fix: improve studio reliability and agent heartbeat

- Fix environment variable: TF_CLIENT_INFO_PATH -> TF_CONNECTION_INFO_PATH
  to match tensor-fusion-worker expectations
- Add TF_CONNECTION_INFO_PATH to studio containers with connections directory mount
  at /var/run/tensor-fusion/connections to enable worker connection tracking
- Add default "sleep infinity" command for studio containers to prevent immediate exit
  when no command is specified
- Reduce agent heartbeat interval from 60s to 30s for faster dashboard updates
  and more responsive worker status reporting

These changes fix the "studio failed to stay running" issue and improve
real-time status synchronization between ggo and the web dashboard.

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

Author: andystime

go.mod

@@ -152,7 +152,7 @@ require (
 	github.com/pmezard/go-difflib v1.0.1-0.20181226105442-5d4384ee4fb2 // indirect
 	github.com/spf13/pflag v1.0.10 // indirect
 	golang.org/x/net v0.50.0 // indirect
-	golang.org/x/sys v0.41.0 // indirect
+	golang.org/x/sys v0.41.0
 	golang.org/x/term v0.40.0
 	golang.org/x/time v0.14.0 // indirect
 	gopkg.in/yaml.v3 v3.0.1 // indirect

internal/agent/agent.go

@@ -23,11 +23,11 @@ import (
 )
 
 const (
-	statusReportInterval = 1 * time.Minute
+	statusReportInterval = 30 * time.Second
 	forceRefreshInterval = 6 * time.Hour
 	// EnvConnectionInfoPath is the environment variable name for connection info directory path
-	// Workers should write their connections to: {TF_CLIENT_INFO_PATH}/{workerID}.txt
-	EnvConnectionInfoPath = "TF_CLIENT_INFO_PATH"
+	// Workers should write their connections to: {TF_CONNECTION_INFO_PATH}/{workerID}.txt
+	EnvConnectionInfoPath = "TF_CONNECTION_INFO_PATH"
 
 	// Worker status constants
 	workerStatusRunning  = "running"
@@ -230,8 +230,8 @@ func (a *Agent) Start() error {
 		klog.Warningf("Failed to create connections directory: path=%s error=%v", a.connectionsDir, err)
 	}
 
-	// Set TF_CLIENT_INFO_PATH environment variable for worker processes
-	// Workers will write connection info to: {TF_CLIENT_INFO_PATH}/{workerID}.txt
+	// Set TF_CONNECTION_INFO_PATH environment variable for worker processes
+	// Workers will write connection info to: {TF_CONNECTION_INFO_PATH}/{workerID}.txt
 	// Each worker has its own file with format: clientIP,clientPort,clientPID (one per line)
 	if err := os.Setenv(EnvConnectionInfoPath, a.connectionsDir); err != nil {
 		klog.Warningf("Failed to set %s env var: error=%v", EnvConnectionInfoPath, err)

internal/studio/backend_docker.go

@@ -264,7 +264,12 @@ func (b *DockerBackend) Create(ctx context.Context, opts *CreateOptions) (*Envir
 
 	// Add command args (supplements ENTRYPOINT or overrides CMD)
 	// FormatContainerCommand handles wrapping single shell commands with "sh -c"
-	if formattedCmd := FormatContainerCommand(opts.Command); len(formattedCmd) > 0 {
+	// If no command is provided, use "sleep infinity" to keep container running
+	cmdToUse := opts.Command
+	if len(cmdToUse) == 0 {
+		cmdToUse = []string{"sleep", "infinity"}
+	}
+	if formattedCmd := FormatContainerCommand(cmdToUse); len(formattedCmd) > 0 {
 		args = append(args, formattedCmd...)
 	}
 

internal/studio/env.go

@@ -200,6 +200,21 @@ func SetupGPUEnv(paths *platform.Paths, config *GPUEnvConfig) (*GPUEnvResult, er
 	result.EnvVars["TF_LOG_LEVEL"] = getEnvDefault("TF_LOG_LEVEL", "info")
 	result.EnvVars["TF_ENABLE_LOG"] = getEnvDefault("TF_ENABLE_LOG", "1")
 
+	// Get connections directory (for tensor-fusion-worker to write connection info)
+	connectionsDir := filepath.Join(paths.StateDir(), "connections")
+	if err := os.MkdirAll(connectionsDir, 0755); err != nil {
+		return nil, fmt.Errorf("failed to create connections directory: %w", err)
+	}
+
+	// Set TF_CONNECTION_INFO_PATH for tensor-fusion-worker
+	if !config.IsContainer {
+		// On host, use actual connections directory
+		result.EnvVars["TF_CONNECTION_INFO_PATH"] = connectionsDir
+	} else {
+		// In container, use mounted path
+		result.EnvVars["TF_CONNECTION_INFO_PATH"] = "/var/run/tensor-fusion/connections"
+	}
+
 	// Add cache path to PATH (for tensor-fusion-worker binary)
 	if !config.IsContainer {
 		// On host, prepend cache path to existing PATH
@@ -236,6 +251,13 @@ func SetupGPUEnv(paths *platform.Paths, config *GPUEnvConfig) (*GPUEnvResult, er
 
 	// Set up volume mounts for container mode
 	if config.IsContainer {
+		// Mount connections directory (for tensor-fusion-worker to write connection info)
+		result.VolumeMounts = append(result.VolumeMounts, VolumeMount{
+			HostPath:      connectionsDir,
+			ContainerPath: "/var/run/tensor-fusion/connections",
+			ReadOnly:      false,
+		})
+
 		// Mount libs directory (contains only .so files for LD_LIBRARY_PATH/LD_PRELOAD)
 		result.VolumeMounts = append(result.VolumeMounts, VolumeMount{
 			HostPath:      libsPath,