Explain daemon log levels and journal filtering

[onechat] / turn.go

diff --git a/turn.go b/turn.go
index 02a4a28513d3b6419d24b1e1b5929e1652a6a315..637d783b70dfb8c5eb80eb521e7e0510b1db13fc 100644 (file)
--- a/turn.go
+++ b/turn.go
@@ -4,15 +4,17 @@ import (
        "context"
        "errors"
        "fmt"
        "context"
        "errors"
        "fmt"

+       "log/slog"

        "path/filepath"
        "sync"
        "time"
 )
 
        "path/filepath"
        "sync"
        "time"
 )
 

-// newTurnLoop returns the loop of agent with its memory and its subagents. Its
-// calls go through client, and notify gets each event. The clock gives the date
-// of a turn.
-func newTurnLoop(agent *Agent, memory *Memory, subagents *Subagents, client ChainClient, clock func() time.Time, notify func(Event)) (*TurnLoop, error) {
+// newTurnLoop returns the loop of agent with its memory, its subagents, and the
+// WhatsApp link and the schedules of its tools. Its calls go through client,
+// and notify gets each event. The clock gives the date of a turn. logger gets
+// the failed turns and the ends of the subagents.
+func newTurnLoop(agent *Agent, memory *Memory, subagents *Subagents, whatsapp *WhatsApp, schedules *Schedules, client ChainClient, clock func() time.Time, logger *slog.Logger, notify func(Event)) (*TurnLoop, error) {

        layout, err := loadViewLayout(filepath.Join(agent.dir, viewLayoutFileName))
        if err != nil {
                return nil, err
        layout, err := loadViewLayout(filepath.Join(agent.dir, viewLayoutFileName))
        if err != nil {
                return nil, err


@@ -24,6 +26,7 @@ func newTurnLoop(agent *Agent, memory *Memory, subagents *Subagents, client Chai
                client: client,
                layout: layout,
                clock:  clock,
                client: client,
                layout: layout,
                clock:  clock,

+               logger: logger,

                notify: notify,
                wake:   make(chan struct{}, 1),
        }
                notify: notify,
                wake:   make(chan struct{}, 1),
        }


@@ -36,6 +39,9 @@ func newTurnLoop(agent *Agent, memory *Memory, subagents *Subagents, client Chai
                client:    client,
                state:     l.state,
                report:    l.Report,
                client:    client,
                state:     l.state,
                report:    l.Report,

+               note:      l.note,
+               fail:      l.fail,
+               logger:    logger,

                notify:    notify,
                running:   make(map[string]*SubagentRun),
        }
                notify:    notify,
                running:   make(map[string]*SubagentRun),
        }


@@ -45,6 +51,8 @@ func newTurnLoop(agent *Agent, memory *Memory, subagents *Subagents, client Chai
                chat:      memory,
                subagents: subagents,
                runs:      l.runs,
                chat:      memory,
                subagents: subagents,
                runs:      l.runs,

+               whatsapp:  whatsapp,
+               schedules: schedules,

                workdir:   agent.config.Workdir,
        }
 
                workdir:   agent.config.Workdir,
        }
 


@@ -52,15 +60,18 @@ func newTurnLoop(agent *Agent, memory *Memory, subagents *Subagents, client Chai
 }
 
 // Event tells the clients what the agent does as it happens. Kind is the kind
 }
 
 // Event tells the clients what the agent does as it happens. Kind is the kind

-// of a logged message (user, reply, tool, echo, work), or one of ready, busy,
-// idle, error, and subagent, which tells that the subagent Text starts a run. ID is the ID of a logged message. Source names the client that
-// sent a user message, such as terminal or telegram, so that each client can
-// show the messages of the others.
+// of a logged message (user, reply, tool, echo, note, work), or one of ready,
+// busy, idle, error, and subagent, which tells that the subagent Text starts a
+// run. ID is the ID of a logged message. Source names the client that sent a
+// user message, such as terminal or telegram, so that each client can show the
+// messages of the others, or the subagent of a report. Status tells how the run
+// of a report ended: finished, stopped, or failed.

 type Event struct {
        Kind   string `json:"kind"`
        ID     int    `json:"id,omitempty"`
        Text   string `json:"text,omitempty"`
        Source string `json:"source,omitempty"`
 type Event struct {
        Kind   string `json:"kind"`
        ID     int    `json:"id,omitempty"`
        Text   string `json:"text,omitempty"`
        Source string `json:"source,omitempty"`

+       Status string `json:"status,omitempty"`

 }
 
 // TurnLoop runs the turns of one agent. A user message goes to the log when it
 }
 
 // TurnLoop runs the turns of one agent. A user message goes to the log when it


@@ -81,6 +92,7 @@ type TurnLoop struct {
        runs   *SubagentRuns
        layout *ViewLayout
        clock  func() time.Time
        runs   *SubagentRuns
        layout *ViewLayout
        clock  func() time.Time

+       logger *slog.Logger

        notify func(Event)
        wake   chan struct{}
 
        notify func(Event)
        wake   chan struct{}
 


@@ -100,6 +112,7 @@ func (l *TurnLoop) Submit(source, text string) (Message, error) {
        }
 
        l.mutex.Lock()
        }
 
        l.mutex.Lock()

+

        l.pending = append(l.pending, message)
        l.mutex.Unlock()
 
        l.pending = append(l.pending, message)
        l.mutex.Unlock()
 


@@ -111,20 +124,52 @@ func (l *TurnLoop) Submit(source, text string) (Message, error) {
        return message, nil
 }
 
        return message, nil
 }
 

-// Report logs the report text of a subagent as a work message and tells the
-// clients. Unless the subagent was stopped, the report joins the waiting
-// messages and wakes the loop, as a user message of Submit does. A failed write
-// stops the loop, because the log refuses all writes after it.
-func (l *TurnLoop) Report(text string, stopped bool) {
-       message, err := l.record("work", "", text)
-
-       l.mutex.Lock()
+// Report logs the report text of the subagent name as a work message and tells
+// the clients, with the status of the run. Unless the subagent was stopped, the
+// report joins the waiting messages and wakes the loop, as a user message of
+// Submit does. A failed write stops the loop, because the log refuses all
+// writes after it.
+func (l *TurnLoop) Report(name, text, status string) {
+       message, err := l.Work(name, text, status)

        switch {
        case err != nil:
        switch {
        case err != nil:

-               l.failure = fmt.Errorf("%w: %w", errChainLogWrite, err)
-       case !stopped:
-               l.pending = append(l.pending, message)
+               l.fail(fmt.Errorf("%w: %w", errChainLogWrite, err))
+       case status != subagentStopped:
+               l.Wake(message)
+       }
+}
+
+// Work logs text as a work message of source, such as the name of a subagent,
+// and tells the clients, with status. The message starts no turn until it goes
+// to Wake.
+func (l *TurnLoop) Work(source, text, status string) (Message, error) {
+       message, err := l.memory.Append("work", text)
+       if err != nil {
+               return Message{}, err

        }
        }

+
+       l.notify(Event{
+               Kind:   "work",
+               ID:     message.ID,
+               Text:   text,
+               Source: source,
+               Status: status,
+       })
+
+       return message, nil
+}
+
+// Wake adds the logged messages to the waiting messages and wakes the loop
+// once, as a user message of Submit does. They join the first request of a new
+// turn together, or the next request of a running turn after the tool results.
+func (l *TurnLoop) Wake(messages ...Message) {
+       if len(messages) == 0 {
+               return
+       }
+
+       l.mutex.Lock()
+
+       l.pending = append(l.pending, messages...)

        l.mutex.Unlock()
 
        select {
        l.mutex.Unlock()
 
        select {


@@ -133,6 +178,12 @@ func (l *TurnLoop) Report(text string, stopped bool) {
        }
 }
 
        }
 }
 

+// StopSubagent ends the run of the subagent name and tells if it ran. Its
+// report is logged, but it starts no turn.
+func (l *TurnLoop) StopSubagent(name string) (bool, error) {
+       return l.runs.Stop(name)
+}
+

 // Interrupt ends the running turn. A running tool is killed, and the turn sends
 // no further request.
 func (l *TurnLoop) Interrupt() bool {
 // Interrupt ends the running turn. A running tool is killed, and the turn sends
 // no further request.
 func (l *TurnLoop) Interrupt() bool {


@@ -168,11 +219,17 @@ func (l *TurnLoop) Flush(ctx context.Context) error {
 }
 
 // Run runs a turn each time user messages or reports wait, until ctx ends. It
 }
 
 // Run runs a turn each time user messages or reports wait, until ctx ends. It

-// returns the error of a failed log write, which stops the agent. At its end
-// it ends the runs of the subagents and waits for them.
+// returns the error of a failed log write, which stops the agent. At its start
+// it resumes the subagents that a stop of the daemon cut, and at its end it
+// ends the runs of the subagents and waits for them.

 func (l *TurnLoop) Run(ctx context.Context) error {
        defer l.runs.Close()
 
 func (l *TurnLoop) Run(ctx context.Context) error {
        defer l.runs.Close()
 

+       err := l.runs.Resume(ctx)
+       if err != nil {
+               return fmt.Errorf("resume subagents: %w", err)
+       }
+

        for {
                for l.waiting() {
                        err := l.turn(ctx)
        for {
                for l.waiting() {
                        err := l.turn(ctx)


@@ -194,6 +251,13 @@ func (l *TurnLoop) Run(ctx context.Context) error {
        }
 }
 
        }
 }
 

+// note logs a note and tells the clients. A note starts no turn.
+func (l *TurnLoop) note(text string) error {
+       _, err := l.record("note", "", text)
+
+       return err
+}
+

 // record logs a message and tells the clients, with the client source of a user
 // message.
 func (l *TurnLoop) record(kind, source, text string) (Message, error) {
 // record logs a message and tells the clients, with the client source of a user
 // message.
 func (l *TurnLoop) record(kind, source, text string) (Message, error) {


@@ -220,6 +284,19 @@ func (l *TurnLoop) waiting() bool {
        return len(l.pending) > 0
 }
 
        return len(l.pending) > 0
 }
 

+// fail stops the loop with err.
+func (l *TurnLoop) fail(err error) {
+       l.mutex.Lock()
+
+       l.failure = errors.Join(l.failure, err)
+       l.mutex.Unlock()
+
+       select {
+       case l.wake <- struct{}{}:
+       default:
+       }
+}
+

 // failed returns the error of a failed write of a report.
 func (l *TurnLoop) failed() error {
        l.mutex.Lock()
 // failed returns the error of a failed write of a report.
 func (l *TurnLoop) failed() error {
        l.mutex.Lock()


@@ -229,19 +306,23 @@ func (l *TurnLoop) failed() error {
 }
 
 // turn runs one turn. A failed or interrupted model call ends the turn and
 }
 
 // turn runs one turn. A failed or interrupted model call ends the turn and

-// tells the clients. The messages of the turn stay in the log and the view, and
-// the next turn starts only with a new message, so a lasting failure such as a
-// needed login does not repeat. Only a failed log write is an error of turn.
+// tells the clients. A failure also goes to the logger, so that it is not lost
+// when no client is connected, but an interrupt does not. The messages of the
+// turn stay in the log and the view, and the next turn starts only with a new
+// message, so a lasting failure such as a needed login does not repeat. Only a
+// failed log write is an error of turn.

 func (l *TurnLoop) turn(ctx context.Context) error {
        ctx, cancel := context.WithCancel(ctx)
        defer cancel()
 
        l.mutex.Lock()
 func (l *TurnLoop) turn(ctx context.Context) error {
        ctx, cancel := context.WithCancel(ctx)
        defer cancel()
 
        l.mutex.Lock()

+

        l.cancel = cancel
        l.mutex.Unlock()
 
        defer func() {
                l.mutex.Lock()
        l.cancel = cancel
        l.mutex.Unlock()
 
        defer func() {
                l.mutex.Lock()

+

                l.cancel = nil
                l.mutex.Unlock()
        }()
                l.cancel = nil
                l.mutex.Unlock()
        }()


@@ -252,6 +333,10 @@ func (l *TurnLoop) turn(ctx context.Context) error {
 
        err := l.chain(ctx)
        if err != nil && !errors.Is(err, errChainLogWrite) {
 
        err := l.chain(ctx)
        if err != nil && !errors.Is(err, errChainLogWrite) {

+               if ctx.Err() == nil {
+                       l.logger.Error("turn failed", "err", err)
+               }
+

                l.notify(Event{
                        Kind: "error",
                        Text: turnError(ctx, err),
                l.notify(Event{
                        Kind: "error",
                        Text: turnError(ctx, err),