Add MaxRunningInstances to InstancesConfig and implement IsRunning method

pkg/config/config.go

@@ -55,6 +55,9 @@ type InstancesConfig struct {
 	// Maximum number of instances that can be created
 	MaxInstances int `yaml:"max_instances"`
 
+	// Maximum number of instances that can be running at the same time
+	MaxRunningInstances int `yaml:"max_running_instances,omitempty"`
+
 	// Path to llama-server executable
 	LlamaExecutable string `yaml:"llama_executable"`
 
@@ -113,6 +116,7 @@ func LoadConfig(configPath string) (AppConfig, error) {
 			LogsDir:              filepath.Join(getDefaultDataDirectory(), "logs"),
 			AutoCreateDirs:       true,
 			MaxInstances:         -1, // -1 means unlimited
+			MaxRunningInstances:  -1, // -1 means unlimited
 			LlamaExecutable:      "llama-server",
 			DefaultAutoRestart:   true,
 			DefaultMaxRestarts:   3,
@@ -211,6 +215,11 @@ func loadEnvVars(cfg *AppConfig) {
 			cfg.Instances.MaxInstances = m
 		}
 	}
+	if maxRunning := os.Getenv("LLAMACTL_MAX_RUNNING_INSTANCES"); maxRunning != "" {
+		if m, err := strconv.Atoi(maxRunning); err == nil {
+			cfg.Instances.MaxRunningInstances = m
+		}
+	}
 	if llamaExec := os.Getenv("LLAMACTL_LLAMA_EXECUTABLE"); llamaExec != "" {
 		cfg.Instances.LlamaExecutable = llamaExec
 	}

pkg/instance/lifecycle.go

@@ -11,6 +11,12 @@ import (
 	"time"
 )
 
+func (i *Process) IsRunning() bool {
+	i.mu.RLock()
+	defer i.mu.RUnlock()
+	return i.Running
+}
+
 // Start starts the llama server instance and returns an error if it fails.
 func (i *Process) Start() error {
 	i.mu.Lock()

pkg/manager/manager.go

@@ -30,6 +30,7 @@ type InstanceManager interface {
 type instanceManager struct {
 	mu               sync.RWMutex
 	instances        map[string]*instance.Process
+	runningInstances map[*instance.Process]struct{}
 	ports            map[int]bool
 	instancesConfig  config.InstancesConfig
 
@@ -47,6 +48,7 @@ func NewInstanceManager(instancesConfig config.InstancesConfig) InstanceManager
 	}
 	im := &instanceManager{
 		instances:        make(map[string]*instance.Process),
+		runningInstances: make(map[*instance.Process]struct{}),
 		ports:            make(map[int]bool),
 		instancesConfig:  instancesConfig,