
 			subsystems := []codersdk.AgentSubsystem{}
 			for _, s := range strings.Split(subsystemsRaw, ",") {
@@ -324,45 +326,70 @@ func (r *RootCmd) workspaceAgent() *serpent.Command {
 				logger.Info(ctx, "agent devcontainer detection not enabled")
 			}
 
-			agnt := agent.New(agent.Options{
-				Client:        client,
-				Logger:        logger,
-				LogDir:        logDir,
-				ScriptDataDir: scriptDataDir,
-				// #nosec G115 - Safe conversion as tailnet listen port is within uint16 range (0-65535)
-				TailnetListenPort: uint16(tailnetListenPort),
-				ExchangeToken: func(ctx context.Context) (string, error) {
-					if exchangeToken == nil {
-						return client.SDK.SessionToken(), nil
-					}
-					resp, err := exchangeToken(ctx)
-					if err != nil {
-						return "", err
-					}
-					client.SetSessionToken(resp.SessionToken)
-					return resp.SessionToken, nil
-				},
-				EnvironmentVariables: environmentVariables,
-				IgnorePorts:          ignorePorts,
-				SSHMaxTimeout:        sshMaxTimeout,
-				Subsystems:           subsystems,
-
-				PrometheusRegistry: prometheusRegistry,
-				BlockFileTransfer:  blockFileTransfer,
-				Execer:             execer,
-
-				ExperimentalDevcontainersEnabled: experimentalDevcontainersEnabled,
-			})
-
-			promHandler := agent.PrometheusMetricsHandler(prometheusRegistry, logger)
-			prometheusSrvClose := ServeHandler(ctx, logger, promHandler, prometheusAddress, "prometheus")
-			defer prometheusSrvClose()
-
-			debugSrvClose := ServeHandler(ctx, logger, agnt.HTTPDebug(), debugAddress, "debug")
-			defer debugSrvClose()
-
-			<-ctx.Done()
-			return agnt.Close()
+			reinitEvents := agentsdk.WaitForReinitLoop(ctx, logger, client)
+
+			var (
+				lastErr  error
+				mustExit bool
+			)
+			for {
+				prometheusRegistry := prometheus.NewRegistry()
+
+				agnt := agent.New(agent.Options{
+					Client:        client,
+					Logger:        logger,
+					LogDir:        logDir,
+					ScriptDataDir: scriptDataDir,
+					// #nosec G115 - Safe conversion as tailnet listen port is within uint16 range (0-65535)
+					TailnetListenPort: uint16(tailnetListenPort),
+					ExchangeToken: func(ctx context.Context) (string, error) {
+						if exchangeToken == nil {
+							return client.SDK.SessionToken(), nil
+						}
+						resp, err := exchangeToken(ctx)
+						if err != nil {
+							return "", err
+						}
+						client.SetSessionToken(resp.SessionToken)
+						return resp.SessionToken, nil
+					},
+					EnvironmentVariables: environmentVariables,
+					IgnorePorts:          ignorePorts,
+					SSHMaxTimeout:        sshMaxTimeout,
+					Subsystems:           subsystems,
+
+					PrometheusRegistry: prometheusRegistry,
+					BlockFileTransfer:  blockFileTransfer,
+					Execer:             execer,
+
+					ExperimentalDevcontainersEnabled: experimentalDevcontainersEnabled,
+				})
+
+				promHandler := agent.PrometheusMetricsHandler(prometheusRegistry, logger)
+				prometheusSrvClose := ServeHandler(ctx, logger, promHandler, prometheusAddress, "prometheus")
+
+				debugSrvClose := ServeHandler(ctx, logger, agnt.HTTPDebug(), debugAddress, "debug")
+
+				select {
+				case <-ctx.Done():
+					logger.Info(ctx, "agent shutting down", slog.Error(context.Cause(ctx)))
+					mustExit = true
+				case event := <-reinitEvents:
+					logger.Info(ctx, "agent received instruction to reinitialize",
+						slog.F("workspace_id", event.WorkspaceID), slog.F("reason", event.Reason))
+				}
+
+				lastErr = agnt.Close()
+				debugSrvClose()
+				prometheusSrvClose()
+
+				if mustExit {
+					break
+				}
+
+				logger.Info(ctx, "agent reinitializing")
+			}
+			return lastErr
 		},
 	}
 

diff --git a/cli/server.go b/cli/server.go
@@ -910,6 +910,37 @@ func (r *RootCmd) Server(newAPI func(context.Context, *coderd.Options) (*coderd.
 			options.StatsBatcher = batcher
 			defer closeBatcher()
 
+			// Manage notifications.
+			var (
+				notificationsCfg     = options.DeploymentValues.Notifications
+				notificationsManager *notifications.Manager
+			)
+
+			metrics := notifications.NewMetrics(options.PrometheusRegistry)
+			helpers := templateHelpers(options)
+
+			// The enqueuer is responsible for enqueueing notifications to the given store.
+			enqueuer, err := notifications.NewStoreEnqueuer(notificationsCfg, options.Database, helpers, logger.Named("notifications.enqueuer"), quartz.NewReal())
+			if err != nil {
+				return xerrors.Errorf("failed to instantiate notification store enqueuer: %w", err)
+			}
+			options.NotificationsEnqueuer = enqueuer
+
+			// The notification manager is responsible for:
+			//   - creating notifiers and managing their lifecycles (notifiers are responsible for dequeueing/sending notifications)
+			//   - keeping the store updated with status updates
+			notificationsManager, err = notifications.NewManager(notificationsCfg, options.Database, options.Pubsub, helpers, metrics, logger.Named("notifications.manager"))
+			if err != nil {
+				return xerrors.Errorf("failed to instantiate notification manager: %w", err)
+			}
+
+			// nolint:gocritic // We need to run the manager in a notifier context.
+			notificationsManager.Run(dbauthz.AsNotifier(ctx))
+
+			// Run report generator to distribute periodic reports.
+			notificationReportGenerator := reports.NewReportGenerator(ctx, logger.Named("notifications.report_generator"), options.Database, options.NotificationsEnqueuer, quartz.NewReal())
+			defer notificationReportGenerator.Close()
+
 			// We use a separate coderAPICloser so the Enterprise API
 			// can have its own close functions. This is cleaner
 			// than abstracting the Coder API itself.
@@ -957,37 +988,6 @@ func (r *RootCmd) Server(newAPI func(context.Context, *coderd.Options) (*coderd.
 				return xerrors.Errorf("write config url: %w", err)
 			}
 
-			// Manage notifications.
-			var (
-				notificationsCfg     = options.DeploymentValues.Notifications
-				notificationsManager *notifications.Manager
-			)
-
-			metrics := notifications.NewMetrics(options.PrometheusRegistry)
-			helpers := templateHelpers(options)
-
-			// The enqueuer is responsible for enqueueing notifications to the given store.
-			enqueuer, err := notifications.NewStoreEnqueuer(notificationsCfg, options.Database, helpers, logger.Named("notifications.enqueuer"), quartz.NewReal())
-			if err != nil {
-				return xerrors.Errorf("failed to instantiate notification store enqueuer: %w", err)
-			}
-			options.NotificationsEnqueuer = enqueuer
-
-			// The notification manager is responsible for:
-			//   - creating notifiers and managing their lifecycles (notifiers are responsible for dequeueing/sending notifications)
-			//   - keeping the store updated with status updates
-			notificationsManager, err = notifications.NewManager(notificationsCfg, options.Database, options.Pubsub, helpers, metrics, logger.Named("notifications.manager"))
-			if err != nil {
-				return xerrors.Errorf("failed to instantiate notification manager: %w", err)
-			}
-
-			// nolint:gocritic // We need to run the manager in a notifier context.
-			notificationsManager.Run(dbauthz.AsNotifier(ctx))
-
-			// Run report generator to distribute periodic reports.
-			notificationReportGenerator := reports.NewReportGenerator(ctx, logger.Named("notifications.report_generator"), options.Database, options.NotificationsEnqueuer, quartz.NewReal())
-			defer notificationReportGenerator.Close()
-
 			// Since errCh only has one buffered slot, all routines
 			// sending on it must be wrapped in a select/default to
 			// avoid leaving dangling goroutines waiting for the

diff --git a/cli/testdata/coder_provisioner_list_--output_json.golden b/cli/testdata/coder_provisioner_list_--output_json.golden
@@ -7,7 +7,7 @@
     "last_seen_at": "====[timestamp]=====",
     "name": "test",
     "version": "v0.0.0-devel",
-    "api_version": "1.4",
+    "api_version": "1.5",
     "provisioners": [
       "echo"
     ],

diff --git a/coderd/apidoc/docs.go b/coderd/apidoc/docs.go
diff --git a/coderd/apidoc/swagger.json b/coderd/apidoc/swagger.json