Retry daemon remote polls

Author: SqlRush

cmd/claude/main.go

@@ -1240,11 +1240,7 @@ func fetchDaemonRemoteEvents(ctx context.Context, registration remotepkg.Registr
 		if ws.Error == "" || pollURL == "" {
 			return fetch
 		}
-		poll := remotepkg.FetchPollEvents(ctx, remotepkg.PollOptions{
-			PollURL:   pollURL,
-			Cursor:    cursor,
-			AuthToken: authToken,
-		})
+		poll := remotepkg.FetchPollEvents(ctx, daemonRemotePollOptions(pollURL, cursor, authToken))
 		return daemonRemoteFetch{
 			Transport:     "poll",
 			StatusCode:    poll.StatusCode,
@@ -1255,11 +1251,7 @@ func fetchDaemonRemoteEvents(ctx context.Context, registration remotepkg.Registr
 			FallbackError: ws.Error,
 		}
 	}
-	poll := remotepkg.FetchPollEvents(ctx, remotepkg.PollOptions{
-		PollURL:   pollURL,
-		Cursor:    cursor,
-		AuthToken: authToken,
-	})
+	poll := remotepkg.FetchPollEvents(ctx, daemonRemotePollOptions(pollURL, cursor, authToken))
 	return daemonRemoteFetch{
 		Transport:    "poll",
 		StatusCode:   poll.StatusCode,
@@ -1270,6 +1262,17 @@ func fetchDaemonRemoteEvents(ctx context.Context, registration remotepkg.Registr
 	}
 }
 
+func daemonRemotePollOptions(pollURL string, cursor string, authToken string) remotepkg.PollOptions {
+	return remotepkg.PollOptions{
+		PollURL:           pollURL,
+		Cursor:            cursor,
+		AuthToken:         authToken,
+		RetryAttempts:     1,
+		RetryInitialDelay: 100 * time.Millisecond,
+		RetryMaxDelay:     500 * time.Millisecond,
+	}
+}
+
 type daemonRemoteDeliveryResult struct {
 	Delivered        int
 	Duplicates       int

cmd/claude/main_test.go

@@ -404,6 +404,57 @@ func TestRunDaemonRemotePollInjectsRemoteTriggers(t *testing.T) {
 	}
 }
 
+func TestRunDaemonRemotePollRetriesTransientPoll(t *testing.T) {
+	dir := t.TempDir()
+	transcriptPath := filepath.Join(dir, "session.jsonl")
+	sessionID := contracts.ID("sess_daemon_remote_poll_retry")
+	calls := 0
+	var auths []string
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		calls++
+		auths = append(auths, r.Header.Get("Authorization"))
+		if calls == 1 {
+			w.Header().Set("Retry-After", "0")
+			http.Error(w, "try again", http.StatusServiceUnavailable)
+			return
+		}
+		w.Header().Set("content-type", "application/json")
+		_, _ = w.Write([]byte(`{"next_cursor":"cursor-retry","events":[]}`))
+	}))
+	defer server.Close()
+	if err := remotepkg.WriteRegistrationState(remotepkg.SessionRegistrationPath(transcriptPath, sessionID), remotepkg.RegistrationState{
+		SessionID:       sessionID,
+		RuntimeState:    remotepkg.RegistrationRegistered,
+		PollURL:         server.URL + "/poll",
+		RemoteSessionID: "remote-session",
+	}); err != nil {
+		t.Fatal(err)
+	}
+	runner := conversation.Runner{
+		SessionID:        sessionID,
+		SessionPath:      transcriptPath,
+		WorkingDirectory: dir,
+		MCP: &conversation.MCPConfig{UserSettings: contracts.Settings{
+			Remote: &contracts.RemoteSetting{AuthToken: "poll-token"},
+		}},
+	}
+
+	result := runDaemonRemotePoll(context.Background(), runner, time.Unix(200, 0).UTC())
+	if result.StructuredContent["runtime_state"] != remotepkg.PumpRunning || result.StructuredContent["status_code"] != http.StatusOK || result.StructuredContent["attempt_count"] != 2 || result.StructuredContent["last_cursor"] != "cursor-retry" || result.StructuredContent["event_count"] != 0 || result.StructuredContent["error_count"] != 0 {
+		t.Fatalf("retry poll = %#v", result.StructuredContent)
+	}
+	if calls != 2 || len(auths) != 2 || auths[0] != "Bearer poll-token" || auths[1] != "Bearer poll-token" {
+		t.Fatalf("calls/auths = %d %#v", calls, auths)
+	}
+	pump, err := remotepkg.LoadPumpState(remotepkg.SessionPumpPath(transcriptPath, sessionID))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if pump.StatusCode != http.StatusOK || pump.AttemptCount != 2 || pump.LastCursor != "cursor-retry" || pump.ErrorCount != 0 {
+		t.Fatalf("pump = %#v", pump)
+	}
+}
+
 func TestRunDaemonRemotePollSkipsExpiredLease(t *testing.T) {
 	dir := t.TempDir()
 	transcriptPath := filepath.Join(dir, "session.jsonl")

docs/cc-100-roadmap.md

@@ -135,6 +135,8 @@ M10 补充：remote ack 和 lease renew 的 transient retry 现在会优先遵
 
 M10 补充：remote poll fetch 现在也支持 transient retry，遇到 transport error、408、429 或 5xx 时可按 PollOptions 重试，并同样优先遵守服务端 `Retry-After` header；PollResult 暴露 `attempt_count` 便于 daemon/pump 审计。
 
+M10 补充：daemon remote poll 与 WebSocket fallback poll 现在实际启用 transient retry（一次短退避，优先遵守 `Retry-After`），并把 poll `attempt_count` 写入 structured result、`remote-pump.json` 和 `/status show remote`。
+
 M10 补充：remote pump state 现在持久化 `attempt_count`，`/status show remote` 会显示 `attempts N`，让 poll/WebSocket pump 的重试行为能被 CLI 状态页审计。
 
 M10 补充：remote WebSocket upgrade 失败后的重连退避现在会读取服务端 `Retry-After` header（秒数或 HTTP-date），Fetch/Stream 两条 WebSocket 路径都会优先遵守云端限流退避，再回退到本地指数退避。

docs/claude-code-go-rewrite-plan.md

@@ -758,6 +758,7 @@ test/parity/                 # golden tests against TS/official behavior
 - 本轮补充：remote poll/WebSocket 事件现在会解析 `ack_url`/`ackUrl`/`acknowledge_url`/`receipt_url` 和 `lease_id`/`lease_expires_at` 及 `ack`/`lease` nested object 元数据；daemon 会在 delivered/duplicate/failed 后对注册 poll/websocket 同源的 ack URL 做 best-effort POST，带 Bearer auth 和 event/status/sent_count/duplicate/error payload，并对 transport error、408/429/5xx 做一次短退避重试；pump state、structured result 和 `/status show remote` 会记录 ack event/sent/error 与 lease event 计数；非同源 ack URL 会被拒绝且脱敏；已过期 lease 会被跳过投递并 ack `expired`，同时记录 `lease_expired_count`。更深的服务端协议协商仍未完成。
 - 本轮补充：remote ack 和 lease renew 的 transient retry 现在会优先遵守服务端 `Retry-After` header（秒数或 HTTP-date），再回退到本地指数退避，并继续受最大退避上限约束，减少云端 429/503 限流时的协议偏差。
 - 本轮补充：remote poll fetch 现在也支持 transient retry，遇到 transport error、408、429 或 5xx 时可按 PollOptions 重试，并同样优先遵守服务端 `Retry-After` header；PollResult 暴露 `attempt_count` 便于 daemon/pump 审计。
+- 本轮补充：daemon remote poll 与 WebSocket fallback poll 现在实际启用 transient retry（一次短退避，优先遵守 `Retry-After`），并把 poll `attempt_count` 写入 structured result、`remote-pump.json` 和 `/status show remote`。
 - 本轮补充：remote pump state 现在持久化 `attempt_count`，`/status show remote` 会显示 `attempts N`，让 poll/WebSocket pump 的重试行为能被 CLI 状态页审计。
 - 本轮补充：remote WebSocket upgrade 失败后的重连退避现在会读取服务端 `Retry-After` header（秒数或 HTTP-date），Fetch/Stream 两条 WebSocket 路径都会优先遵守云端限流退避，再回退到本地指数退避。
 - 本轮补充：remote WebSocket `FetchWebSocketEvents`/`StreamWebSocketEvents` result 现在暴露 `status_code` 和 `attempt_count`；成功握手记录 101，upgrade 失败记录服务端 HTTP status，daemon tick/常驻 stream 会把这些字段写入 `remote-pump.json`、structured result 和 `/status show remote` 审计面。