fix(responses): preserve incomplete terminal state

changelog/unreleased/responses-incomplete-terminal-state.md

@@ -0,0 +1,7 @@
+### English
+
+- Map upstream `finish_reason: length` results to the Responses API `incomplete` terminal state, including streaming events, request logs, statistics, and UI filters.
+
+### 中文
+
+- 将上游 `finish_reason: length` 结果映射为 Responses API 的 `incomplete` 终止状态，并同步支持流式事件、请求日志、统计与界面筛选。

frontend/src/api/logs.ts

@@ -107,6 +107,7 @@ export type RequestStatsWindow = {
 export type RequestStatsTotals = {
   requests: number
   ok: number
+  incomplete: number
   error: number
   canceled: number
   streaming: number

frontend/src/i18n/messages.ts

@@ -201,6 +201,7 @@ export const messages: Record<Lang, Dict> = {
     logsStreamNo: 'sync',
     logsFilterAll: 'All',
     logsFilterOk: 'OK',
+    logsFilterIncomplete: 'Incomplete',
     logsFilterError: 'Error',
     logsFilterCanceled: 'Canceled',
     logsFilterAccountAll: 'All accounts',
@@ -842,6 +843,7 @@ export const messages: Record<Lang, Dict> = {
     logsStreamNo: '同步',
     logsFilterAll: '全部',
     logsFilterOk: '成功',
+    logsFilterIncomplete: '未完成',
     logsFilterError: '失败',
     logsFilterCanceled: '取消',
     logsFilterAccountAll: '全部账号',

frontend/src/pages/LogsPage.tsx

@@ -46,7 +46,7 @@ import { useI18n } from '@/hooks/useI18n'
 import { accountProviderLabel } from '@/lib/provider'
 
 type PageTab = 'requests' | 'runtime'
-type RequestFilter = 'all' | 'ok' | 'error' | 'canceled'
+type RequestFilter = 'all' | 'ok' | 'incomplete' | 'error' | 'canceled'
 type RuntimeFilter = 'all' | 'info' | 'warn' | 'error'
 type StreamFilter = 'all' | 'stream' | 'sync'
 type TimeRange = 'all' | '1h' | '24h' | '7d' | 'custom'
@@ -56,7 +56,7 @@ type DateRangeValue = { start: DateValue; end: DateValue }
 
 function statusColor(status?: string): 'success' | 'warning' | 'danger' | 'default' {
   if (status === 'ok') return 'success'
-  if (status === 'streaming' || status === 'started') return 'warning'
+  if (status === 'streaming' || status === 'started' || status === 'incomplete') return 'warning'
   if (status === 'error' || status === 'canceled') return 'danger'
   return 'default'
 }
@@ -511,6 +511,7 @@ export function LogsPage() {
                 options={[
                   { id: 'all', label: t('logsFilterAll') },
                   { id: 'ok', label: t('logsFilterOk') },
+                  { id: 'incomplete', label: t('logsFilterIncomplete') },
                   { id: 'error', label: t('logsFilterError') },
                   { id: 'canceled', label: t('logsFilterCanceled') },
                 ]}

frontend/src/pages/OverviewPage.tsx

@@ -28,7 +28,7 @@ type StatsWindow = 1 | 24 | 168
 
 const EMPTY_STATS: RequestStats = {
   window: { from: '', to: '', hours: 24 },
-  totals: { requests: 0, ok: 0, error: 0, canceled: 0, streaming: 0, success_rate: 0 },
+  totals: { requests: 0, ok: 0, incomplete: 0, error: 0, canceled: 0, streaming: 0, success_rate: 0 },
   latency: {},
   tokens: { prompt: 0, completion: 0, cache_read: 0, total: 0 },
   status: [],
@@ -121,7 +121,7 @@ export function OverviewPage() {
 
   const metrics = [
     { label: t('metricRequests'), value: traffic.totals.requests as number | null, kind: 'compact' as const, detail: t('statsWindowHint', { window: t(hours === 1 ? 'statsWindow1h' : hours === 168 ? 'statsWindow7d' : 'statsWindow24h') }), ok: traffic.totals.requests > 0 },
-    { label: t('metricSuccess'), value: traffic.totals.success_rate, kind: 'percent' as const, detail: `${traffic.totals.ok} ${t('logsFilterOk')} · ${traffic.totals.error} ${t('logsFilterError')}`, ok: traffic.totals.requests === 0 || traffic.totals.success_rate >= 0.9 },
+    { label: t('metricSuccess'), value: traffic.totals.success_rate, kind: 'percent' as const, detail: `${traffic.totals.ok} ${t('logsFilterOk')} · ${traffic.totals.incomplete} ${t('logsFilterIncomplete')} · ${traffic.totals.error} ${t('logsFilterError')}`, ok: traffic.totals.requests === 0 || traffic.totals.success_rate >= 0.9 },
     { label: t('metricLatency'), value: traffic.latency.p95_ms ?? traffic.latency.avg_ms, kind: 'ms' as const, detail: `p50 ${formatLatency(traffic.latency.p50_ms)} · avg ${formatLatency(traffic.latency.avg_ms)}`, ok: traffic.latency.p95_ms == null || traffic.latency.p95_ms < 8000 },
     { label: t('metricTokens'), value: traffic.tokens.total, kind: 'compact' as const, detail: `${formatCompact(traffic.tokens.prompt)} / ${formatCompact(traffic.tokens.completion)}`, ok: true },
   ]

internal/accounts/request_logs.go

@@ -9,11 +9,12 @@ import (
 )
 
 const (
-	RequestStatusStarted   = "started"
-	RequestStatusStreaming = "streaming"
-	RequestStatusOK        = "ok"
-	RequestStatusError     = "error"
-	RequestStatusCanceled  = "canceled"
+	RequestStatusStarted    = "started"
+	RequestStatusStreaming  = "streaming"
+	RequestStatusOK         = "ok"
+	RequestStatusIncomplete = "incomplete"
+	RequestStatusError      = "error"
+	RequestStatusCanceled   = "canceled"
 
 	AttemptStatusStarted  = "started"
 	AttemptStatusOK       = "ok"
@@ -140,6 +141,7 @@ type RequestStatsWindow struct {
 type RequestStatsTotals struct {
 	Requests    int     `json:"requests"`
 	OK          int     `json:"ok"`
+	Incomplete  int     `json:"incomplete"`
 	Error       int     `json:"error"`
 	Canceled    int     `json:"canceled"`
 	Streaming   int     `json:"streaming"`

internal/gateway/compat_stream.go

@@ -538,6 +538,7 @@ func RelayResponsesStream(writer io.Writer, body io.Reader, requestID, model str
 	if err != nil {
 		return stats, err
 	}
+	stats.FinishReason = output.finishReason
 	if err := closeReasoning(); err != nil {
 		return stats, err
 	}
@@ -608,8 +609,9 @@ func RelayResponsesStream(writer io.Writer, body io.Reader, requestID, model str
 			return stats, err
 		}
 	}
-	completed := responsesResponse(requestID, model, content, output.reasoning.String(), calls, derefInt(stats.PromptTokens), derefInt(stats.CompletionTokens))
-	if err := eventWriter.write("response.completed", map[string]any{"type": "response.completed", "response": completed}); err != nil {
+	terminal := responsesTerminalForFinishReason(output.finishReason)
+	response := responsesResponse(requestID, model, content, output.reasoning.String(), calls, derefInt(stats.PromptTokens), derefInt(stats.CompletionTokens), output.finishReason)
+	if err := eventWriter.write(terminal.event, map[string]any{"type": terminal.event, "response": response}); err != nil {
 		return stats, err
 	}
 	return stats, nil

internal/gateway/openai_stream.go

@@ -38,6 +38,7 @@ type StreamRelayStats struct {
 	Credits          *float64
 	ConsumedCredits  *float64
 	Model            string
+	FinishReason     string
 	FirstTokenAt     *time.Time
 	SSEEventCount    int
 	BytesRead        int64

internal/gateway/responses.go

@@ -43,13 +43,14 @@ func (h *Handler) HandleResponses(w http.ResponseWriter, r *http.Request) {
 		writeCompatibilityOpenAIError(w, err)
 		return
 	}
-	h.finishCompatibility(execution, result.AccountID, result.Provider, result.Routing, accounts.RequestStatusOK, 0, &StreamRelayStats{
+	requestStatus := responsesRequestStatus(result.FinishReason)
+	h.finishCompatibility(execution, result.AccountID, result.Provider, result.Routing, requestStatus, 0, &StreamRelayStats{
 		PromptTokens: ptrInt(result.PromptTokens), CompletionTokens: ptrInt(result.CompletionTokens),
 		CacheReadTokens: result.CacheReadTokens, CacheWriteTokens: result.CacheWriteTokens,
 		CachedTokens: result.CachedTokens, UsageSource: result.UsageSource, Credits: result.Credits,
-		ConsumedCredits: result.ConsumedCredits, Model: result.Model,
+		ConsumedCredits: result.ConsumedCredits, Model: result.Model, FinishReason: result.FinishReason,
 	}, nil, result.AttemptCount, result.ReasoningLevel)
-	response := responsesResponse(execution.RequestID, firstNonEmpty(result.Model, execution.PublicModel), result.Content, result.Reasoning, decodeOpenAIToolCalls(result.ToolCalls), result.PromptTokens, result.CompletionTokens)
+	response := responsesResponse(execution.RequestID, firstNonEmpty(result.Model, execution.PublicModel), result.Content, result.Reasoning, decodeOpenAIToolCalls(result.ToolCalls), result.PromptTokens, result.CompletionTokens, result.FinishReason)
 	translate.RestoreResponseToolNames(response, execution.Request.ResponseToolNames)
 	writeJSON(w, http.StatusOK, response)
 }
@@ -71,6 +72,9 @@ func (h *Handler) handleResponsesStream(w http.ResponseWriter, r *http.Request,
 	writer := compatibilityStreamWriter(w)
 	stats, relayErr := RelayResponsesStream(writer, upstream.Response.Body, execution.RequestID, firstNonEmpty(execution.PublicModel, execution.Request.Model), execution.Request.ResponseToolNames)
 	status := streamRequestStatus(relayErr)
+	if relayErr == nil {
+		status = responsesRequestStatus(stats.FinishReason)
+	}
 	if r.Context().Err() != nil || errors.Is(relayErr, context.Canceled) || errors.Is(relayErr, context.DeadlineExceeded) {
 		status = accounts.RequestStatusCanceled
 	}
@@ -100,12 +104,43 @@ func writeCompatibilityOpenAIError(w http.ResponseWriter, err error) {
 	WriteClassifiedErr(w, err)
 }
 
-func responsesResponse(requestID, model, content, reasoning string, toolCalls []proxyToolCall, promptTokens, completionTokens int) map[string]any {
-	return map[string]any{
-		"id": "resp_" + requestID, "object": "response", "created_at": time.Now().Unix(), "status": "completed", "model": model,
+type responsesTerminal struct {
+	status            string
+	event             string
+	incompleteDetails map[string]any
+}
+
+func responsesTerminalForFinishReason(finishReason string) responsesTerminal {
+	if finishReason == "length" {
+		return responsesTerminal{
+			status: "incomplete",
+			event:  "response.incomplete",
+			incompleteDetails: map[string]any{
+				"reason": "max_output_tokens",
+			},
+		}
+	}
+	return responsesTerminal{status: "completed", event: "response.completed"}
+}
+
+func responsesRequestStatus(finishReason string) string {
+	if responsesTerminalForFinishReason(finishReason).status == "incomplete" {
+		return accounts.RequestStatusIncomplete
+	}
+	return accounts.RequestStatusOK
+}
+
+func responsesResponse(requestID, model, content, reasoning string, toolCalls []proxyToolCall, promptTokens, completionTokens int, finishReason string) map[string]any {
+	terminal := responsesTerminalForFinishReason(finishReason)
+	response := map[string]any{
+		"id": "resp_" + requestID, "object": "response", "created_at": time.Now().Unix(), "status": terminal.status, "model": model,
 		"output": responsesOutputItems(requestID, content, reasoning, toolCalls),
 		"usage":  responsesUsage(promptTokens, completionTokens),
 	}
+	if terminal.incompleteDetails != nil {
+		response["incomplete_details"] = terminal.incompleteDetails
+	}
+	return response
 }
 
 func responsesOutputItems(requestID, content, reasoning string, toolCalls []proxyToolCall) []any {

internal/gateway/responses_test.go

@@ -0,0 +1,56 @@
+package gateway
+
+import (
+	"encoding/json"
+	"strings"
+	"testing"
+)
+
+func TestResponsesResponseMapsLengthToIncomplete(t *testing.T) {
+	response := responsesResponse("req", "model", "", "reasoning", nil, 10, 32, "length")
+	if response["status"] != "incomplete" {
+		t.Fatalf("status=%v", response["status"])
+	}
+	details := response["incomplete_details"].(map[string]any)
+	if details["reason"] != "max_output_tokens" {
+		t.Fatalf("details=%#v", details)
+	}
+	if responsesRequestStatus("length") != "incomplete" || responsesRequestStatus("stop") != "ok" {
+		t.Fatal("request status mapping changed")
+	}
+}
+
+func TestRelayResponsesStreamEmitsIncomplete(t *testing.T) {
+	upstream := strings.NewReader("data: {\"choices\":[{\"delta\":{\"reasoning_content\":\"thinking\"}}]}\n\n" +
+		"data: {\"choices\":[{\"finish_reason\":\"length\"}],\"usage\":{\"prompt_tokens\":100,\"completion_tokens\":32}}\n\n" +
+		"data: [DONE]\n\n")
+	var output strings.Builder
+	stats, err := RelayResponsesStream(&output, upstream, "req", "model", nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if stats.FinishReason != "length" {
+		t.Fatalf("finish reason=%q", stats.FinishReason)
+	}
+	body := output.String()
+	if !strings.Contains(body, "event: response.incomplete") || strings.Contains(body, "event: response.completed") {
+		t.Fatalf("wrong terminal event:\n%s", body)
+	}
+	var terminal map[string]any
+	for _, line := range strings.Split(body, "\n") {
+		if !strings.HasPrefix(line, "data: ") {
+			continue
+		}
+		var event map[string]any
+		if json.Unmarshal([]byte(strings.TrimPrefix(line, "data: ")), &event) == nil && event["type"] == "response.incomplete" {
+			terminal = event
+		}
+	}
+	if terminal == nil {
+		t.Fatal("missing response.incomplete payload")
+	}
+	response := terminal["response"].(map[string]any)
+	if response["status"] != "incomplete" || response["incomplete_details"].(map[string]any)["reason"] != "max_output_tokens" {
+		t.Fatalf("response=%#v", response)
+	}
+}

internal/runtime/manager_test.go

@@ -853,7 +853,19 @@ func TestManagerEscalatesConsecutiveRestartBackoffSeparately(t *testing.T) {
 	if third == second {
 		t.Fatal("manager reused exited process")
 	}
-	item, ok := manager.Pool().ByID(account.ID)
+	// fakeStarter announces the process before Start returns. Wait until the
+	// manager has registered that process and copied the restart count into the
+	// pool instead of racing the remainder of startAccount.
+	deadline := time.Now().Add(time.Second)
+	var item executor.Item
+	var ok bool
+	for time.Now().Before(deadline) {
+		item, ok = manager.Pool().ByID(account.ID)
+		if ok && item.Restarts == 2 && item.RestartBackoffLevel == 2 {
+			break
+		}
+		time.Sleep(time.Millisecond)
+	}
 	if !ok {
 		t.Fatal("account disappeared during consecutive recovery")
 	}

internal/store/request_logs.go

@@ -181,9 +181,10 @@ func (s *Store) SummarizeRequestLogs(ctx context.Context, from, to time.Time) (a
 
 	row := s.db.QueryRowContext(ctx, `
 SELECT
-  COUNT(*),
-  COALESCE(SUM(CASE WHEN status = 'ok' THEN 1 ELSE 0 END), 0),
-  COALESCE(SUM(CASE WHEN status = 'error' THEN 1 ELSE 0 END), 0),
+	  COUNT(*),
+	  COALESCE(SUM(CASE WHEN status = 'ok' THEN 1 ELSE 0 END), 0),
+	  COALESCE(SUM(CASE WHEN status = 'incomplete' THEN 1 ELSE 0 END), 0),
+	  COALESCE(SUM(CASE WHEN status = 'error' THEN 1 ELSE 0 END), 0),
   COALESCE(SUM(CASE WHEN status = 'canceled' THEN 1 ELSE 0 END), 0),
   COALESCE(SUM(CASE WHEN stream = 1 THEN 1 ELSE 0 END), 0),
   AVG(CASE WHEN latency_ms IS NOT NULL THEN latency_ms END),
@@ -194,7 +195,7 @@ SELECT
 FROM request_logs`+where, args...)
 	var avgLatency, avgTTFB sql.NullFloat64
 	if err := row.Scan(
-		&stats.Totals.Requests, &stats.Totals.OK, &stats.Totals.Error, &stats.Totals.Canceled, &stats.Totals.Streaming,
+		&stats.Totals.Requests, &stats.Totals.OK, &stats.Totals.Incomplete, &stats.Totals.Error, &stats.Totals.Canceled, &stats.Totals.Streaming,
 		&avgLatency, &avgTTFB, &stats.Tokens.Prompt, &stats.Tokens.Completion, &stats.Tokens.CacheRead,
 	); err != nil {
 		return accounts.RequestStats{}, fmt.Errorf("summarize request logs: %w", err)

internal/store/request_logs_test.go

@@ -347,6 +347,7 @@ func TestSummarizeRequestLogs(t *testing.T) {
 	insert(base.Add(20*time.Minute), accounts.RequestStatusOK, "glm-5.3", "acc_a", "qoder", "", 200, 10, 20, true)
 	insert(base.Add(90*time.Minute), accounts.RequestStatusError, "qwen3.7-plus", "acc_b", "workbuddy", accounts.KindRateLimit, 400, 8, 0, false)
 	insert(base.Add(2*time.Hour), accounts.RequestStatusCanceled, "glm-5.3", "acc_a", "qoder", "", 0, 0, 0, false)
+	insert(base.Add(2*time.Hour+15*time.Minute), accounts.RequestStatusIncomplete, "glm-5.3", "acc_a", "qoder", "", 0, 0, 0, false)
 	insert(base.Add(-30*time.Hour), accounts.RequestStatusOK, "old", "acc_a", "qoder", "", 50, 1, 1, false)
 
 	from := time.Date(2026, 8, 28, 10, 0, 0, 0, time.UTC)
@@ -355,10 +356,10 @@ func TestSummarizeRequestLogs(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	if stats.Totals.Requests != 4 || stats.Totals.OK != 2 || stats.Totals.Error != 1 || stats.Totals.Canceled != 1 || stats.Totals.Streaming != 1 {
+	if stats.Totals.Requests != 5 || stats.Totals.OK != 2 || stats.Totals.Incomplete != 1 || stats.Totals.Error != 1 || stats.Totals.Canceled != 1 || stats.Totals.Streaming != 1 {
 		t.Fatalf("totals = %+v", stats.Totals)
 	}
-	if stats.Totals.SuccessRate != 0.5 {
+	if stats.Totals.SuccessRate != 0.4 {
 		t.Fatalf("success rate = %v", stats.Totals.SuccessRate)
 	}
 	if stats.Tokens.Prompt != 30 || stats.Tokens.Completion != 54 || stats.Tokens.Total != 84 {
@@ -376,19 +377,19 @@ func TestSummarizeRequestLogs(t *testing.T) {
 	if len(stats.Errors) != 1 || stats.Errors[0].Key != accounts.KindRateLimit || stats.Errors[0].Count != 1 {
 		t.Fatalf("errors = %+v", stats.Errors)
 	}
-	if len(stats.Models) == 0 || stats.Models[0].Key != "glm-5.3" || stats.Models[0].Count != 3 {
+	if len(stats.Models) == 0 || stats.Models[0].Key != "glm-5.3" || stats.Models[0].Count != 4 {
 		t.Fatalf("models = %+v", stats.Models)
 	}
-	if len(stats.Accounts) == 0 || stats.Accounts[0].Key != "acc_a" || stats.Accounts[0].Count != 3 {
+	if len(stats.Accounts) == 0 || stats.Accounts[0].Key != "acc_a" || stats.Accounts[0].Count != 4 {
 		t.Fatalf("accounts = %+v", stats.Accounts)
 	}
-	if len(stats.Providers) == 0 || stats.Providers[0].Key != "qoder" || stats.Providers[0].Count != 3 {
+	if len(stats.Providers) == 0 || stats.Providers[0].Key != "qoder" || stats.Providers[0].Count != 4 {
 		t.Fatalf("providers = %+v", stats.Providers)
 	}
 	if len(stats.Series) != 3 {
 		t.Fatalf("series len = %d %+v", len(stats.Series), stats.Series)
 	}
-	if stats.Series[0].Requests != 2 || stats.Series[1].Requests != 1 || stats.Series[2].Requests != 1 {
+	if stats.Series[0].Requests != 2 || stats.Series[1].Requests != 1 || stats.Series[2].Requests != 2 {
 		t.Fatalf("series = %+v", stats.Series)
 	}
 

internal/webui/static/index.html

@@ -22,7 +22,7 @@
     <link rel="preconnect" href="https://fonts.googleapis.com" />
     <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
     <link href="https://fonts.googleapis.com/css2?family=IBM+Plex+Mono:wght@400;500&family=Outfit:wght@400;500;600;700&display=swap" rel="stylesheet" />
-    <script type="module" crossorigin src="/assets/index-DPnzAZji.js"></script>
+    <script type="module" crossorigin src="/assets/index-cijSDZXp.js"></script>
     <link rel="stylesheet" crossorigin href="/assets/index-WwQDPEz6.css">
   </head>
   <body>