Keep retries after a rejected pair request

[onechat] / subagent_run.go

diff --git a/subagent_run.go b/subagent_run.go
index 16cf2bd03340a7271c985859f739777811810b46..67cf79d7ac14949738e2e167ee8269d45b9409a2 100644 (file)
--- a/subagent_run.go
+++ b/subagent_run.go
@@ -5,12 +5,14 @@ import (
        "encoding/json"
        "errors"
        "fmt"
        "encoding/json"
        "errors"
        "fmt"

+       "log"

        "maps"
        "os"
        "path/filepath"
        "slices"
        "strings"
        "sync"
        "maps"
        "os"
        "path/filepath"
        "slices"
        "strings"
        "sync"

+       "time"

 )
 
 const (
 )
 
 const (


@@ -43,9 +45,10 @@ var errSubagentClosed = errors.New("the daemon stops, so no subagent starts")
 // of the subagent. A message of Tell that comes during a chain joins its next
 // request after the tool results. One that comes after the final answer starts
 // one more chain, which reads the log again. The report of the run goes to
 // of the subagent. A message of Tell that comes during a chain joins its next
 // request after the tool results. One that comes after the final answer starts
 // one more chain, which reads the log again. The report of the run goes to

-// report: the final answer of each chain, or the error of a failed call. Stop
-// ends one run, and its report says that it stopped. At the stop of the daemon
-// the runs end with no report.
+// report: the final answer of each chain, or the error of a failed call. The
+// logger gets one line at the end of a run: done, failed with the cause, or
+// stopped. Stop ends one run, and its report says that it stopped. At the stop
+// of the daemon the runs end with no report.

 //
 // work.json in the directory of the agent holds the names of the running
 // subagents. Each start and each end of a run writes it whole, but Close does
 //
 // work.json in the directory of the agent holds the names of the running
 // subagents. Each start and each end of a run writes it whole, but Close does


@@ -66,6 +69,7 @@ type SubagentRuns struct {
        report    func(name, text, status string)
        note      func(text string) error
        fail      func(err error)
        report    func(name, text, status string)
        note      func(text string) error
        fail      func(err error)

+       logger    *log.Logger

        notify    func(Event)
 
        mutex   sync.Mutex
        notify    func(Event)
 
        mutex   sync.Mutex


@@ -237,6 +241,7 @@ func (r *SubagentRuns) Resume(ctx context.Context) error {
        for _, name := range names {
                _, err = r.Tell(ctx, name, text)
                if err != nil {
        for _, name := range names {
                _, err = r.Tell(ctx, name, text)
                if err != nil {

+                       r.logger.Printf("subagent %s lost at restart: %v", name, err)

                        r.report(name, subagentReport(name, []string{"failed: " + err.Error()}), subagentFailed)
                }
        }
                        r.report(name, subagentReport(name, []string{"failed: " + err.Error()}), subagentFailed)
                }
        }


@@ -313,8 +318,8 @@ func (r *SubagentRuns) effort(subagent *Subagent) (string, error) {
        return "", fmt.Errorf("the log of %s does not start with a subagent note", subagent.name)
 }
 
        return "", fmt.Errorf("the log of %s does not start with a subagent note", subagent.name)
 }
 

-// start runs run in the background. Its context keeps the values of ctx, but
-// only Stop and Close end it.
+// start runs run in the background and logs its start. Its context keeps the
+// values of ctx, but only Stop and Close end it.

 func (r *SubagentRuns) start(ctx context.Context, run *SubagentRun) error {
        ctx, run.cancel = context.WithCancel(context.WithoutCancel(ctx))
 
 func (r *SubagentRuns) start(ctx context.Context, run *SubagentRun) error {
        ctx, run.cancel = context.WithCancel(context.WithoutCancel(ctx))
 


@@ -344,6 +349,8 @@ func (r *SubagentRuns) start(ctx context.Context, run *SubagentRun) error {
        r.workers.Add(1)
        r.mutex.Unlock()
 
        r.workers.Add(1)
        r.mutex.Unlock()
 

+       run.started = r.subagents.clock()
+       r.logger.Printf("subagent %s started", run.subagent.name)

        r.notify(Event{
                Kind: "subagent",
                Text: run.subagent.name,
        r.notify(Event{
                Kind: "subagent",
                Text: run.subagent.name,


@@ -398,14 +405,15 @@ func (r *SubagentRuns) workPath() string {
        return filepath.Join(r.agent.dir, subagentWorkFileName)
 }
 
        return filepath.Join(r.agent.dir, subagentWorkFileName)
 }
 

-// run runs the chains of run and reports its end, unless Close ended it. A
-// chain follows while messages wait after the final answer of the chain before
+// run runs the chains of run and reports and logs its end with the length of
+// the run, unless Close ended it. A chain follows while messages wait after the final answer of the chain before

 // it, and no chain failed. A chain that Stop ends gives the reply "stopped".
 func (r *SubagentRuns) run(ctx context.Context, run *SubagentRun) {
        defer r.workers.Done()
        defer run.cancel()
 
        var replies []string
 // it, and no chain failed. A chain that Stop ends gives the reply "stopped".
 func (r *SubagentRuns) run(ctx context.Context, run *SubagentRun) {
        defer r.workers.Done()
        defer run.cancel()
 
        var replies []string

+       var failure error

        stopped := false
        status := subagentFinished
        for more := true; more; {
        stopped := false
        status := subagentFinished
        for more := true; more; {


@@ -419,6 +427,7 @@ func (r *SubagentRuns) run(ctx context.Context, run *SubagentRun) {
                case err != nil:
                        reply = "failed: " + err.Error()
                        status = subagentFailed
                case err != nil:
                        reply = "failed: " + err.Error()
                        status = subagentFailed

+                       failure = err

                }
 
                more = err == nil && ctx.Err() == nil && len(run.pending) > 0
                }
 
                more = err == nil && ctx.Err() == nil && len(run.pending) > 0


@@ -447,6 +456,16 @@ func (r *SubagentRuns) run(ctx context.Context, run *SubagentRun) {
                status = subagentStopped
        }
 
                status = subagentStopped
        }
 

+       length := durationText(r.subagents.clock().Sub(run.started))
+       switch status {
+       case subagentStopped:
+               r.logger.Printf("subagent %s stopped after %s", run.subagent.name, length)
+       case subagentFailed:
+               r.logger.Printf("subagent %s failed after %s: %v", run.subagent.name, length, failure)
+       default:
+               r.logger.Printf("subagent %s done after %s", run.subagent.name, length)
+       }
+

        r.report(run.subagent.name, subagentReport(run.subagent.name, replies), status)
 }
 
        r.report(run.subagent.name, subagentReport(run.subagent.name, replies), status)
 }
 


@@ -505,7 +524,7 @@ func (r *SubagentRuns) take(run *SubagentRun) []Message {
 
 // SubagentRun is one run of a subagent: the view that its requests get, its
 // effort, the end of its context, the messages of Tell that wait for a request,
 
 // SubagentRun is one run of a subagent: the view that its requests get, its
 // effort, the end of its context, the messages of Tell that wait for a request,

-// and whether Stop ended it. The mutex of SubagentRuns protects the messages
+// whether Stop ended it, and the time it started. The mutex of SubagentRuns protects the messages

 // and the stop.
 type SubagentRun struct {
        subagent *Subagent
 // and the stop.
 type SubagentRun struct {
        subagent *Subagent


@@ -514,6 +533,7 @@ type SubagentRun struct {
        cancel   context.CancelFunc
        pending  []Message
        stopped  bool
        cancel   context.CancelFunc
        pending  []Message
        stopped  bool

+       started  time.Time

 }
 
 // SubagentWorkFile is the content of work.json.
 }
 
 // SubagentWorkFile is the content of work.json.