fix(pending-messages): run compaction in background, return 202 Accepted

LLM summarization (30-120s) was blocking HTTP response causing browser
timeouts. Now runs in goroutine and returns immediately. UI polls every
5s until completion.

internal/http/pending_messages.go

@@ -151,22 +151,25 @@ func (h *PendingMessagesHandler) handleCompact(w http.ResponseWriter, r *http.Re
 		return
 	}
 
-	// Detach from HTTP request context so LLM summarization isn't
-	// cancelled when the browser closes the connection.
-	ctx, cancel := context.WithTimeout(context.WithoutCancel(r.Context()), 180*time.Second)
-	defer cancel()
-
+	// Run compaction in background so the HTTP response returns immediately.
+	// The long-running LLM call (30-120s) was blocking the response, which
+	// caused browser WebSocket connections to drop (pong timeout).
 	keepRecent := h.keepRecent
 	if keepRecent <= 0 {
 		keepRecent = 15
 	}
-	remaining, err := channels.CompactGroup(ctx, h.store, req.ChannelName, req.HistoryKey, provider, model, keepRecent, h.maxTokens)
-	if err != nil {
-		slog.Warn("compact.failed", "channel", req.ChannelName, "key", req.HistoryKey, "error", err)
-		writeJSON(w, http.StatusInternalServerError, map[string]string{"error": err.Error()})
-		return
-	}
-	writeJSON(w, http.StatusOK, map[string]interface{}{"status": "ok", "method": "summarized", "remaining": remaining})
+	go func() {
+		ctx, cancel := context.WithTimeout(context.Background(), 180*time.Second)
+		defer cancel()
+		remaining, err := channels.CompactGroup(ctx, h.store, req.ChannelName, req.HistoryKey, provider, model, keepRecent, h.maxTokens)
+		if err != nil {
+			slog.Warn("compact.failed", "channel", req.ChannelName, "key", req.HistoryKey, "error", err)
+		} else {
+			slog.Info("compact.done", "channel", req.ChannelName, "key", req.HistoryKey, "remaining", remaining)
+		}
+	}()
+
+	writeJSON(w, http.StatusAccepted, map[string]interface{}{"status": "accepted", "method": "summarizing"})
 }
 
 // resolveProviderAndModel resolves the LLM provider+model for pending message compaction.

ui/web/src/i18n/locales/en/pending-messages.json

@@ -41,6 +41,8 @@
     "compacted": "Compaction complete",
     "compactedSummarized": "Summarized to {{count}} messages",
     "compactedDeleted": "All messages deleted (no LLM provider available)",
+    "compacting": "Compacting…",
+    "compactingDesc": "LLM summarization running in background. List will refresh automatically.",
     "failedCompact": "Compaction failed",
     "cleared": "Messages cleared",
     "failedClear": "Failed to clear messages"

ui/web/src/i18n/locales/vi/pending-messages.json

@@ -41,6 +41,8 @@
     "compacted": "Tóm tắt hoàn tất",
     "compactedSummarized": "Đã tóm tắt còn {{count}} tin nhắn",
     "compactedDeleted": "Đã xóa tất cả (không có LLM provider)",
+    "compacting": "Đang tóm tắt…",
+    "compactingDesc": "LLM đang tóm tắt ở nền. Danh sách sẽ tự động cập nhật.",
     "failedCompact": "Tóm tắt thất bại",
     "cleared": "Đã xóa tin nhắn",
     "failedClear": "Không thể xóa tin nhắn"

ui/web/src/i18n/locales/zh/pending-messages.json

@@ -41,6 +41,8 @@
     "compacted": "压缩完成",
     "compactedSummarized": "已压缩为 {{count}} 条消息",
     "compactedDeleted": "已删除全部（无可用LLM Provider）",
+    "compacting": "压缩中…",
+    "compactingDesc": "LLM正在后台进行摘要。列表将自动刷新。",
     "failedCompact": "压缩失败",
     "cleared": "消息已清除",
     "failedClear": "清除消息失败"

ui/web/src/pages/pending-messages/hooks/use-pending-messages.ts

@@ -48,13 +48,19 @@ export function usePendingMessages() {
           "/v1/pending-messages/compact",
           { channel_name: channel, history_key: key },
         );
-        const method = res?.method ?? "summarized";
-        toast.success(
-          i18next.t("pending-messages:toast.compacted"),
-          method === "deleted"
-            ? i18next.t("pending-messages:toast.compactedDeleted")
-            : i18next.t("pending-messages:toast.compactedSummarized", { count: res?.remaining ?? 0 }),
-        );
+        const method = res?.method ?? "summarizing";
+        if (method === "deleted") {
+          toast.success(
+            i18next.t("pending-messages:toast.compacted"),
+            i18next.t("pending-messages:toast.compactedDeleted"),
+          );
+        } else {
+          // Backend runs LLM in background — show info toast and let caller poll
+          toast.info(
+            i18next.t("pending-messages:toast.compacting"),
+            i18next.t("pending-messages:toast.compactingDesc"),
+          );
+        }
         return true;
       } catch (err) {
         toast.error(

ui/web/src/pages/pending-messages/pending-messages-page.tsx

@@ -45,10 +45,17 @@ export function PendingMessagesPage() {
     const key = `${group.channel_name}/${group.history_key}`;
     setActionLoading(key);
     setCompactingKey(key);
-    await compactGroup(group.channel_name, group.history_key);
-    setCompactingKey(null);
-    setActionLoading(null);
-    loadGroups();
+    const ok = await compactGroup(group.channel_name, group.history_key);
+    if (!ok) {
+      setCompactingKey(null);
+      setActionLoading(null);
+      return;
+    }
+    // Backend runs LLM in background — poll until done (has_summary changes).
+    // Compact button is already disabled when has_summary=true.
+    const poll = setInterval(() => loadGroups(), 5000);
+    setTimeout(() => loadGroups(), 2000); // quick first check
+    setTimeout(() => { clearInterval(poll); setCompactingKey(null); setActionLoading(null); }, 120_000);
   };
 
   const handleClear = async (group: PendingMessageGroup) => {