Resume queued appends at their saved log ID

[onechat] / daemon.go

diff --git a/daemon.go b/daemon.go
index 2a23dc28a4405380137213b21f9c25566c575844..3a54b1ae94374c7d53c3938f1eeff0ae8fcb9768 100644 (file)
--- a/daemon.go
+++ b/daemon.go
@@ -7,7 +7,6 @@ import (
        "errors"
        "fmt"
        "io"
        "errors"
        "fmt"
        "io"

-       "log"

        "net"
        "os"
        "path/filepath"
        "net"
        "os"
        "path/filepath"


@@ -25,10 +24,13 @@ const (
        daemonClientEvents = 1024
 )
 
        daemonClientEvents = 1024
 )
 

-// daemon runs "onechat daemon": it owns the root agent and its files and serves
-// the clients on the Unix socket in the data directory until ctx ends. When
-// telegram.json exists, it also runs the Telegram bot. It holds the lock of the
-// token file, so it is the only process that refreshes the tokens.
+// daemon runs "onechat daemon": it owns the root, its children, and their
+// files, and serves the clients on the Unix socket in the data directory until
+// ctx ends. When telegram.json exists, it also runs the Telegram bot. It also
+// keeps a linked WhatsApp device connected. It holds the lock of the token
+// file, so it is the only process that refreshes the tokens and opens the
+// WhatsApp session. Its log lines go to out, through the loggers of
+// daemonLogger.

 func daemon(ctx context.Context, out io.Writer) (err error) {
        dir, err := dataDir()
        if err != nil {
 func daemon(ctx context.Context, out io.Writer) (err error) {
        dir, err := dataDir()
        if err != nil {


@@ -62,52 +64,68 @@ func daemon(ctx context.Context, out io.Writer) (err error) {
                err = errors.Join(err, tokens.Close())
        }()
 
                err = errors.Join(err, tokens.Close())
        }()
 

-       address, auth, err := modelEndpoint(config, dir)
+       whatsappSetDeviceProps()
+
+       logger := daemonLogger(out, "onechat: ")
+       whatsapp, err := openWhatsApp(ctx, dir, time.Now, daemonLogger(out, "whatsapp: "), whatsappNewClient)

        if err != nil {
                return err
        }
 
        if err != nil {
                return err
        }
 

-       agent, err := openAgent(filepath.Join(dir, daemonAgentDirName))
+       defer func() {
+               err = errors.Join(err, whatsapp.Close())
+       }()
+
+       address, auth, err := modelEndpoint(config, dir)

        if err != nil {
                return err
        }
 
        if err != nil {
                return err
        }
 

-       client := newModelClient(address, auth, agent.CacheKey(), filepath.Join(agent.dir, "usage.jsonl"))
-       memory, err := openMemory(ctx, agent.dir, time.Now, newCompactor(client, agent.Prompt, config.Compactor))
+       agents, err := openDaemonAgents(ctx, dir, address, auth, config.Compactor, time.Now, logger)

        if err != nil {
                return err
        }
 
        defer func() {
        if err != nil {
                return err
        }
 
        defer func() {

-               err = errors.Join(err, memory.Close())
+               err = errors.Join(err, agents.Close())

        }()
 
        }()
 

+       root := agents[0]
+       agent := root.agent
+       schedules := root.schedules
+

        listener, err := daemonListenSocket(socket)
        if err != nil {
                return err
        }
 
        listener, err := daemonListenSocket(socket)
        if err != nil {
                return err
        }
 

-       _, err = fmt.Fprintf(out, "onechat: %s on %s, %s at %s effort, in %s\n", agent.config.Name, listener.Addr(), agent.config.Model, agent.config.Effort, agent.config.Workdir)
-       if err != nil {
-               return errors.Join(err, listener.Close())
+       linked := "not linked"
+       if whatsapp.isLinked() {
+               linked = "linked"

        }
 
        }
 

+       logger.Info("ready", "version", version, "socket", listener.Addr(), "model", agent.config.Model, "effort", agent.config.Effort, "workdir", agent.config.Workdir, "whatsapp", linked, "schedules", schedules.Len(), "children", agents.ChildNames())
+

        var telegram *TelegramBot
        var telegram *TelegramBot

+

        if hasBot {
        if hasBot {

-               telegram = newTelegramBot(newTelegramAPI(telegramAPIURL, botSettings.Token), botSettings.Owner, filepath.Join(agent.dir, telegramOffsetFileName), log.New(out, "telegram: ", log.LstdFlags))
-               _, err = fmt.Fprintln(out, "onechat: the Telegram bot waits for messages of its owner")
-               if err != nil {
-                       return errors.Join(err, listener.Close())
-               }
+               telegram = newTelegramBot(newTelegramAPI(telegramAPIURL, botSettings.Token), botSettings.Owner, filepath.Join(agent.dir, telegramOffsetFileName), daemonLogger(out, "telegram: "))
+
+               logger.Info("Telegram bot started")

        }
 
        }
 

-       return daemonServe(ctx, listener, telegram, func(notify func(Event)) (*TurnLoop, error) {
-               return newTurnLoop(agent, memory, client, time.Now, notify)
-       })
+       return daemonServe(ctx, listener, telegram, whatsapp, schedules, func(notify func(Event)) (*TurnLoop, error) {
+               return root.Loop(whatsapp, notify)
+       }, agents[1:]...)

 }
 
 }
 

-// DaemonHub holds the connected clients and sends each event to all of them.
+// DaemonHub holds the connected clients and sends each event to all of them. It
+// runs the WhatsApp commands of the clients on whatsapp. snapshot locks all
+// agents and subagents during a backup copy.

 type DaemonHub struct {
 type DaemonHub struct {

+       whatsapp *WhatsApp
+       snapshot func(context.Context, func() error) error
+

        mutex   sync.Mutex
        clients map[*DaemonClient]bool
        workers sync.WaitGroup
        mutex   sync.Mutex
        clients map[*DaemonClient]bool
        workers sync.WaitGroup


@@ -129,10 +147,12 @@ func (h *DaemonHub) Accept(ctx context.Context, listener net.Listener, l *TurnLo
                }
 
                h.mutex.Lock()
                }
 
                h.mutex.Lock()

+

                h.clients[client] = true
                h.mutex.Unlock()
 
                h.workers.Add(2)
                h.clients[client] = true
                h.mutex.Unlock()
 
                h.workers.Add(2)

+

                go h.write(client)
                go h.read(ctx, client, l)
 
                go h.write(client)
                go h.read(ctx, client, l)
 


@@ -169,11 +189,30 @@ func (h *DaemonHub) read(ctx context.Context, client *DaemonClient, l *TurnLoop)
 
        scanner := bufio.NewScanner(client.conn)
        scanner.Buffer(make([]byte, 64*1024), socketRequestBytes)
 
        scanner := bufio.NewScanner(client.conn)
        scanner.Buffer(make([]byte, 64*1024), socketRequestBytes)

+

        for ctx.Err() == nil && scanner.Scan() {
                var request SocketRequest
        for ctx.Err() == nil && scanner.Scan() {
                var request SocketRequest

+

                err := json.Unmarshal(scanner.Bytes(), &request)
                if err == nil {
                err := json.Unmarshal(scanner.Bytes(), &request)
                if err == nil {

-                       err = daemonRunRequest(l, request)
+                       if request.Type == "backup" {
+                               var size int64
+
+                               size, err = backupData(ctx, filepath.Dir(l.agent.dir), request.Text, h.whatsapp, h.snapshot)
+                               if err == nil {
+                                       h.Send(client, Event{
+                                               Kind: "backup",
+                                               Text: fmt.Sprintf("%s (%d bytes)", request.Text, size),
+                                       })
+                               }
+                       } else {
+                               err = daemonRunRequest(l, h.whatsapp, request, func(text string) {
+                                       h.Send(client, Event{
+                                               Kind: "notice",
+                                               Text: text,
+                                       })
+                               })
+                       }

                }
 
                if err != nil {
                }
 
                if err != nil {


@@ -221,6 +260,7 @@ func (h *DaemonHub) remove(client *DaemonClient) {
        delete(h.clients, client)
        client.once.Do(func() {
                close(client.events)
        delete(h.clients, client)
        client.once.Do(func() {
                close(client.events)

+

                _ = client.conn.Close()
        })
 }
                _ = client.conn.Close()
        })
 }


@@ -238,9 +278,11 @@ func (h *DaemonHub) Broadcast(event Event) {
 // CloseAll drops every client and waits for their readers and writers.
 func (h *DaemonHub) CloseAll() {
        h.mutex.Lock()
 // CloseAll drops every client and waits for their readers and writers.
 func (h *DaemonHub) CloseAll() {
        h.mutex.Lock()

+

        for client := range h.clients {
                h.remove(client)
        }
        for client := range h.clients {
                h.remove(client)
        }

+

        h.mutex.Unlock()
 
        h.workers.Wait()
        h.mutex.Unlock()
 
        h.workers.Wait()


@@ -282,14 +324,20 @@ func daemonListenSocket(path string) (net.Listener, error) {
 }
 
 // daemonServe runs the loop that start makes and serves the clients of listener
 }
 
 // daemonServe runs the loop that start makes and serves the clients of listener

-// until ctx ends or the loop fails. It closes listener. When bot is not nil, it
-// runs the bot too, and a failure of the bot stops the agent.
-func daemonServe(ctx context.Context, listener net.Listener, bot *TelegramBot, start func(notify func(Event)) (*TurnLoop, error)) error {
+// until ctx ends or the loop fails. It closes listener. It runs whatsapp, which
+// logs its batches and notes in the loop and tells all clients its notices, and
+// a failure of whatsapp stops the agent. It runs schedules, which log their
+// runs in the loop, and a failure of schedules stops the agent too. When bot is
+// not nil, it runs the bot too, and a failure of the bot stops the agent. Each
+// child runs its own loop and schedules with no client events. A failed child
+// stops the daemon, and backups lock all loops before they copy data.
+func daemonServe(ctx context.Context, listener net.Listener, bot *TelegramBot, whatsapp *WhatsApp, schedules *Schedules, start func(notify func(Event)) (*TurnLoop, error), children ...*DaemonAgent) error {

        ctx, cancel := context.WithCancel(ctx)
        defer cancel()
 
        hub := &DaemonHub{
        ctx, cancel := context.WithCancel(ctx)
        defer cancel()
 
        hub := &DaemonHub{

-               clients: make(map[*DaemonClient]bool),
+               whatsapp: whatsapp,
+               clients:  make(map[*DaemonClient]bool),

        }
 
        notify := hub.Broadcast
        }
 
        notify := hub.Broadcast


@@ -305,21 +353,79 @@ func daemonServe(ctx context.Context, listener net.Listener, bot *TelegramBot, s
                return errors.Join(err, listener.Close())
        }
 
                return errors.Join(err, listener.Close())
        }
 

+       loops := []*TurnLoop{l}
+       for _, child := range children {
+               loop, err := child.Loop(nil, func(Event) {
+               })
+               if err != nil {
+                       return errors.Join(fmt.Errorf("start child %q: %w", child.agent.config.Name, err), listener.Close())
+               }
+
+               loops = append(loops, loop)
+       }
+
+       hub.snapshot = func(ctx context.Context, snapshot func() error) error {
+               return daemonSnapshot(ctx, loops, snapshot)
+       }
+

        var workers sync.WaitGroup
        var workers sync.WaitGroup

-       workers.Add(1)
+
+       childErrors := make([]error, len(children))
+       for i, child := range children {
+               workers.Add(1)
+
+               go func() {
+                       defer workers.Done()
+
+                       childErrors[i] = daemonRunChild(ctx, loops[i+1], child.schedules)
+                       if childErrors[i] != nil {
+                               child.logger.Error("child stopped", "err", childErrors[i])
+                       }
+
+                       cancel()
+               }()
+       }
+
+       workers.Add(2)
+

        go func() {
                defer workers.Done()
 
                hub.Accept(ctx, listener, l)
        }()
 
        go func() {
                defer workers.Done()
 
                hub.Accept(ctx, listener, l)
        }()
 

+       var whatsappErr error
+
+       go func() {
+               defer workers.Done()
+
+               whatsappErr = whatsapp.Run(ctx, l, notify)
+
+               cancel()
+       }()
+
+       var schedulesErr error
+
+       workers.Add(1)
+
+       go func() {
+               defer workers.Done()
+
+               schedulesErr = schedules.Run(ctx, l)
+
+               cancel()
+       }()
+

        var botErr error
        var botErr error

+

        if bot != nil {
                workers.Add(1)
        if bot != nil {
                workers.Add(1)

+

                go func() {
                        defer workers.Done()
 
                go func() {
                        defer workers.Done()
 

-                       botErr = bot.Run(ctx, l)
+                       botErr = bot.Run(ctx, l, whatsapp)
+

                        cancel()
                }()
        }
                        cancel()
                }()
        }


@@ -330,15 +436,41 @@ func daemonServe(ctx context.Context, listener net.Listener, bot *TelegramBot, s
        }
 
        cancel()
        }
 
        cancel()

+

        err = errors.Join(err, listener.Close())
        err = errors.Join(err, listener.Close())

+

        hub.CloseAll()
        workers.Wait()
 
        hub.CloseAll()
        workers.Wait()
 

-       return errors.Join(err, botErr)
+       return errors.Join(err, whatsappErr, schedulesErr, botErr, errors.Join(childErrors...))

 }
 
 }
 

-// daemonRunRequest runs one request of a client.
-func daemonRunRequest(l *TurnLoop, request SocketRequest) error {
+// daemonRunChild runs a child's turn loop and its schedules until either stops.
+// Its events stay in its own log.
+func daemonRunChild(ctx context.Context, loop *TurnLoop, schedules *Schedules) error {
+       ctx, cancel := context.WithCancel(ctx)
+       defer cancel()
+
+       done := make(chan error, 1)
+       go func() {
+               done <- schedules.Run(ctx, loop)
+
+               cancel()
+       }()
+
+       err := loop.Run(ctx)
+       if errors.Is(err, context.Canceled) {
+               err = nil
+       }
+
+       cancel()
+
+       return errors.Join(err, <-done)
+}
+
+// daemonRunRequest runs one request of a client. The answers of a WhatsApp
+// command go to answer.
+func daemonRunRequest(l *TurnLoop, whatsapp *WhatsApp, request SocketRequest, answer func(text string)) error {

        switch request.Type {
        case "send":
                if request.Text == "" {
        switch request.Type {
        case "send":
                if request.Text == "" {


@@ -354,6 +486,17 @@ func daemonRunRequest(l *TurnLoop, request SocketRequest) error {
                }
 
                return nil
                }
 
                return nil

+       case "whatsapp":
+               whatsapp.Command(request.Text, answer)
+
+               return nil
+       case "stop":
+               found, err := l.StopSubagent(request.Text)
+               if err == nil && !found {
+                       err = fmt.Errorf("no subagent %s runs", request.Text)
+               }
+
+               return err

        default:
                return fmt.Errorf("unknown request %q", request.Type)
        }
        default:
                return fmt.Errorf("unknown request %q", request.Type)
        }