This repository has been archived on 2026-08-16. You can view files and clone it, but cannot push or open issues or pull requests.
ollie-9p/session/persist.go

361 lines
8.4 KiB
Go

package session
import (
"context"
"fmt"
"os"
"path/filepath"
"strings"
"sync"
"time"
"ollie/agent"
"ollie/backend"
coresession "ollie/session"
"ollie/toolsrv"
"ollie/tools"
"olliesrv/fs"
"olliesrv/prompts"
)
// Kill terminates a session and removes it from the tree.
func Kill(rs *rootState, id string) {
rs.mu.Lock()
sess := rs.sessions[id]
delete(rs.sessions, id)
rs.mu.Unlock()
if sess != nil {
sess.Cancel()
sess.Core.Close()
removePersistedAgent(rs, id)
rs.cfg.Log.Info("killed session %s", id)
if rs.cfg.OnSessionKilled != nil {
rs.cfg.OnSessionKilled(id)
}
}
}
// KillFromRoot is the public API for callers with *fs.Tree.
func KillFromRoot(root *fs.Tree, id string) {
Kill(state(root), id)
}
// Rename renames a session.
func Rename(rs *rootState, old, new string) error {
rs.mu.Lock()
sess, ok := rs.sessions[old]
if !ok {
rs.mu.Unlock()
return fmt.Errorf("session not found: %s", old)
}
if _, exists := rs.sessions[new]; exists {
rs.mu.Unlock()
return fmt.Errorf("session already exists: %s", new)
}
if sess.Core.Agent().IsRunning() {
rs.mu.Unlock()
return fmt.Errorf("cannot rename while agent is running")
}
if err := sess.Core.SetSessionID(new); err != nil {
rs.mu.Unlock()
return err
}
sess.id = new
rs.sessions[new] = sess
delete(rs.sessions, old)
rs.mu.Unlock()
sess.AppendLog([]byte(fmt.Sprintf(":: session renamed: %s -> %s\n", old, new)))
rs.cfg.Log.Info("renamed session %s -> %s", old, new)
if rs.cfg.OnSessionRenamed != nil {
rs.cfg.OnSessionRenamed(old, new)
}
return nil
}
// RenameFromRoot is the public API for callers with *fs.Tree.
func RenameFromRoot(root *fs.Tree, old, new string) error {
return Rename(state(root), old, new)
}
// Shutdown interrupts all sessions, waits for idle, persists, and closes.
func Shutdown(root *fs.Tree) {
rs := state(root)
InterruptAll(root)
waitIdle(rs, 100*time.Millisecond, 5*time.Second)
saveAllSessions(rs)
rs.mu.Lock()
ids := make([]string, 0, len(rs.sessions))
for id := range rs.sessions {
ids = append(ids, id)
}
rs.mu.Unlock()
for _, id := range ids {
rs.mu.Lock()
sess := rs.sessions[id]
delete(rs.sessions, id)
rs.mu.Unlock()
if sess != nil {
sess.Cancel()
sess.Core.Close()
rs.cfg.Log.Info("shutdown session %s", id)
}
}
}
func waitIdle(rs *rootState, poll, timeout time.Duration) {
deadline := time.Now().Add(timeout)
for time.Now().Before(deadline) {
allIdle := true
rs.mu.RLock()
for _, sess := range rs.sessions {
if sess.Core.Agent().State() != "idle" {
allIdle = false
break
}
}
rs.mu.RUnlock()
if allIdle {
return
}
time.Sleep(poll)
}
rs.cfg.Log.Warn("shutdown: timed out waiting for sessions to become idle, saving anyway")
}
// --- Persistence ---
func activeSessionsDir(rs *rootState) string {
return filepath.Join(rs.cfg.SessionsDir, "active")
}
func persistSession(rs *rootState, id string) {
rs.mu.RLock()
sess, ok := rs.sessions[id]
rs.mu.RUnlock()
if !ok {
return
}
dir := activeSessionsDir(rs)
os.MkdirAll(dir, 0700)
path := filepath.Join(dir, id+".json")
if err := sess.Core.SaveSession(path); err != nil {
rs.cfg.Log.Error("persist session %s: %v", id, err)
}
}
func removePersistedAgent(rs *rootState, id string) {
path := filepath.Join(activeSessionsDir(rs), id+".json")
os.Remove(path)
}
func saveAllSessions(rs *rootState) {
rs.mu.RLock()
ids := make([]string, 0, len(rs.sessions))
for id := range rs.sessions {
ids = append(ids, id)
}
rs.mu.RUnlock()
for _, id := range ids {
persistSession(rs, id)
}
}
func restoreAllSessions(rs *rootState) {
dir := activeSessionsDir(rs)
entries, err := os.ReadDir(dir)
if err != nil {
return
}
type loadedSession struct {
ps *agent.PersistedAgent
name string
}
var loaded []loadedSession
for _, e := range entries {
if !strings.HasSuffix(e.Name(), ".json") {
continue
}
path := filepath.Join(dir, e.Name())
ps, err := agent.LoadPersistedAgent(path)
if err != nil {
rs.cfg.Log.Error("restore session %s: %v", e.Name(), err)
continue
}
loaded = append(loaded, loadedSession{ps: ps, name: e.Name()})
}
if len(loaded) == 0 {
return
}
var wg sync.WaitGroup
for _, ls := range loaded {
wg.Add(1)
go func(ls loadedSession) {
defer wg.Done()
if err := restoreSession(rs, ls.ps); err != nil {
rs.cfg.Log.Error("restore session %s: %v", ls.ps.ID, err)
}
}(ls)
}
wg.Wait()
}
func restoreSession(rs *rootState, ps *agent.PersistedAgent) error {
cwd := ps.CWD
if cwd == "" {
cwd, _ = os.Getwd()
}
agentName := ps.Agent
if agentName == "" {
agentName = "default"
}
sessID := ps.ID
cfg := LoadAgentConfig(rs.cfg.AgentsDir, agentName, nil)
backendName := ps.Backend
if backendName == "" && cfg != nil && cfg.Backend != "" {
backendName = cfg.Backend
}
be, err := backend.NewWithName(backendName)
if err != nil {
return fmt.Errorf("backend: %w", err)
}
modelName := ps.Model
if modelName == "" && cfg != nil && cfg.Model != "" {
modelName = cfg.Model
}
if modelName == "" {
modelName = os.Getenv("OLLIE_MODEL")
}
if modelName != "" {
be.SetModel(modelName)
}
uname := rs.nextUname()
var newToolServer func() toolsrv.Runner
var promptEnv []string
remoteTarget := ps.Remote
if remoteTarget != "" {
rsrv, dialErr := toolsrv.RemoteDial(rs.cfg.Ctx, toolsrv.RemoteConfig{
SSHTarget: remoteTarget,
CWD: cwd,
})
if dialErr != nil {
return fmt.Errorf("remote dial: %w", dialErr)
}
newToolServer = func() toolsrv.Runner { return rsrv }
promptEnv = []string{
"PRIME_CWD=" + cwd,
"PRIME_PLATFORM=" + rsrv.Info.Platform,
"PRIME_IS_GIT_REPO=" + fmt.Sprintf("%v", rsrv.Info.IsGitRepo),
}
} else {
var execOpts []toolsrv.Option
if rs.cfg.Strict {
execOpts = append(execOpts, toolsrv.WithStrict())
}
if rs.cfg.Yolo {
execOpts = append(execOpts, toolsrv.WithYolo())
}
if rs.cfg.ToolRegistry != nil {
execOpts = append(execOpts, toolsrv.WithToolRegistry(rs.cfg.ToolRegistry, sessID))
}
if rs.cfg.SkillsRegistry != nil {
execOpts = append(execOpts, toolsrv.WithSkillsRegistry(rs.cfg.SkillsRegistry))
execOpts = append(execOpts, toolsrv.WithBuiltins(tools.Builtins()))
}
newToolServer = func() toolsrv.Runner {
srv, _ := toolsrv.Dial(rs.cfg.Ctx, toolsrv.LocalAddr(cwd), execOpts...)
return srv
}
promptEnv = agent.PromptEnv(cwd)
}
env := []string{"OLLIE_SESSION_ID=" + sessID, "OLLIE_UNAME=" + uname}
env = append(env, promptEnv...)
var spOverride string
if cfg != nil {
spOverride = cfg.SystemPrompt
}
sysPrompt := prompts.ResolveSystemPrompt(spOverride)
envMap := make(map[string]string)
for _, e := range env {
if k, v, ok := strings.Cut(e, "="); ok {
envMap[k] = v
}
}
opModel := prompts.OperationalModel(rs.cfg.Enable9P, rs.cfg.EnableDBus, envMap)
platform := "linux"
isGitRepo := false
for _, e := range promptEnv {
if k, v, ok := strings.Cut(e, "="); ok {
switch k {
case "PRIME_PLATFORM":
platform = v
case "PRIME_IS_GIT_REPO":
isGitRepo = v == "true"
}
}
}
envBlock := prompts.Environment(cwd, platform, isGitRepo, "")
toolSrv := newToolServer()
rt := agent.BuildRuntime(cfg, toolSrv, cwd, env, sysPrompt, opModel, envBlock)
restoredSession := agent.RestoreHistory(ps)
var sessPtr *Session
core := coresession.New(coresession.Config{
Backend: be,
AgentName: agentName,
AgentsDir: rs.cfg.AgentsDir,
SessionsDir: rs.cfg.SessionsDir,
SessionID: sessID,
AgentID: uname,
CWD: cwd,
Remote: remoteTarget,
History: restoredSession,
Runtime: rt,
NewToolServer: newToolServer,
PromptEnvExtra: promptEnv,
BaseLayers: []string{sysPrompt, opModel, envBlock},
Log: rs.cfg.Sink.NewLogger("core"),
ReadPlanStep: func() string {
if sessPtr == nil {
return ""
}
sessPtr.mu.RLock()
data := make([]byte, len(sessPtr.plan))
copy(data, sessPtr.plan)
sessPtr.mu.RUnlock()
return coresession.NextUncheckedStep(data)
},
})
sessionCtx, sessionCancel := context.WithCancel(rs.cfg.Ctx)
sess := NewSession(sessID, core, sessionCtx, sessionCancel)
sessPtr = sess
sess.uname = uname
sess.remote = remoteTarget
replayMessagesToLog(sess, ps.Messages)
rs.mu.Lock()
rs.sessions[sessID] = sess
rs.mu.Unlock()
rs.cfg.Log.Info("restored session %s (backend=%s model=%s agent=%s)", sessID, backendName, modelName, agentName)
if rs.cfg.OnSessionCreated != nil {
rs.cfg.OnSessionCreated(sessID, sess)
}
return nil
}