From f85612a4da858806d2f345e6db275ceca487e6a3 Mon Sep 17 00:00:00 2001
From: Levi Neely <lneely@pascom.net>
Date: Thu, 1 Oct 2026 13:34:33 +0200
Subject: [PATCH] session: pause all sessions on shutdown, fix agent config
 loading

Shutdown changes:
- Shutdown() now calls PauseAll() before closing sessions
- PauseAll() interrupts agents then pauses each session
- Sessions persist their state, allowing resume on next startup

Agent loading fix:
- LoadConfig returns (*AgentConfig, error) instead of *AgentConfig
- Parse errors are no longer silently discarded
- Callers handle nil config gracefully (no nil pointer dereferences)
---
 cmd/olliesrv/internal/agent/config_paths.go | 12 ++++---
 cmd/olliesrv/internal/session/registry.go   | 35 ++++++++++++++++++---
 cmd/olliesrv/internal/session/session.go    | 14 +++++++--
 3 files changed, 49 insertions(+), 12 deletions(-)

diff --git a/cmd/olliesrv/internal/agent/config_paths.go b/cmd/olliesrv/internal/agent/config_paths.go
index b421b2a..0ddd32f 100644
--- a/cmd/olliesrv/internal/agent/config_paths.go
+++ b/cmd/olliesrv/internal/agent/config_paths.go
@@ -17,16 +17,20 @@ func AgentConfigPath(agentsDir, name string) string {
 }
 
 // LoadConfig loads an agent configuration file by profile name.
-func LoadConfig(agentsDir, name string, open func(string) (*os.File, error)) *AgentConfig {
+// Returns (nil, nil) if the file does not exist.
+// Returns (nil, err) for parse errors.
+func LoadConfig(agentsDir, name string, open func(string) (*os.File, error)) (*AgentConfig, error) {
 	if open == nil {
 		open = os.Open
 	}
 	path := AgentConfigPath(agentsDir, name)
 	f, err := open(path)
 	if err != nil {
-		return nil
+		if os.IsNotExist(err) {
+			return nil, nil
+		}
+		return nil, err
 	}
 	defer f.Close()
-	cfg, _ := Load(f)
-	return cfg
+	return Load(f)
 }
diff --git a/cmd/olliesrv/internal/session/registry.go b/cmd/olliesrv/internal/session/registry.go
index b9c51fd..3359611 100644
--- a/cmd/olliesrv/internal/session/registry.go
+++ b/cmd/olliesrv/internal/session/registry.go
@@ -170,9 +170,10 @@ func Rename(oldName, newName string) error {
 	return nil
 }
 
-// Shutdown interrupts all sessions, waits for idle, and closes them.
+// Shutdown pauses all sessions (to allow later resume), waits for idle, then closes them.
 func Shutdown() {
-	InterruptAll()
+	// First, pause all sessions to persist their state.
+	PauseAll()
 	waitIdle(100*time.Millisecond, 5*time.Second)
 
 	mu.Lock()
@@ -188,6 +189,8 @@ func Shutdown() {
 		delete(sessions, name)
 		mu.Unlock()
 		if sess != nil {
+			// Session is already paused, so Cancel/Keeper/Proc should be nil.
+			// But close anything still open defensively.
 			if sess.Cancel != nil {
 				sess.Cancel()
 			}
@@ -202,6 +205,25 @@ func Shutdown() {
 	}
 }
 
+// PauseAll pauses every active session. Agents are interrupted first.
+func PauseAll() {
+	InterruptAll()
+	mu.RLock()
+	var toProcess []*Session
+	for _, sess := range sessions {
+		if !sess.IsPaused() {
+			toProcess = append(toProcess, sess)
+		}
+	}
+	mu.RUnlock()
+
+	for _, sess := range toProcess {
+		if err := sess.Pause(); err != nil {
+			pkgLog.Warn("pause session %s: %v", sess.ID, err)
+		}
+	}
+}
+
 // InterruptAll interrupts every active session.
 func InterruptAll() {
 	mu.RLock()
@@ -380,7 +402,10 @@ func buildAgent(sess *Session, p AgentParams) (*agent.Agent, error) {
 		return nil, fmt.Errorf("session cwd is required")
 	}
 
-	cfg := agent.LoadConfig(pkgAgentsDir, p.Profile, nil)
+	cfg, err := agent.LoadConfig(pkgAgentsDir, p.Profile, nil)
+	if err != nil {
+		return nil, fmt.Errorf("load agent config %q: %w", p.Profile, err)
+	}
 	sessID := sess.ID
 
 	// Set up tool server — reuse existing session infrastructure
@@ -431,7 +456,7 @@ func buildAgent(sess *Session, p AgentParams) (*agent.Agent, error) {
 
 	// Create backend
 	beName := p.Backend
-	if beName == "" {
+	if beName == "" && cfg != nil {
 		beName = cfg.Backend
 	}
 	be, err := backend.NewWithName(beName)
@@ -442,7 +467,7 @@ func buildAgent(sess *Session, p AgentParams) (*agent.Agent, error) {
 
 	if p.Model != "" {
 		rt.Backend.SetModel(p.Model)
-	} else if cfg.Model != "" {
+	} else if cfg != nil && cfg.Model != "" {
 		rt.Backend.SetModel(cfg.Model)
 	}
 
diff --git a/cmd/olliesrv/internal/session/session.go b/cmd/olliesrv/internal/session/session.go
index 5e1721e..9bc59ee 100644
--- a/cmd/olliesrv/internal/session/session.go
+++ b/cmd/olliesrv/internal/session/session.go
@@ -496,15 +496,23 @@ func (s *Session) Resume() error {
 
 // rebuildAgentRuntime rebuilds the full runtime for an agent (used on resume).
 func rebuildAgentRuntime(ag *agent.Agent, sessID string) error {
-	cfg := agent.LoadConfig(pkgAgentsDir, ag.Profile(), os.Open)
+	cfg, err := agent.LoadConfig(pkgAgentsDir, ag.Profile(), os.Open)
+	if err != nil {
+		return fmt.Errorf("load agent config %q: %w", ag.Profile(), err)
+	}
 	layers := BuildPromptLayers(cfg, ag.Cwd(), sessID, ag.ID(), "linux", false, "")
 	env := []string{"OLLIE_SESSION_ID=" + sessID, "OLLIE_UNAME=" + ag.ID()}
 	rt := agent.BuildRuntime(cfg, ag.ToolServer(), ag.Cwd(), env, layers.SystemPrompt, layers.EnvBlock)
-	be, err := backend.NewWithName(cfg.Backend)
+
+	var beName string
+	if cfg != nil {
+		beName = cfg.Backend
+	}
+	be, err := backend.NewWithName(beName)
 	if err != nil {
 		return fmt.Errorf("backend: %w", err)
 	}
-	if cfg.Model != "" {
+	if cfg != nil && cfg.Model != "" {
 		be.SetModel(cfg.Model)
 	}
 	rt.Backend = be