package vip.mate.channel.web; import io.swagger.v3.oas.annotations.Operation; import io.swagger.v3.oas.annotations.tags.Tag; import com.fasterxml.jackson.databind.ObjectMapper; import lombok.RequiredArgsConstructor; import lombok.extern.slf4j.Slf4j; import org.springframework.core.io.FileSystemResource; import org.springframework.core.io.Resource; import org.springframework.http.HttpHeaders; import org.springframework.http.MediaType; import org.springframework.http.ResponseEntity; import org.springframework.security.core.Authentication; import org.springframework.web.bind.annotation.*; import org.springframework.web.multipart.MultipartFile; import org.springframework.web.servlet.mvc.method.annotation.SseEmitter; import vip.mate.common.result.R; import vip.mate.agent.AgentService; import vip.mate.agent.model.AgentEntity; import vip.mate.approval.ApprovalWorkflowService; import vip.mate.approval.MetadataDecision; import vip.mate.approval.PendingApproval; import vip.mate.approval.ResolveOutcome; import vip.mate.memory.event.ConversationCompletionPublisher; import vip.mate.workspace.conversation.ConversationService; import vip.mate.workspace.conversation.model.MessageContentPart; import vip.mate.workspace.conversation.model.MessageEntity; import java.net.URLEncoder; import java.nio.charset.StandardCharsets; import java.nio.file.Files; import java.nio.file.Path; import java.nio.file.Paths; import java.io.IOException; import reactor.core.Disposable; import java.util.ArrayList; import java.util.LinkedHashMap; import java.util.List; import java.util.Map; import java.util.concurrent.ExecutorService; import java.util.concurrent.Executors; import java.util.concurrent.atomic.AtomicBoolean; /** * Web 渠道聊天接口 * 提供 SSE 流式对话和同步对话能力 * * @author MateClaw Team */ @Tag(name = "Web聊天") @Slf4j @RestController @RequestMapping("/api/v1/chat") @RequiredArgsConstructor public class ChatController { private final AgentService agentService; private final ConversationService conversationService; private final ApprovalWorkflowService approvalService; private final ChatStreamTracker streamTracker; private final ObjectMapper objectMapper; private final ConversationCompletionPublisher completionPublisher; private final Path uploadRoot = Paths.get("data", "chat-uploads"); // 使用虚拟线程池处理 SSE（Java 17+ 兼容，Java 21 可用 Executors.newVirtualThreadPerTaskExecutor()） private final ExecutorService sseExecutor = Executors.newCachedThreadPool(); /** * SSE 流式对话（支持断线重连） *

* 正常请求：保存用户消息，启动 Flux 生产者，通过 StreamTracker 广播事件。 * 重连请求（reconnect=true）：附着到仍在运行的流，回放已缓冲事件后接收实时增量。 */ @Operation(summary = "结构化 SSE 流式对话（支持重连）") @PostMapping(value = "/stream", produces = MediaType.TEXT_EVENT_STREAM_VALUE) public SseEmitter chatStream( @RequestBody ChatStreamRequest request, @RequestHeader(value = "X-Workspace-Id", required = false) Long workspaceId, Authentication auth) { String conversationId = request.getConversationId() != null ? request.getConversationId() : "default"; // SSE 超时设为 10 分钟，覆盖 servlet 默认的 30s，避免长回答被中断 // RFC-058 PR-1: Utf8SseEmitter 显式声明 charset=UTF-8，防止中文在 Windows 中文 Chrome / 部分代理处乱码 SseEmitter emitter = new Utf8SseEmitter(10 * 60 * 1000L); // ---- 分支 A：断线重连 ---- if (Boolean.TRUE.equals(request.getReconnect())) { String reconnectUser = auth != null ? auth.getName() : "anonymous"; log.info("SSE reconnect: conversationId={}, user={}", conversationId, reconnectUser); // 校验会话归属 if (!conversationService.isConversationOwner(conversationId, reconnectUser)) { try { sendEvent(emitter, "error", Map.of("message", "无权访问该会话")); } catch (IOException e) { log.warn("SSE reconnect auth error send failed: {}", e.getMessage()); } emitter.complete(); return emitter; } registerEmitterCallbacks(emitter, conversationId); // Issue #17 — distinguish "stream truly completed on this node" // from "stream is running on another node (multi-node deployment // without sticky session)". They look identical from attach()'s // boolean return, but the user-facing remediation is different. boolean existsLocally = streamTracker.streamExistsOnThisNode(conversationId); long lastEventId = request.getLastEventId() == null ? 0L : request.getLastEventId(); boolean attached = streamTracker.attach(conversationId, emitter, lastEventId); if (!attached) { try { if (existsLocally) { // RunState exists here but is done — stream finished normally sendEvent(emitter, "done", Map.of("status", "completed")); } else { // No RunState on this node. Either: // (a) The stream finished long ago and was cleaned up, OR // (b) Multi-node deployment routed this reconnect to a // DIFFERENT node than the originating one. The CE // build assumes single-instance (see ChatStreamTracker // class javadoc and rfc-054 §0); LB must be configured // for sticky session by conversationId. // We can't tell (a) from (b) at this layer, so emit an // explicit code the front-end can surface to operators. log.info("SSE reconnect: no RunState locally for conversationId={} — " + "either completed-and-cleaned or running on another node", conversationId); sendEvent(emitter, "done", Map.of( "status", "stream_not_local", "message", "Stream is not active on this node. " + "If you're running a multi-node deployment, " + "verify the load balancer is configured for sticky " + "session by conversationId. See deploy/multi-node-deployment.md." )); } } catch (IOException e) { log.warn("SSE reconnect done send error: {}", e.getMessage()); } emitter.complete(); } return emitter; } // ---- 分支 B：正常请求 ---- Long agentId = request.getAgentId(); String message = request.getMessage() != null ? request.getMessage() : ""; if (auth == null) { try { sendEvent(emitter, "error", Map.of("message", "未登录，请先登录")); } catch (IOException e) { log.warn("SSE auth error send failed: {}", e.getMessage()); } emitter.complete(); return emitter; } String username = auth.getName(); log.info("SSE chat: agentId={}, conversationId={}, user={}", agentId, conversationId, username); // ---- Workspace 边界校验：确保 agent 属于当前 workspace ---- if (agentId != null) { AgentEntity agent = agentService.getAgent(agentId); if (agent != null && agent.getWorkspaceId() != null) { long wsId = workspaceId != null ? workspaceId : 1L; if (!agent.getWorkspaceId().equals(wsId)) { log.warn("Chat workspace mismatch: agent {} belongs to workspace {}, request workspace {}", agentId, agent.getWorkspaceId(), wsId); try { sendEvent(emitter, "error", Map.of("message", "Agent 不属于当前工作区")); sendEvent(emitter, "done", Map.of("status", "completed")); } catch (IOException e) { log.warn("SSE workspace error send failed: {}", e.getMessage()); } emitter.complete(); return emitter; } } } // ---- 审批命令拦截：/approve、/deny 走 SSE 流式 replay ---- String normalizedMsg = message.trim().toLowerCase(); boolean isApprovalCommand = "/approve".equals(normalizedMsg) || "approve".equals(normalizedMsg); boolean isDenyCommand = "/deny".equals(normalizedMsg) || "deny".equals(normalizedMsg); if (isApprovalCommand || isDenyCommand) { PendingApproval pending = approvalService.findPendingByConversation(conversationId); if (pending == null) { try { sendEvent(emitter, "error", Map.of("message", "当前没有待审批的工具调用")); sendEvent(emitter, "done", Map.of("status", "completed")); } catch (IOException e) { /* ignore */ } emitter.complete(); return emitter; } // deny: workflow.resolve handles DB + metadata + memory atomically. if (isDenyCommand) { ResolveOutcome denyOutcome = approvalService.resolve(pending.getPendingId(), username, "denied"); conversationService.removeApprovalPlaceholders(conversationId); log.info("[Approval-Stream] User {} denied pending {} for conversation {} (dbSynced={}, msgRewritten={})", username, pending.getPendingId(), conversationId, denyOutcome.dbSynced(), denyOutcome.messagesRewritten()); } // approve: atomic resolveAndConsume; workflow handles DB + metadata + memory. PendingApproval consumed = null; if (isApprovalCommand) { ResolveOutcome consumeOutcome = approvalService.resolveAndConsume(pending.getPendingId(), username); if (consumeOutcome.isAlreadyResolved()) { try { sendEvent(emitter, "error", Map.of("message", "审批记录已过期或已被处理")); sendEvent(emitter, "done", Map.of("status", "completed")); } catch (IOException e2) { /* ignore */ } emitter.complete(); return emitter; } consumed = consumeOutcome.consumedSnapshot(); // Clear residual approval placeholder messages so the LLM context for // replay doesn't include "[Awaiting approval]" text artifacts. conversationService.removeApprovalPlaceholders(conversationId); log.info("[Approval-Stream] User {} approved pending {} for conversation {} (msgRewritten={})", username, consumed.getPendingId(), conversationId, consumeOutcome.messagesRewritten()); } final PendingApproval finalConsumed = consumed; final String decision = isApprovalCommand ? "approved" : "denied"; streamTracker.register(conversationId); Long approvalAgentId = parseLongOrNull(pending.getAgentId()); streamTracker.bindRunMeta(conversationId, approvalAgentId, username); registerEmitterCallbacks(emitter, conversationId); streamTracker.attach(conversationId, emitter); AtomicBoolean approvalEmitterDone = new AtomicBoolean(false); sseExecutor.execute(() -> { StreamAccumulator accumulator = new StreamAccumulator(); AtomicBoolean finalized = new AtomicBoolean(false); try { // 广播 approval_resolved 事件 broadcastEvent(conversationId, "tool_approval_resolved", Map.of( "pendingId", pending.getPendingId(), "decision", decision, "toolName", pending.getToolName(), "timestamp", System.currentTimeMillis() )); if ("denied".equals(decision)) { String denyMsg = "用户拒绝执行工具 " + pending.getToolName(); MessageEntity savedAssistant = conversationService.saveMessage(conversationId, "assistant", denyMsg); broadcastEvent(conversationId, "message_start", Map.of("role", "assistant")); broadcastEvent(conversationId, "content_delta", Map.of("delta", denyMsg)); broadcastEvent(conversationId, "message_complete", Map.of("status", "completed")); broadcastEvent(conversationId, "done", buildDonePayload( conversationId, "completed", savedAssistant, 0, 0, isAssistantPersisted(savedAssistant), conversationService.getMessageCount(conversationId))); // deny 是正常 turn 终结，用户可能在 awaiting_approval 阶段排了消息 ChatStreamTracker.CompletionResult denyCr = streamTracker.completeAndConsumeIfLast(conversationId); if (denyCr.allDone() && denyCr.queuedInput() != null) { startQueuedMessage(conversationId, emitter, approvalEmitterDone, denyCr.queuedInput(), username); } else { completeEmitterQuietly(emitter, approvalEmitterDone); } return; } // approved: 使用已原子消费的记录触发 replay 流 if (finalConsumed == null) { broadcastEvent(conversationId, "error", Map.of("message", "审批记录已被消费")); broadcastEvent(conversationId, "done", Map.of("status", "completed")); // 审批记录被另一个请求消费，但用户可能在等待期间排了消息 ChatStreamTracker.CompletionResult consumedNullCr = streamTracker.completeAndConsumeIfLast(conversationId); if (consumedNullCr.allDone() && consumedNullCr.queuedInput() != null) { startQueuedMessage(conversationId, emitter, approvalEmitterDone, consumedNullCr.queuedInput(), username); } else { completeEmitterQuietly(emitter, approvalEmitterDone); } return; } Long replayAgentId = finalConsumed.getAgentId() != null ? Long.parseLong(finalConsumed.getAgentId()) : agentId; broadcastEvent(conversationId, "message_start", Map.of("role", "assistant")); // 不含工具名的中性 prompt（对齐 IM 渠道，防止 fallthrough 时误导 LLM） String replayPrompt = "继续执行已批准的工具调用。"; streamTracker.incrementFlux(conversationId); // RFC-063r §2.12: prefer the persisted Memento snapshot // (covers cross-restart approval where the original channel // is gone) and fall back to a fresh web-origin // ChatOrigin when none was captured. vip.mate.agent.context.ChatOrigin replayOrigin = approvalService.restoreChatOrigin(finalConsumed.getChatOrigin()); if (replayOrigin == vip.mate.agent.context.ChatOrigin.EMPTY) { replayOrigin = vip.mate.agent.context.ChatOrigin.web( conversationId, username, workspaceId, null); } Disposable disposable = agentService.chatWithReplayStream( replayAgentId, replayPrompt, conversationId, finalConsumed.getToolCallPayload(), username, replayOrigin) .doOnNext(delta -> { if (approvalEmitterDone.get()) return; try { accumulator.accept(delta, conversationId); } catch (Exception e) { log.warn("SSE replay broadcast error: {}", e.getMessage()); } }) .doOnComplete(() -> { if (!finalized.compareAndSet(false, true)) return; // Force-recycle short-circuit: see main doOnComplete below. if (streamTracker.isRecycled(conversationId)) { log.info("SSE replay doOnComplete skipped for force-recycled conversation: {}", conversationId); try { conversationService.updateStreamStatus(conversationId, "idle"); } catch (Exception e) { log.debug("recycled-skip: stream_status reset failed for {}: {}", conversationId, e.getMessage()); } ChatStreamTracker.CompletionResult cr = streamTracker.completeAndConsumeIfLast(conversationId); if (cr.allDone()) { completeEmitterQuietly(emitter, approvalEmitterDone); } return; } // Replay can re-trigger an approval (the approved tool // call may chain into another guarded tool). Derive status the same // way as the normal stream so awaiting_approval doesn't get masked // as completed. boolean replayWasStopped = streamTracker.isStopRequested(conversationId); ChatStreamTracker.InterruptType replayInterrupt = streamTracker.getInterruptType(conversationId); boolean replayIsError = accumulator.getContent() != null && accumulator.getContent().startsWith("[错误] "); String persistStatus = derivePersistStatus( accumulator.isAwaitingApproval(), replayIsError, replayWasStopped, replayInterrupt); try { MessageEntity savedAssistant = null; List parts = accumulator.toAssistantParts(); String text = accumulator.getContent(); if (!text.isBlank() || !parts.isEmpty()) { savedAssistant = conversationService.saveMessage(conversationId, "assistant", text, parts, persistStatus, accumulator.getPromptTokens(), accumulator.getCompletionTokens(), accumulator.getRuntimeModelName(), accumulator.getRuntimeProviderId(), accumulator.toMetadataJson()); // includes toolCalls metadata } else if (replayWasStopped) { boolean replayIsFollowup = replayInterrupt == ChatStreamTracker.InterruptType.USER_INTERRUPT_WITH_FOLLOWUP; savedAssistant = conversationService.saveMessage(conversationId, "assistant", replayIsFollowup ? "[已中断]" : "[已停止生成]", null, persistStatus); } else { savedAssistant = saveEmptyAssistantPlaceholder( conversationId, persistStatus, accumulator, "SSE replay doOnComplete"); } broadcastEvent(conversationId, "message_complete", Map.of( "status", persistStatus, "hasThinking", !accumulator.getThinking().isBlank(), "hasContent", !text.isBlank() )); int msgCount = conversationService.getMessageCount(conversationId); broadcastEvent(conversationId, "done", buildDonePayload( conversationId, persistStatus, savedAssistant, 0, 0, isAssistantPersisted(savedAssistant), msgCount)); } catch (Exception e) { log.warn("SSE replay complete error: {}", e.getMessage()); } finally { ChatStreamTracker.CompletionResult cr = streamTracker.completeAndConsumeIfLast(conversationId); if (cr.allDone()) { if (cr.queuedInput() != null) { startQueuedMessage(conversationId, emitter, approvalEmitterDone, cr.queuedInput(), username); } else { conversationService.updateStreamStatus(conversationId, "idle"); completeEmitterQuietly(emitter, approvalEmitterDone); } } } }) .doOnError(e -> { if (!finalized.compareAndSet(false, true)) return; // Force-recycle short-circuit: see main doOnComplete below. if (streamTracker.isRecycled(conversationId)) { log.info("SSE replay doOnError skipped for force-recycled conversation: {}, cause={}", conversationId, e.getMessage()); try { conversationService.updateStreamStatus(conversationId, "idle"); } catch (Exception ex) { log.debug("recycled-skip: stream_status reset failed for {}: {}", conversationId, ex.getMessage()); } ChatStreamTracker.CompletionResult cr = streamTracker.completeAndConsumeIfLast(conversationId); if (cr.allDone()) { completeEmitterQuietly(emitter, approvalEmitterDone); } return; } boolean isUserStop = e instanceof java.util.concurrent.CancellationException || (e.getCause() instanceof java.util.concurrent.CancellationException); ChatStreamTracker.InterruptType replayInterruptType = streamTracker.getInterruptType(conversationId); boolean replayIsFollowup = replayInterruptType == ChatStreamTracker.InterruptType.USER_INTERRUPT_WITH_FOLLOWUP; String errStatus = !isUserStop ? "failed" : replayIsFollowup ? "interrupted" : "stopped"; if (replayIsFollowup) { log.info("SSE replay stream interrupted for follow-up: conversationId={}", conversationId); } else if (isUserStop) { log.info("SSE replay stream stopped by user: conversationId={}", conversationId); } else { log.error("SSE replay error: {}", e.getMessage()); } try { MessageEntity savedAssistant = null; List replayParts = accumulator.toAssistantParts(); String replayText = accumulator.getContent(); if (!replayText.isBlank() || !replayParts.isEmpty()) { String savedText = replayText.isBlank() && isUserStop ? (replayIsFollowup ? "[已中断]" : "[已停止生成]") : replayText; savedAssistant = conversationService.saveMessage(conversationId, "assistant", savedText, replayParts, errStatus, accumulator.getPromptTokens(), accumulator.getCompletionTokens(), accumulator.getRuntimeModelName(), accumulator.getRuntimeProviderId(), accumulator.toMetadataJson()); } else if (isUserStop) { savedAssistant = conversationService.saveMessage(conversationId, "assistant", replayIsFollowup ? "[已中断]" : "[已停止生成]", null, errStatus); } else { savedAssistant = conversationService.saveMessage(conversationId, "assistant", "[错误] " + (e.getMessage() != null ? e.getMessage() : "replay error"), null, "failed"); } if (replayIsFollowup) { broadcastEvent(conversationId, "message_complete", Map.of( "status", "interrupted", "hasThinking", !accumulator.getThinking().isBlank(), "hasContent", !replayText.isBlank() )); broadcastEvent(conversationId, "turn_interrupted", Map.of( "conversationId", conversationId, "hasQueuedMessage", streamTracker.hasQueuedMessage(conversationId) )); } else if (isUserStop) { broadcastEvent(conversationId, "message_complete", Map.of( "status", "stopped", "hasThinking", !accumulator.getThinking().isBlank(), "hasContent", !replayText.isBlank() )); int stoppedMsgCount = conversationService.getMessageCount(conversationId); broadcastEvent(conversationId, "done", buildDonePayload( conversationId, "stopped", savedAssistant, 0, 0, isAssistantPersisted(savedAssistant), stoppedMsgCount)); } else { broadcastEvent(conversationId, "error", buildErrorPayload( conversationId, e.getMessage() != null ? e.getMessage() : "replay error", savedAssistant)); } } catch (Exception ex) { log.warn("SSE replay error finalize failed: {}", ex.getMessage()); } streamTracker.clearInterruptState(conversationId); ChatStreamTracker.CompletionResult cr = streamTracker.completeAndConsumeIfLast(conversationId); if (cr.allDone()) { if (cr.queuedInput() != null) { startQueuedMessage(conversationId, emitter, approvalEmitterDone, cr.queuedInput(), username); } else { conversationService.updateStreamStatus(conversationId, "idle"); completeEmitterQuietly(emitter, approvalEmitterDone); } } }) .subscribe( chunk -> { }, err -> log.debug("SSE replay subscription terminated: {}", err.getMessage()), () -> log.debug("SSE replay subscription completed: conversationId={}", conversationId)); streamTracker.setDisposable(conversationId, disposable); streamTracker.setEmergencySaveCallback(conversationId, () -> emergencySaveAccumulator(conversationId, accumulator)); } catch (Exception e) { log.error("SSE approval replay setup error: {}", e.getMessage()); streamTracker.complete(conversationId); completeEmitterQuietly(emitter, approvalEmitterDone); } }); return emitter; } // ---- 正常请求：注册流状态并附着首个订阅者 ---- streamTracker.register(conversationId); streamTracker.bindRunMeta(conversationId, agentId, username); registerEmitterCallbacks(emitter, conversationId); streamTracker.attach(conversationId, emitter); // Per-emitter "the SSE channel is open and you should reset any // pending placeholder UI". Sent directly to the emitter rather than // broadcast so reconnecting subscribers don't see a duplicate marker // for an already-open conversation. try { sendEvent(emitter, "stream_started", Map.of( "conversationId", conversationId, "timestamp", System.currentTimeMillis() )); } catch (IOException e) { log.debug("Failed to send stream_started event for {}: {}", conversationId, e.getMessage()); } // 标记 emitter 是否已结束，防止 Flux 回调再次写入已关闭的 emitter AtomicBoolean emitterDone = new AtomicBoolean(false); sseExecutor.execute(() -> { StreamAccumulator accumulator = new StreamAccumulator(); AtomicBoolean finalized = new AtomicBoolean(false); try { conversationService.getOrCreateConversation(conversationId, agentId, username, workspaceId); List requestParts = normalizeRequestParts(request); String promptText = buildPromptText(message, requestParts); conversationService.saveMessage(conversationId, "user", message, requestParts); conversationService.updateStreamStatus(conversationId, "running"); broadcastEvent(conversationId, "session", Map.of( "conversationId", conversationId, "agentId", agentId )); broadcastEvent(conversationId, "message_start", Map.of( "role", "assistant" )); streamTracker.incrementFlux(conversationId); // RFC-063r §2.5: web entry — null channelId / no ChannelTarget; // tools that need a workspace path read it from the agent (origin // is enriched with workspaceBasePath in StateGraph buildInitialState). vip.mate.agent.context.ChatOrigin webOrigin = vip.mate.agent.context.ChatOrigin.web(conversationId, username, workspaceId, null); Disposable disposable = agentService.chatStructuredStream(agentId, promptText, conversationId, username, request.getThinkingLevel(), webOrigin) .doOnNext(delta -> { if (emitterDone.get()) return; try { accumulator.accept(delta, conversationId); } catch (Exception e) { log.warn("SSE broadcast error: {}", e.getMessage()); } }) .doOnComplete(() -> { if (!finalized.compareAndSet(false, true)) return; // Force-recycle: the recycle path already wrote a // "[已被用户中止]" placeholder (or the partial // content via emergencySave). The agent's flux may // have completed the same millisecond — skip its // save + broadcast so we don't append a duplicate // assistant row below the placeholder. Cleanup // still runs so queue draining + emitter close // happen normally. if (streamTracker.isRecycled(conversationId)) { log.info("SSE doOnComplete skipped for force-recycled conversation: {}", conversationId); streamTracker.clearInterruptState(conversationId); // Defensive: keep DB stream_status consistent with the // "this turn is over" reality even when we skip the // save. Force-recycle's controller path already wrote // 'idle' for the recycled run, so this is normally a // no-op — but if a register() ever fails to clear the // marker (e.g. a different turn snuck through), this // prevents the row leaking at 'running' across refresh. try { conversationService.updateStreamStatus(conversationId, "idle"); } catch (Exception e) { log.debug("recycled-skip: stream_status reset failed for {}: {}", conversationId, e.getMessage()); } ChatStreamTracker.CompletionResult cr = streamTracker.completeAndConsumeIfLast(conversationId); if (cr.allDone()) { completeEmitterQuietly(emitter, emitterDone); } return; } // 区分四种完成语义： // 1. 正常完成（stopRequested=false）→ completed // 2. 用户主动停止 → stopped // 3. 用户中断后续跑（interrupt-with-followup）→ interrupted // 4. LLM 客户端错误 / typed error → error // （NodeStreamingChatHelper 把 typed error 序列化为 "[错误] " // 前缀的文本作为 content_delta 注入，accumulator 不区分， // 这里靠前缀识别。打标 status='error' 后，BaseAgent // 的 history sanitization 阶段会跳过这类消息，避免下次 // prompt 被污染——DeepSeek thinking mode 缺 // reasoning_content 立即 400，Claude 不接受 assistant // prefill 也 400，二者循环复制错误。） boolean wasStopped = streamTracker.isStopRequested(conversationId); ChatStreamTracker.InterruptType interruptType = streamTracker.getInterruptType(conversationId); boolean isInterruptFollowup = interruptType == ChatStreamTracker.InterruptType.USER_INTERRUPT_WITH_FOLLOWUP; boolean isError = accumulator.getContent() != null && accumulator.getContent().startsWith("[错误] "); String persistStatus = derivePersistStatus( accumulator.isAwaitingApproval(), isError, wasStopped, interruptType); try { MessageEntity savedAssistant = null; List assistantParts = accumulator.toAssistantParts(); String assistantText = accumulator.getContent(); if (!assistantText.isBlank() || !assistantParts.isEmpty()) { String savedText = assistantText.isBlank() && wasStopped ? (isInterruptFollowup ? "[已中断]" : "[已停止生成]") : assistantText; savedAssistant = conversationService.saveMessage(conversationId, "assistant", savedText, assistantParts, persistStatus, accumulator.getPromptTokens(), accumulator.getCompletionTokens(), accumulator.getRuntimeModelName(), accumulator.getRuntimeProviderId(), accumulator.toMetadataJson()); } else if (wasStopped) { savedAssistant = conversationService.saveMessage(conversationId, "assistant", isInterruptFollowup ? "[已中断]" : "[已停止生成]", null, persistStatus); } else { savedAssistant = saveEmptyAssistantPlaceholder( conversationId, persistStatus, accumulator, "SSE doOnComplete"); } // 发布对话完成事件（仅正常完成时；停止/中断/错误均不触发记忆提取） // RFC-049 follow-up: also skip on isError — error turns persist // garbage like "[错误] Bad request..." as the assistant reply, // which would pollute the memory extraction pipeline if propagated. if (!wasStopped && !isError) { completionPublisher.publish(agentId, conversationId, message, assistantText, "web"); } if (isInterruptFollowup) { broadcastEvent(conversationId, "message_complete", Map.of( "status", "interrupted", "hasThinking", !accumulator.getThinking().isBlank(), "hasContent", !assistantText.isBlank() )); broadcastEvent(conversationId, "turn_interrupted", Map.of( "conversationId", conversationId, "hasQueuedMessage", streamTracker.hasQueuedMessage(conversationId) )); } else { broadcastEvent(conversationId, "message_complete", Map.of( "status", persistStatus, "hasThinking", !accumulator.getThinking().isBlank(), "hasContent", !assistantText.isBlank() )); int msgCount = conversationService.getMessageCount(conversationId); broadcastEvent(conversationId, "done", buildDonePayload( conversationId, persistStatus, savedAssistant, accumulator.getPromptTokens(), accumulator.getCompletionTokens(), isAssistantPersisted(savedAssistant), msgCount)); } } catch (Exception e) { log.warn("SSE complete error: {}", e.getMessage()); } finally { streamTracker.clearInterruptState(conversationId); ChatStreamTracker.CompletionResult cr = streamTracker.completeAndConsumeIfLast(conversationId); if (cr.allDone()) { // RFC follow-up (2026-04-27): the previous guard // cr.queuedInput() != null && (isInterruptFollowup || !wasStopped) // dropped legitimate queued messages when the user stopped // the running turn and then sent a new message via the // enqueue path (not the interrupt-with-followup path) — // wasStopped=true + isInterruptFollowup=false made the // guard false, the consumed queuedInput was discarded, // and the user's new message vanished. The other 4 sites // in this controller already use the simpler "if queued, // run it" condition; align with them. If the user // genuinely doesn't want continuation, no message would // have been in messageQueue to begin with. if (cr.queuedInput() != null) { startQueuedMessage(conversationId, emitter, emitterDone, cr.queuedInput(), username); } else { conversationService.updateStreamStatus(conversationId, "idle"); // 延迟关闭 emitter，确保最后的事件都已发送 sseExecutor.execute(() -> { try { Thread.sleep(100); } catch (InterruptedException ignored) {} completeEmitterQuietly(emitter, emitterDone); }); } } else { log.info("Original stream completed but replay still active, " + "keeping SSE emitter alive: conversationId={}", conversationId); } } }) .doOnCancel(() -> { boolean wasFirst = finalized.compareAndSet(false, true); log.info("SSE doOnCancel fired: conversationId={}, wasFirst={}", conversationId, wasFirst); if (!wasFirst) return; // Force-recycle short-circuit: see doOnComplete above. if (streamTracker.isRecycled(conversationId)) { log.info("SSE doOnCancel skipped for force-recycled conversation: {}", conversationId); streamTracker.clearInterruptState(conversationId); try { conversationService.updateStreamStatus(conversationId, "idle"); } catch (Exception e) { log.debug("recycled-skip: stream_status reset failed for {}: {}", conversationId, e.getMessage()); } ChatStreamTracker.CompletionResult cr = streamTracker.completeAndConsumeIfLast(conversationId); if (cr.allDone()) { completeEmitterQuietly(emitter, emitterDone); } return; } // 区分用户主动停止和 interrupt-with-followup ChatStreamTracker.InterruptType interruptType = streamTracker.getInterruptType(conversationId); boolean isInterruptFollowup = interruptType == ChatStreamTracker.InterruptType.USER_INTERRUPT_WITH_FOLLOWUP; String status = isInterruptFollowup ? "interrupted" : "stopped"; log.info("SSE stream cancelled ({}): conversationId={}", status, conversationId); try { MessageEntity savedAssistant = null; List assistantParts = accumulator.toAssistantParts(); String assistantText = accumulator.getContent(); if (!assistantText.isBlank() || !assistantParts.isEmpty()) { String savedText = assistantText.isBlank() ? (isInterruptFollowup ? "[已中断]" : "[已停止生成]") : assistantText; savedAssistant = conversationService.saveMessage(conversationId, "assistant", savedText, assistantParts, status, accumulator.getPromptTokens(), accumulator.getCompletionTokens(), accumulator.getRuntimeModelName(), accumulator.getRuntimeProviderId(), accumulator.toMetadataJson()); } else { savedAssistant = conversationService.saveMessage(conversationId, "assistant", isInterruptFollowup ? "[已中断]" : "[已停止生成]", null, status); } if (isInterruptFollowup) { broadcastEvent(conversationId, "message_complete", Map.of( "status", "interrupted", "hasThinking", !accumulator.getThinking().isBlank(), "hasContent", !assistantText.isBlank() )); broadcastEvent(conversationId, "turn_interrupted", Map.of( "conversationId", conversationId, "hasQueuedMessage", streamTracker.hasQueuedMessage(conversationId) )); } else { broadcastEvent(conversationId, "message_complete", Map.of( "status", "stopped", "hasThinking", !accumulator.getThinking().isBlank(), "hasContent", !assistantText.isBlank() )); int stoppedMsgCount = conversationService.getMessageCount(conversationId); broadcastEvent(conversationId, "done", buildDonePayload( conversationId, "stopped", savedAssistant, 0, 0, isAssistantPersisted(savedAssistant), stoppedMsgCount)); } } catch (Exception e) { log.warn("SSE stop finalize error: {}", e.getMessage()); } finally { streamTracker.clearInterruptState(conversationId); ChatStreamTracker.CompletionResult cr = streamTracker.completeAndConsumeIfLast(conversationId); if (cr.allDone()) { if (cr.queuedInput() != null) { // 无论中断类型，都消费排队消息（修复 Disposable 不可用时队列被丢弃的 bug） startQueuedMessage(conversationId, emitter, emitterDone, cr.queuedInput(), username); } else { conversationService.updateStreamStatus(conversationId, "idle"); completeEmitterQuietly(emitter, emitterDone); } } } }) .doOnError(e -> { boolean wasFirst = finalized.compareAndSet(false, true); if (!wasFirst) { log.info("SSE doOnError skipped (finalized by doOnCancel): conversationId={}", conversationId); return; } // Force-recycle short-circuit: see doOnComplete above. if (streamTracker.isRecycled(conversationId)) { log.info("SSE doOnError skipped for force-recycled conversation: {}, cause={}", conversationId, e.getMessage()); streamTracker.clearInterruptState(conversationId); try { conversationService.updateStreamStatus(conversationId, "idle"); } catch (Exception ex) { log.debug("recycled-skip: stream_status reset failed for {}: {}", conversationId, ex.getMessage()); } ChatStreamTracker.CompletionResult cr = streamTracker.completeAndConsumeIfLast(conversationId); if (cr.allDone()) { completeEmitterQuietly(emitter, emitterDone); } return; } // CancellationException = 用户主动停止或中断续跑 boolean isUserStop = e instanceof java.util.concurrent.CancellationException || (e.getCause() instanceof java.util.concurrent.CancellationException); ChatStreamTracker.InterruptType interruptType = streamTracker.getInterruptType(conversationId); boolean isInterruptFollowup = interruptType == ChatStreamTracker.InterruptType.USER_INTERRUPT_WITH_FOLLOWUP; // 三态：interrupted > stopped > failed String status = !isUserStop ? "failed" : isInterruptFollowup ? "interrupted" : "stopped"; if (isInterruptFollowup) { log.info("SSE stream interrupted for follow-up (CancellationException): conversationId={}", conversationId); } else if (isUserStop) { log.info("SSE stream stopped by user (CancellationException): conversationId={}", conversationId); } else if (isClientDisconnect(e)) { log.warn("SSE client disconnected: conversationId={}, cause={}", conversationId, e.getMessage()); } else { log.error("SSE stream error: conversationId={}, cause={}", conversationId, e.getMessage()); } try { List assistantParts = accumulator.toAssistantParts(); String assistantText = accumulator.getContent(); log.info("SSE doOnError saving: conversationId={}, status={}, textLen={}, partsCount={}", conversationId, status, assistantText.length(), assistantParts.size()); String errorMsg = e.getMessage() != null ? e.getMessage() : "unknown error"; MessageEntity savedAssistant = null; if (!assistantText.isBlank() || !assistantParts.isEmpty()) { String savedText = assistantText.isBlank() && isUserStop ? (isInterruptFollowup ? "[已中断]" : "[已停止生成]") : assistantText; savedAssistant = conversationService.saveMessage(conversationId, "assistant", savedText, assistantParts, status, accumulator.getPromptTokens(), accumulator.getCompletionTokens(), accumulator.getRuntimeModelName(), accumulator.getRuntimeProviderId(), accumulator.toMetadataJson()); } else if (isUserStop) { savedAssistant = conversationService.saveMessage(conversationId, "assistant", isInterruptFollowup ? "[已中断]" : "[已停止生成]", null, status); } else { savedAssistant = conversationService.saveMessage(conversationId, "assistant", "[错误] " + errorMsg, null, "failed"); } if (isInterruptFollowup) { broadcastEvent(conversationId, "message_complete", Map.of( "status", "interrupted", "hasThinking", !accumulator.getThinking().isBlank(), "hasContent", !assistantText.isBlank() )); broadcastEvent(conversationId, "turn_interrupted", Map.of( "conversationId", conversationId, "hasQueuedMessage", streamTracker.hasQueuedMessage(conversationId) )); } else if (isUserStop) { broadcastEvent(conversationId, "message_complete", Map.of( "status", "stopped", "hasThinking", !accumulator.getThinking().isBlank(), "hasContent", !assistantText.isBlank() )); int stoppedMsgCount = conversationService.getMessageCount(conversationId); broadcastEvent(conversationId, "done", buildDonePayload( conversationId, "stopped", savedAssistant, 0, 0, isAssistantPersisted(savedAssistant), stoppedMsgCount)); } else { broadcastEvent(conversationId, "error", buildErrorPayload(conversationId, errorMsg, savedAssistant)); } } catch (Exception ioException) { log.error("SSE doOnError save/broadcast failed: conversationId={}, error={}", conversationId, ioException.getMessage(), ioException); } streamTracker.clearInterruptState(conversationId); ChatStreamTracker.CompletionResult cr = streamTracker.completeAndConsumeIfLast(conversationId); log.info("SSE doOnError cleanup: conversationId={}, allDone={}, isInterruptFollowup={}, hasQueued={}", conversationId, cr.allDone(), isInterruptFollowup, cr.queuedInput() != null); if (cr.allDone()) { // RFC follow-up (2026-04-27): the previous guard // cr.queuedInput()!=null && !(isUserStop && !isInterruptFollowup) // tried to suppress continuation when the user "explicitly // stopped" without an interrupt-with-followup. But the // frontend's enqueue path doesn't set interruptType — it // just calls requestStop + offers to messageQueue. From the // server's POV that's "isUserStop=true, isInterruptFollowup= // false, queue has content", which the guard mis-classified // as "abort" and silently dropped the user's freshly-typed // follow-up. Whoever puts a message in messageQueue means it // — just run it. Aligns with doOnComplete and the 4 other // queue-launch sites in this controller. if (cr.queuedInput() != null) { startQueuedMessage(conversationId, emitter, emitterDone, cr.queuedInput(), username); } else { conversationService.updateStreamStatus(conversationId, "idle"); completeEmitterQuietly(emitter, emitterDone); } } }) .subscribe( chunk -> { }, error -> log.debug("SSE stream subscription terminated with error: {}", error.getMessage()), () -> log.debug("SSE stream subscription completed: conversationId={}", conversationId)); // 将 Disposable 注册到 StreamTracker，以便 stop 端点可以取消它 streamTracker.setDisposable(conversationId, disposable); // JVM 关闭时优雅落盘：避免 mvn spring-boot:run 重启 / SIGTERM 把 // 进行中 turn 的 assistant 消息丢失（doOnError 来不及在 Hikari 关闭前执行） streamTracker.setEmergencySaveCallback(conversationId, () -> emergencySaveAccumulator(conversationId, accumulator)); } catch (Exception e) { log.error("SSE setup error: {}", e.getMessage()); try { broadcastEvent(conversationId, "error", Map.of("message", e.getMessage() != null ? e.getMessage() : "unknown error")); } catch (Exception ioException) { log.warn("SSE setup failure event broadcast error: {}", ioException.getMessage()); } streamTracker.complete(conversationId); conversationService.updateStreamStatus(conversationId, "idle"); completeEmitterQuietly(emitter, emitterDone); } }); return emitter; } /** * 停止指定会话的流式生成。 * 取消 Flux 订阅（底层 HTTP 连接也会随之关闭），已生成的部分内容以 stopped 状态入库。 *

* Stop 同时清理所有未 resolve 的 pending approval：当 LLM 在一个 turn 里连发了 * 多个需要审批的工具调用、用户在中间 Stop 时，这些 pending 会一直留在 in-memory * pendingMap 里。下次刷新页面时 frontend 的 hydrate 链路（`getPendingApprovals` API * + 消息 metadata 里的 `pendingApproval` 字段）会反复弹出"允许 xxx 执行？"banner。 * Stop 端点现在 deny 所有 pending、同步 update 受影响 message 的 metadata，并广播 * tool_approval_resolved 让前端实时清理 UI。 */ @Operation(summary = "停止流式生成") @PostMapping("/{conversationId}/stop") public R> stopStream(@PathVariable String conversationId, Authentication auth) { String username = auth != null ? auth.getName() : "anonymous"; // 权限校验：已认证用户需验证会话归属，匿名用户（permitAll）直接放行 if (auth != null && !conversationService.isConversationOwner(conversationId, username)) { return R.fail("无权操作该会话"); } boolean stopped = streamTracker.requestStop(conversationId); // Sweep ghost approvals — workflow.denyAllByConversation owns DB + metadata + memory // atomically; we only need to broadcast SSE events on the resulting outcomes. List denied = approvalService.denyAllByConversation(conversationId, username); int messagesRewritten = denied.stream().mapToInt(ResolveOutcome::messagesRewritten).sum(); for (ResolveOutcome o : denied) { broadcastEvent(conversationId, "tool_approval_resolved", Map.of( "pendingId", o.pendingId(), "decision", "denied", "toolName", o.toolName() != null ? o.toolName() : "", "timestamp", System.currentTimeMillis() )); } log.info("Stop requested: conversationId={}, user={}, stopped={}, ghostPendingsCleared={}, messagesRewritten={}", conversationId, username, stopped, denied.size(), messagesRewritten); return R.ok(Map.of( "stopped", stopped, "ghostPendingsCleared", denied.size(), "messagesRewritten", messagesRewritten )); } /** * 在执行中追加一条后续消息：仅入队，等当前 turn 自然结束后再启动。 *

* 对齐 Claude Code 行为：流式输出过程中收到新输入不会强制 dispose 当前 LLM 调用， * 而是仅入队。当前 turn 跑到 doOnComplete/doOnError 后由 startQueuedMessage 接管。 *

* 旧行为（dispose 当前 disposable + 立即重启）会导致部分 LLM 输出被丢弃 + token 浪费， * 已废弃。如果未来需要"立即打断"语义，应该走 /stop（用户主动取消）+ 重新发起新消息的路径。 */ @Operation(summary = "排队后续消息（不打断当前流）") @PostMapping("/{conversationId}/interrupt") public R> interruptStream( @PathVariable String conversationId, @RequestBody InterruptRequest request, Authentication auth) { String username = auth != null ? auth.getName() : "anonymous"; if (auth != null && !conversationService.isConversationOwner(conversationId, username)) { return R.fail("无权操作该会话"); } if (!streamTracker.isRunning(conversationId)) { return R.ok(Map.of("interrupted", false, "reason", "no_active_stream")); } String message = request.getMessage(); Long agentId = request.getAgentId(); List contentParts = request.getContentParts(); // 判断当前阶段（仅用于 reason 字段，行为对所有阶段一致：仅入队） boolean isAwaitingApproval = approvalService.findPendingByConversation(conversationId) != null; // 仅入队、不 dispose。延迟持久化到 startQueuedMessage（让 Asst-N 先在 doOnComplete 落库， // 否则 listMessages ORDER BY create_time ASC 会把 Q(N+1) 排到 Asst-N 前面） boolean queued = streamTracker.enqueueMessage(conversationId, message, agentId, false, contentParts); log.info("Enqueued follow-up message during running turn: conversationId={}, user={}, queueSize={}, awaitingApproval={}", conversationId, username, streamTracker.getQueueSize(conversationId), isAwaitingApproval); return R.ok(Map.of( "interrupted", false, "queued", queued, "queueSize", streamTracker.getQueueSize(conversationId), "reason", isAwaitingApproval ? "awaiting_approval" : "queued" )); } @lombok.Data public static class InterruptRequest { private String message; private Long agentId; /** 结构化内容片段（含图片等附件），排队消息带附件时由前端传入 */ private List contentParts; } /** * 同步对话 */ @Operation(summary = "同步对话") @PostMapping public R chat( @RequestParam Long agentId, @RequestBody ChatRequest request, @RequestHeader(value = "X-Workspace-Id", required = false) Long workspaceId, Authentication auth) { String username = auth != null ? auth.getName() : null; if (username == null) { return R.fail("未登录，请先登录"); } conversationService.getOrCreateConversation(request.getConversationId(), agentId, username, workspaceId); conversationService.saveMessage(request.getConversationId(), "user", request.getMessage(), request.getContentParts()); String promptText = buildPromptText(request.getMessage(), request.getContentParts()); String response = agentService.chat(agentId, promptText, request.getConversationId()); conversationService.saveMessage(request.getConversationId(), "assistant", response); completionPublisher.publish(agentId, request.getConversationId(), request.getMessage(), response, "web"); return R.ok(response); } @Operation(summary = "上传聊天附件") @PostMapping(value = "/upload", consumes = MediaType.MULTIPART_FORM_DATA_VALUE) public R upload( @RequestParam String conversationId, @RequestPart("file") MultipartFile file, Authentication auth) throws IOException { String username = auth != null ? auth.getName() : "anonymous"; // 校验会话归属（会话可能尚未创建，此时允许上传——后续 stream/chat 会创建并绑定用户） if (conversationService.conversationExists(conversationId) && !conversationService.isConversationOwner(conversationId, username)) { return R.fail("无权操作该会话"); } if (file.isEmpty()) { return R.fail("上传文件不能为空"); } String originalFilename = file.getOriginalFilename() != null ? file.getOriginalFilename() : "file"; String safeFilename = Path.of(originalFilename).getFileName().toString().replaceAll("[^a-zA-Z0-9._-]", "_"); String storedName = System.currentTimeMillis() + "_" + safeFilename; Path conversationDir = uploadRoot.resolve(conversationId); Files.createDirectories(conversationDir); Path target = conversationDir.resolve(storedName); file.transferTo(target); log.info("Chat attachment uploaded: conversationId={}, user={}, file={}", conversationId, username, target); ChatUploadResponse response = new ChatUploadResponse(); response.setConversationId(conversationId); response.setFileName(originalFilename); response.setStoredName(storedName); response.setUrl("/api/v1/chat/files/" + conversationId + "/" + storedName); // 使用相对路径，避免暴露服务端绝对路径 response.setPath(uploadRoot.resolve(conversationId).resolve(storedName).toString()); response.setSize(file.getSize()); response.setContentType(file.getContentType()); return R.ok(response); } @Operation(summary = "读取聊天附件") @GetMapping("/files/{conversationId}/{storedName:.+}") public ResponseEntity readUploadedFile( @PathVariable String conversationId, @PathVariable String storedName, Authentication auth) throws IOException { // 校验当前用户拥有该会话 String username = auth != null ? auth.getName() : "anonymous"; if (!conversationService.isConversationOwner(conversationId, username)) { return ResponseEntity.status(403).build(); } Path filePath = uploadRoot.resolve(conversationId).resolve(storedName).normalize(); if (!Files.exists(filePath) || !filePath.startsWith(uploadRoot.resolve(conversationId).normalize())) { return ResponseEntity.notFound().build(); } Resource resource = new FileSystemResource(filePath); String contentType = Files.probeContentType(filePath); // probeContentType 在部分平台不识别视频格式，通过扩展名 fallback if (contentType == null) { contentType = guessContentTypeByExtension(filePath.getFileName().toString()); } MediaType mediaType = MediaType.APPLICATION_OCTET_STREAM; if (contentType != null) { try { mediaType = MediaType.parseMediaType(contentType); } catch (Exception ignored) { } } String encodedFilename = URLEncoder.encode(filePath.getFileName().toString(), StandardCharsets.UTF_8) .replace("+", "%20"); return ResponseEntity.ok() .contentType(mediaType) .header(HttpHeaders.CONTENT_DISPOSITION, "inline; filename*=UTF-8''" + encodedFilename) .body(resource); } @lombok.Data public static class ChatRequest { private String message; private String conversationId = "default"; private List contentParts; } @lombok.Data public static class ChatUploadResponse { private String conversationId; private String fileName; private String storedName; private String url; private String path; private Long size; private String contentType; } @lombok.Data public static class ChatStreamRequest { private Long agentId; private String message; private String conversationId = "default"; private List contentParts; /** true 表示断线重连，不发送新消息，只附着到已有的流 */ private Boolean reconnect; /** * Last SSE event id the client has already processed. Only meaningful * when {@link #reconnect} is true — the server skips events with * id ≤ this value during buffer replay so the client doesn't * see them twice. 0 (or null) means "replay everything", matching * the legacy attach behavior for backwards compatibility. */ private Long lastEventId; /** 思考深度：off / low / medium / high / max，null 表示跟随 Agent 默认 */ private String thinkingLevel; } /** * 自动启动排队消息（interrupt-with-followup 或自然完成后的续跑逻辑）。 * 接受由 {@link ChatStreamTracker#completeAndConsumeIfLast} 预先消费的 QueuedInput 快照。 * 快照已脱离 RunState 生命周期，不受后续 complete/register 影响。 * 支持链式续跑：queued stream 自身完成时也通过 completeAndConsumeIfLast 检查并递归调用。 */ private void startQueuedMessage(String conversationId, SseEmitter emitter, AtomicBoolean emitterDone, ChatStreamTracker.QueuedInput preConsumedInput, String requesterId) { if (preConsumedInput == null) { conversationService.updateStreamStatus(conversationId, "idle"); completeEmitterQuietly(emitter, emitterDone); return; } // Rate Limit 防护：如果上一轮以 rate limit 错误结束，不立即续跑排队消息（必然再次 429）。 // 改为持久化用户消息 + 通知前端"稍后重试"，避免连锁 429 浪费配额。 String lastMessage = conversationService.getLastMessage(conversationId); if (lastMessage != null && (lastMessage.contains("频率过高") || lastMessage.contains("rate_limit") || lastMessage.contains("429") || lastMessage.contains("速率限制"))) { log.warn("Skipping queued message after rate limit error: conversationId={}, lastMessage={}", conversationId, lastMessage.substring(0, Math.min(50, lastMessage.length()))); // 持久化用户消息不丢失 if (preConsumedInput.message() != null && !preConsumedInput.message().isBlank() && !preConsumedInput.persisted()) { conversationService.saveMessage(conversationId, "user", preConsumedInput.message()); } broadcastEvent(conversationId, "warning", Map.of( "message", "上一轮请求触发了频率限制，排队消息已保存，请稍后重新发送")); broadcastEvent(conversationId, "done", Map.of("status", "rate_limited")); conversationService.updateStreamStatus(conversationId, "idle"); completeEmitterQuietly(emitter, emitterDone); return; } String queuedMessage = preConsumedInput.message(); Long agentId = preConsumedInput.agentId() != null ? preConsumedInput.agentId() : 1L; log.info("Starting queued message: conversationId={}, agentId={}, message={}", conversationId, agentId, queuedMessage.substring(0, Math.min(30, queuedMessage.length()))); // 持久化排队的用户消息（含 contentParts；幂等：如果 /interrupt 已提前持久化则跳过）。 // 这里持久化是为了确保 user 消息在 assistant 消息（doOnError/doOnCancel 已写入）之后落库， // 让 listMessages ORDER BY create_time ASC 后顺序正确：Q1 → Asst1 → Q2 → Asst2。 if (queuedMessage != null && !queuedMessage.isBlank() && !preConsumedInput.persisted()) { conversationService.saveMessage(conversationId, "user", queuedMessage, preConsumedInput.contentParts(), "queued"); } // 广播 queued_input_started 事件 broadcastEvent(conversationId, "queued_input_started", Map.of( "conversationId", conversationId, "message", queuedMessage )); // 重新注册流状态 streamTracker.register(conversationId); streamTracker.attach(conversationId, emitter); // 启动新的流（复用现有 sseExecutor.execute 的逻辑模式） StreamAccumulator accumulator = new StreamAccumulator(); AtomicBoolean finalized = new AtomicBoolean(false); broadcastEvent(conversationId, "message_start", Map.of("role", "assistant")); streamTracker.incrementFlux(conversationId); // RFC-063r §2.5: queued messages land in the same conversation; carry // a web-origin ChatOrigin so any cron job created during the queued // turn keeps a consistent (null-channel) binding. vip.mate.agent.context.ChatOrigin queuedOrigin = vip.mate.agent.context.ChatOrigin.web(conversationId, requesterId, null, null); Disposable disposable = agentService.chatStructuredStream(agentId, queuedMessage, conversationId, requesterId, null, queuedOrigin) .doOnNext(delta -> { if (emitterDone.get()) return; try { accumulator.accept(delta, conversationId); } catch (Exception e) { log.warn("SSE queued broadcast error: {}", e.getMessage()); } }) .doOnComplete(() -> { if (!finalized.compareAndSet(false, true)) return; // RFC-067 §4.6: queued stream can hit a tool_approval_requested event // mid-flight. Derive status via the shared helper so awaiting_approval // is not silently downgraded to completed (which would prematurely fire // expirePendingApprovals on the frontend and ghost-clear the banner). boolean queuedWasStopped = streamTracker.isStopRequested(conversationId); ChatStreamTracker.InterruptType queuedInterrupt = streamTracker.getInterruptType(conversationId); boolean queuedIsError = accumulator.getContent() != null && accumulator.getContent().startsWith("[错误] "); String persistStatus = derivePersistStatus( accumulator.isAwaitingApproval(), queuedIsError, queuedWasStopped, queuedInterrupt); try { MessageEntity savedAssistant = null; List parts = accumulator.toAssistantParts(); String text = accumulator.getContent(); if (!text.isBlank() || !parts.isEmpty()) { savedAssistant = conversationService.saveMessage(conversationId, "assistant", text, parts, persistStatus, accumulator.getPromptTokens(), accumulator.getCompletionTokens(), accumulator.getRuntimeModelName(), accumulator.getRuntimeProviderId(), accumulator.toMetadataJson()); } else if (queuedWasStopped) { boolean queuedIsFollowup = queuedInterrupt == ChatStreamTracker.InterruptType.USER_INTERRUPT_WITH_FOLLOWUP; savedAssistant = conversationService.saveMessage(conversationId, "assistant", queuedIsFollowup ? "[已中断]" : "[已停止生成]", null, persistStatus); } else { savedAssistant = saveEmptyAssistantPlaceholder( conversationId, persistStatus, accumulator, "SSE queued doOnComplete"); } broadcastEvent(conversationId, "message_complete", Map.of( "status", persistStatus, "hasThinking", !accumulator.getThinking().isBlank(), "hasContent", !text.isBlank() )); broadcastEvent(conversationId, "done", buildDonePayload( conversationId, persistStatus, savedAssistant, accumulator.getPromptTokens(), accumulator.getCompletionTokens(), isAssistantPersisted(savedAssistant), conversationService.getMessageCount(conversationId))); } catch (Exception e) { log.warn("SSE queued complete error: {}", e.getMessage()); } finally { ChatStreamTracker.CompletionResult cr = streamTracker.completeAndConsumeIfLast(conversationId); if (cr.allDone()) { if (cr.queuedInput() != null) { // 链式续跑：queued stream 期间又排了新消息 startQueuedMessage(conversationId, emitter, emitterDone, cr.queuedInput(), requesterId); } else { conversationService.updateStreamStatus(conversationId, "idle"); sseExecutor.execute(() -> { try { Thread.sleep(100); } catch (InterruptedException ignored) {} completeEmitterQuietly(emitter, emitterDone); }); } } } }) .doOnError(e -> { if (!finalized.compareAndSet(false, true)) return; log.error("SSE queued stream error: conversationId={}, cause={}", conversationId, e.getMessage()); // 持久化已累积的 assistant 消息（修复：原逻辑未保存导致回答丢失） try { MessageEntity savedAssistant = null; List parts = accumulator.toAssistantParts(); String text = accumulator.getContent(); if (!text.isBlank() || !parts.isEmpty()) { savedAssistant = conversationService.saveMessage(conversationId, "assistant", text, parts, "failed", accumulator.getPromptTokens(), accumulator.getCompletionTokens(), accumulator.getRuntimeModelName(), accumulator.getRuntimeProviderId(), accumulator.toMetadataJson()); } else { String errorMsg = e.getMessage() != null ? e.getMessage() : "queued stream error"; savedAssistant = conversationService.saveMessage(conversationId, "assistant", "[错误] " + errorMsg, null, "failed"); } broadcastEvent(conversationId, "error", buildErrorPayload( conversationId, e.getMessage() != null ? e.getMessage() : "queued stream error", savedAssistant)); } catch (Exception saveEx) { log.error("SSE queued doOnError save failed: {}", saveEx.getMessage()); } ChatStreamTracker.CompletionResult cr = streamTracker.completeAndConsumeIfLast(conversationId); if (cr.allDone()) { if (cr.queuedInput() != null) { startQueuedMessage(conversationId, emitter, emitterDone, cr.queuedInput(), requesterId); } else { conversationService.updateStreamStatus(conversationId, "idle"); completeEmitterQuietly(emitter, emitterDone); } } }) .subscribe(); streamTracker.setDisposable(conversationId, disposable); streamTracker.setEmergencySaveCallback(conversationId, () -> emergencySaveAccumulator(conversationId, accumulator)); } private void sendEvent(SseEmitter emitter, String name, Object data) throws IOException { String payload; try { payload = objectMapper.writeValueAsString(data); } catch (Exception e) { payload = "{\"message\":\"serialization_error\"}"; } emitter.send(SseEmitter.event().name(name).data(payload)); } private void broadcastEvent(String conversationId, String name, Object data) { String payload; try { payload = objectMapper.writeValueAsString(data); } catch (Exception e) { payload = "{\"message\":\"serialization_error\"}"; } streamTracker.broadcast(conversationId, name, payload); } /** * Derive the persistence status for an assistant message at stream finalization * (RFC-067 §4.6). Five-way state machine: *

* Package-private so {@link vip.mate.channel.web.ChatControllerPersistStatusTest} * can exercise the truth table directly without spinning up the controller. */ static String derivePersistStatus(boolean isAwaitingApproval, boolean isError, boolean wasStopped, ChatStreamTracker.InterruptType interruptType) { if (isAwaitingApproval) return "awaiting_approval"; if (isError) return "error"; if (!wasStopped) return "completed"; return interruptType == ChatStreamTracker.InterruptType.USER_INTERRUPT_WITH_FOLLOWUP ? "interrupted" : "stopped"; } static String emptyAssistantPlaceholder(String status) { if ("awaiting_approval".equals(status)) return "[等待审批]"; return "[本次没有输出]"; } static boolean isAssistantPersisted(MessageEntity savedAssistant) { return savedAssistant != null; } private MessageEntity saveEmptyAssistantPlaceholder(String conversationId, String status, StreamAccumulator accumulator, String source) { log.warn("{} with empty accumulator: conversationId={}, status={}, finishReason={}, phase={}, hasSegments={}", source, conversationId, status, accumulator.getFinishReason(), accumulator.getCurrentPhase(), !accumulator.segmentsEmpty()); return conversationService.saveMessage(conversationId, "assistant", emptyAssistantPlaceholder(status), null, status, accumulator.getPromptTokens(), accumulator.getCompletionTokens(), accumulator.getRuntimeModelName(), accumulator.getRuntimeProviderId(), accumulator.toMetadataJson()); } private Map buildDonePayload(String conversationId, String status, MessageEntity savedAssistant, int promptTokens, int completionTokens, boolean persisted, Integer messageCount) { Map payload = new LinkedHashMap<>(); if (conversationId != null && !conversationId.isBlank()) payload.put("conversationId", conversationId); payload.put("status", status); if (savedAssistant != null && savedAssistant.getId() != null) { payload.put("assistantMessageId", savedAssistant.getId()); // Surface runtime model attribution so the chat bubble can show // which model produced this reply without waiting for a history reload. if (savedAssistant.getRuntimeModel() != null && !savedAssistant.getRuntimeModel().isBlank()) { payload.put("runtimeModel", savedAssistant.getRuntimeModel()); } if (savedAssistant.getRuntimeProvider() != null && !savedAssistant.getRuntimeProvider().isBlank()) { payload.put("runtimeProvider", savedAssistant.getRuntimeProvider()); } } if (promptTokens > 0) payload.put("promptTokens", promptTokens); if (completionTokens > 0) payload.put("completionTokens", completionTokens); payload.put("persisted", persisted); if (messageCount != null) payload.put("messageCount", messageCount); return payload; } private Map buildErrorPayload(String conversationId, String message, MessageEntity savedAssistant) { Map payload = new LinkedHashMap<>(); payload.put("message", message); if (conversationId != null && !conversationId.isBlank()) payload.put("conversationId", conversationId); if (savedAssistant != null && savedAssistant.getId() != null) { payload.put("assistantMessageId", savedAssistant.getId()); } return payload; } /** * Snapshot the accumulator and persist it as an assistant message during JVM shutdown. * Invoked from {@link ChatStreamTracker#onShutdown()} so any in-flight turn doesn't * lose its already-streamed content + tool calls when the process exits. *

* Status routing (RFC-067 §4.6): *

    *
  • If the accumulator was awaiting approval at shutdown, the message keeps * {@code awaiting_approval}. After restart, {@code recoverFromDb} re-registers * the pending in memory and the existing approval banner remains coherent * with both DB and metadata. Falling back to {@code interrupted_shutdown} * here would orphan the message metadata (UI would render "interrupted" * while the approval is still recoverable).
    • *
  • Otherwise {@code interrupted_shutdown} as before.
    • *
* Idempotent w.r.t. the normal doOnComplete/doOnError save: if those paths already * persisted the message, this writes a second row, which is rare in practice * (race window is sub-second between dispose and save) and acceptable. Skipping * save when nothing to save avoids empty rows. */ private void emergencySaveAccumulator(String conversationId, StreamAccumulator accumulator) { try { String text = accumulator.getContent(); List parts = accumulator.toAssistantParts(); if (text.isBlank() && parts.isEmpty()) { log.warn("[ChatController] Emergency save skipped (empty accumulator): conversationId={}, finishReason={}, phase={}, hasSegments={}", conversationId, accumulator.getFinishReason(), accumulator.getCurrentPhase(), !accumulator.segmentsEmpty()); return; } boolean awaitingApproval = accumulator.isAwaitingApproval(); String status = awaitingApproval ? "awaiting_approval" : "interrupted_shutdown"; String savedText = text.isBlank() ? (awaitingApproval ? "[等待审批 — 服务重启]" : "[已中断 — 服务重启]") : text; conversationService.saveMessage(conversationId, "assistant", savedText, parts, status, accumulator.getPromptTokens(), accumulator.getCompletionTokens(), accumulator.getRuntimeModelName(), accumulator.getRuntimeProviderId(), accumulator.toMetadataJson()); log.info("[ChatController] Emergency-saved in-flight assistant message: " + "conversationId={}, status={}, textLen={}, partsCount={}", conversationId, status, text.length(), parts.size()); } catch (Exception e) { log.error("[ChatController] Emergency save failed for {}: {}", conversationId, e.getMessage(), e); } } private List normalizeRequestParts(ChatStreamRequest request) { if (request.getContentParts() != null && !request.getContentParts().isEmpty()) { return request.getContentParts(); } if (request.getMessage() == null || request.getMessage().isBlank()) { return List.of(); } MessageContentPart textPart = new MessageContentPart(); textPart.setType("text"); textPart.setText(request.getMessage()); return List.of(textPart); } private String buildPromptText(String message, List parts) { if (parts == null || parts.isEmpty()) { return message != null ? message : ""; } StringBuilder builder = new StringBuilder(); for (MessageContentPart part : parts) { if (part == null || part.getType() == null) { continue; } switch (part.getType()) { case "text", "thinking" -> appendPromptLine(builder, part.getText()); case "file" -> appendPromptLine(builder, "附件: " + safe(part.getFileName()) + " (" + safe(part.getPath()) + ")"); case "image" -> appendPromptLine(builder, "图片附件: " + safe(part.getFileName()) + " (" + safe(part.getPath()) + ")"); case "video" -> appendPromptLine(builder, "视频附件: " + safe(part.getFileName()) + " (" + safe(part.getPath()) + ")"); default -> appendPromptLine(builder, part.getText()); } } return builder.toString().trim(); } private void appendPromptLine(StringBuilder builder, String text) { if (text == null || text.isBlank()) { return; } if (!builder.isEmpty()) { builder.append('\n'); } builder.append(text); } private String safe(String text) { return text == null ? "" : text; } private static final java.util.Map MEDIA_CONTENT_TYPES = java.util.Map.of( "mp4", "video/mp4", "webm", "video/webm", "mov", "video/quicktime", "avi", "video/x-msvideo", "mkv", "video/x-matroska", "mpeg", "video/mpeg", "mp3", "audio/mpeg", "wav", "audio/wav", "ogg", "audio/ogg" ); private static String guessContentTypeByExtension(String fileName) { if (fileName == null) return null; int dot = fileName.lastIndexOf('.'); if (dot < 0 || dot == fileName.length() - 1) return null; return MEDIA_CONTENT_TYPES.get(fileName.substring(dot + 1).toLowerCase()); } /** * 注册 SseEmitter 的完整生命周期回调 */ private void registerEmitterCallbacks(SseEmitter emitter, String conversationId) { emitter.onCompletion(() -> { log.debug("SSE emitter completed: conversationId={}", conversationId); // Detach immediately so a subsequent broadcast (heartbeat / async_task_*) // doesn't waste a send call on the zombie emitter and emit // "Removing dead subscriber ... ResponseBodyEmitter has already completed". streamTracker.detach(conversationId, emitter); }); emitter.onTimeout(() -> { log.debug("SSE emitter timeout: conversationId={}", conversationId); streamTracker.detach(conversationId, emitter); // 超时后显式 complete，防止 servlet 容器再抛 AsyncRequestTimeoutException emitter.complete(); }); emitter.onError(e -> { if (isClientDisconnect(e)) { log.debug("SSE client disconnected: conversationId={}, cause={}", conversationId, e.getMessage()); } else { log.warn("SSE emitter error: conversationId={}, cause={}", conversationId, e.getMessage()); } streamTracker.detach(conversationId, emitter); }); } /** * 安全地完成 emitter，防止重复调用和已关闭连接引发的异常 */ private void completeEmitterQuietly(SseEmitter emitter, AtomicBoolean emitterDone) { if (!emitterDone.compareAndSet(false, true)) return; try { emitter.complete(); } catch (Exception e) { log.debug("Emitter already completed: {}", e.getMessage()); } } /** * 判断异常是否为客户端断开连接（broken pipe、connection reset 等） */ private boolean isClientDisconnect(Throwable e) { if (e instanceof IOException) return true; String msg = e.getMessage(); if (msg == null) return false; String lower = msg.toLowerCase(); return lower.contains("broken pipe") || lower.contains("connection reset") || lower.contains("client abort") || lower.contains("closed"); } /** * 流式累积器 — 收集 StreamDelta 事件，持久化到 DB。 *

* 维护两份数据： *

    *
  • {@code toolCalls} — 兼容旧逻辑（执行面板等 UI 使用）
    • *
  • {@code segments} — 按事件到达顺序记录的有序时间线（前端分段渲染用）
    • *
* 两份数据从同一事件流构建，保证一致。segments 保留了 thinking → tools → content * 的真实交错顺序，toolCalls 是 segments 中 tool_call 类型的平铺视图。 */ private final class StreamAccumulator { private final StringBuilder content = new StringBuilder(); private final StringBuilder thinking = new StringBuilder(); private final List> toolCalls = new ArrayList<>(); /** 有序事件时间线 — 前端分段渲染的权威数据源 */ private final List> segments = new ArrayList<>(); private final List> browserActions = new ArrayList<>(); private final List warnings = new ArrayList<>(); private final List> planStepResults = new ArrayList<>(); /** RFC-052: tool names whose returnDirect output was folded into the assistant message */ private final List directToolNames = new ArrayList<>(); private int segCounter = 0; private int promptTokens = 0; private int completionTokens = 0; private String runtimeModelName = ""; private String runtimeProviderId = ""; private boolean awaitingApproval = false; private String currentPhase = ""; /** * Graph-emitted FinishReason for the turn (e.g. {@code "incomplete"}, * {@code "stopped"}, {@code "evidence_insufficient"}). Sourced from * the {@code finish_reason} {@link vip.mate.agent.GraphEventPublisher} * event that {@code FinalAnswerNode} attaches to its PENDING_EVENTS * output — same pipeline the SSE accumulator already drains, so the * value is delivered alongside the assistant content (not via a * sibling SSE-only broadcast that would bypass this accumulator). * Persisted into message metadata so downstream filters * (memory promotion gate) see a machine-readable status instead of * having to guess from text. Empty string until the event arrives. */ private String finishReason = ""; /** * Recovery affordance payload from {@link * vip.mate.agent.GraphEventPublisher#feedback}. Persisted into * {@code metadata.feedbackEvent} so a page reload still surfaces * the retry/regenerate/report card on the failed assistant * bubble. Null when the turn ended cleanly. */ private Map feedbackEvent = null; private Long planId = null; private List planSteps = List.of(); private Integer currentPlanStep = null; private Map pendingApproval = null; /** * Multimodal sidecar routing decision for this turn (null when no * routing happened). Captured from the {@code _routing_decision} * event emitted before the graph stream and folded into * {@code metadata.routing} on persistence so the chat UI can show * which sidecar (if any) was invoked. */ private Map routingDecision = null; synchronized void accept(AgentService.StreamDelta delta, String conversationId) { if (delta == null) return; if (delta.isEvent()) { if ("_usage_final".equals(delta.eventType())) { Map data = delta.eventData(); promptTokens = ((Number) data.getOrDefault("promptTokens", 0)).intValue(); completionTokens = ((Number) data.getOrDefault("completionTokens", 0)).intValue(); runtimeModelName = String.valueOf(data.getOrDefault("runtimeModelName", "")); runtimeProviderId = String.valueOf(data.getOrDefault("runtimeProviderId", "")); return; } if ("phase".equals(delta.eventType())) { String phase = String.valueOf(delta.eventData().getOrDefault("phase", "")); if (!phase.isBlank()) { currentPhase = phase; streamTracker.updatePhase(conversationId, phase); // phase 切换时关闭 running 的 content/thinking segment，保留边界 finalizeRunningSegments("content", "thinking"); } } if ("finish_reason".equals(delta.eventType())) { Object reason = delta.eventData().get("reason"); if (reason != null) { // Last-write-wins: graph normally fires this exactly once // at FinalAnswerNode completion. Replay paths that re-enter // the graph after approval will emit a fresh value, which // is the correct behavior — the latest reason is what gets // persisted with the assistant message. finishReason = String.valueOf(reason); } } if (vip.mate.agent.GraphEventPublisher.EVENT_FEEDBACK .equals(delta.eventType())) { // Snapshot the affordance payload so it persists into // message metadata. The same event is also rebroadcast // live (via the broadcastEvent fall-through below) so // an already-mounted UI sees it instantly without // waiting for the message-save round trip. feedbackEvent = delta.eventData(); } if (vip.mate.agent.GraphEventPublisher.EVENT_ROUTING_DECISION.equals(delta.eventType())) { // Captured at turn start; persisted under metadata.routing so the // chat UI can render which sidecar (if any) was invoked. Internal // event — return early to skip rebroadcast on IM channels. routingDecision = delta.eventData(); return; } accumulateToolEvent(delta.eventType(), delta.eventData(), conversationId); try { broadcastEvent(conversationId, delta.eventType(), delta.eventData()); } catch (Exception e) { log.warn("Failed to broadcast event {}: {}", delta.eventType(), e.getMessage()); } return; } // content_delta if (delta.content() != null && !delta.content().isBlank()) { content.append(delta.content()); streamTracker.updatePhase(conversationId, "drafting_answer"); if (!delta.persistenceOnly()) { broadcastEvent(conversationId, "content_delta", Map.of("delta", delta.content())); } // segments: 追加到当前 running content segment，或创建新的 var seg = findLastRunning("content"); if (seg != null) { seg.put("text", seg.getOrDefault("text", "") + delta.content()); } else { finalizeRunningSegments("thinking"); var s = newSegment("content"); s.put("text", delta.content()); segments.add(s); } } // thinking_delta if (delta.thinking() != null && !delta.thinking().isBlank()) { thinking.append(delta.thinking()); if (!delta.persistenceOnly()) { broadcastEvent(conversationId, "thinking_delta", Map.of("delta", delta.thinking())); } var seg = findLastRunning("thinking"); if (seg != null) { seg.put("thinkingText", seg.getOrDefault("thinkingText", "") + delta.thinking()); } else { var s = newSegment("thinking"); s.put("thinkingText", delta.thinking()); segments.add(s); } } } boolean isAwaitingApproval() { return awaitingApproval; } private void accumulateToolEvent(String eventType, Map data, String conversationId) { if ("tool_approval_requested".equals(eventType)) { awaitingApproval = true; currentPhase = "awaiting_approval"; pendingApproval = new LinkedHashMap<>(); pendingApproval.put("pendingId", data.getOrDefault("pendingId", "")); pendingApproval.put("toolName", data.getOrDefault("toolName", "")); pendingApproval.put("arguments", data.getOrDefault("arguments", "")); pendingApproval.put("reason", data.getOrDefault("reason", "")); pendingApproval.put("status", "pending_approval"); if (data.containsKey("findings")) pendingApproval.put("findings", data.get("findings")); if (data.containsKey("maxSeverity")) pendingApproval.put("maxSeverity", data.get("maxSeverity")); if (data.containsKey("summary")) pendingApproval.put("summary", data.get("summary")); streamTracker.updatePhase(conversationId, "awaiting_approval"); } else if ("tool_approval_resolved".equals(eventType)) { if (pendingApproval != null) { pendingApproval.put("status", "approved".equals(String.valueOf(data.getOrDefault("decision", ""))) ? "approved" : "denied"); } } else if ("plan_created".equals(eventType)) { Object rawPlanId = data.get("planId"); if (rawPlanId instanceof Number n) { planId = n.longValue(); } else if (rawPlanId != null) { try { planId = Long.valueOf(String.valueOf(rawPlanId)); } catch (Exception ignored) {} } Object steps = data.get("steps"); if (steps instanceof List list) { planSteps = list.stream().map(String::valueOf).toList(); planStepResults.clear(); for (int i = 0; i < planSteps.size(); i++) { planStepResults.add(null); } } currentPlanStep = 0; } else if ("plan_step_started".equals(eventType)) { Object idx = data.get("index"); if (idx instanceof Number n) { currentPlanStep = n.intValue(); } } else if ("plan_step_completed".equals(eventType)) { Object idx = data.get("index"); if (idx instanceof Number n) { int index = n.intValue(); currentPlanStep = index; ensurePlanStepCapacity(index + 1); Map stepResult = new LinkedHashMap<>(); stepResult.put("result", data.getOrDefault("result", "")); stepResult.put("status", "completed"); planStepResults.set(index, stepResult); } } else if ("browser_action".equals(eventType)) { browserActions.add(new LinkedHashMap<>(data)); } else if ("warning".equals(eventType)) { String warning = String.valueOf(data.getOrDefault("message", data.getOrDefault("delta", ""))); if (!warning.isBlank()) { warnings.add(warning); } } else if ("tool_call_started".equals(eventType)) { // toolCalls（兼容） Map tc = new LinkedHashMap<>(); tc.put("name", data.getOrDefault("toolName", "")); tc.put("arguments", data.getOrDefault("arguments", "")); tc.put("status", "running"); toolCalls.add(tc); // segments: 关闭 running thinking/content，插入 tool_call finalizeRunningSegments("thinking", "content"); var seg = newSegment("tool_call"); seg.put("toolName", data.getOrDefault("toolName", "")); seg.put("toolArgs", data.getOrDefault("arguments", "")); segments.add(seg); } else if ("tool_direct_result".equals(eventType)) { // RFC-052: returnDirect tool — track the tool name so history // replay can render a "data returned directly by tool" badge. // The actual textual content reaches the user/persistence layer // through the regular content_delta path (FinalAnswerNode's // FINAL_ANSWER → StateGraphReActAgent → StreamDelta), so we // intentionally do NOT add a content-bearing segment here to // avoid the user seeing the same text twice. String toolName = String.valueOf(data.getOrDefault("toolName", "")); if (!toolName.isBlank() && !directToolNames.contains(toolName)) { directToolNames.add(toolName); } } else if ("tool_call_completed".equals(eventType)) { String toolName = String.valueOf(data.getOrDefault("toolName", "")); // toolCalls（兼容） for (int i = toolCalls.size() - 1; i >= 0; i--) { Map tc = toolCalls.get(i); if ("running".equals(tc.get("status")) && toolName.equals(tc.get("name"))) { tc.put("result", data.getOrDefault("result", "")); tc.put("success", data.getOrDefault("success", true)); tc.put("status", "completed"); break; } } // segments: 标记对应 tool_call 完成 for (int i = segments.size() - 1; i >= 0; i--) { var seg = segments.get(i); if ("tool_call".equals(seg.get("type")) && "running".equals(seg.get("status")) && toolName.equals(seg.get("toolName"))) { seg.put("status", "completed"); seg.put("toolResult", data.getOrDefault("result", "")); seg.put("toolSuccess", data.getOrDefault("success", true)); break; } } } } private void ensurePlanStepCapacity(int size) { while (planStepResults.size() < size) { planStepResults.add(null); } } // ==================== Segment helpers ==================== private Map newSegment(String type) { Map seg = new LinkedHashMap<>(); seg.put("id", type.substring(0, 2) + "-" + segCounter++); seg.put("type", type); seg.put("status", "running"); return seg; } private Map findLastRunning(String type) { for (int i = segments.size() - 1; i >= 0; i--) { var seg = segments.get(i); if (type.equals(seg.get("type")) && "running".equals(seg.get("status"))) return seg; } return null; } private void finalizeRunningSegments(String... types) { var typeSet = java.util.Set.of(types); for (var seg : segments) { if ("running".equals(seg.get("status")) && typeSet.contains(seg.get("type"))) { seg.put("status", "completed"); } } } // ==================== 原有访问器 ==================== String getContent() { return content.toString().trim(); } String getThinking() { return thinking.toString().trim(); } int getPromptTokens() { return promptTokens; } int getCompletionTokens() { return completionTokens; } String getRuntimeModelName() { return runtimeModelName; } String getRuntimeProviderId() { return runtimeProviderId; } String getCurrentPhase() { return currentPhase; } String getFinishReason() { return finishReason; } boolean segmentsEmpty() { return segments.isEmpty(); } synchronized List toAssistantParts() { List parts = new ArrayList<>(); if (!getContent().isBlank()) { MessageContentPart textPart = new MessageContentPart(); textPart.setType("text"); textPart.setText(getContent()); parts.add(textPart); } if (!getThinking().isBlank()) { MessageContentPart thinkingPart = new MessageContentPart(); thinkingPart.setType("thinking"); thinkingPart.setText(getThinking()); parts.add(thinkingPart); } for (Map tc : toolCalls) { try { parts.add(MessageContentPart.toolCall(objectMapper.writeValueAsString(tc))); } catch (Exception e) { log.warn("Failed to serialize tool call: {}", e.getMessage()); } } return parts; } void finalizeToolCalls() { for (Map tc : toolCalls) { if ("running".equals(tc.get("status"))) tc.put("status", "completed"); } } /** * 生成 metadata JSON：包含 toolCalls + segments。 * toolCalls 保留兼容旧 UI，segments 是按事件顺序的完整时间线。 */ synchronized String toMetadataJson() { finalizeToolCalls(); finalizeRunningSegments("thinking", "content", "tool_call"); try { Map metadata = new LinkedHashMap<>(); if (!toolCalls.isEmpty()) { metadata.put("toolCalls", toolCalls); } if (!segments.isEmpty()) { metadata.put("segments", segments); } if (!currentPhase.isBlank()) { metadata.put("currentPhase", currentPhase); } if (planId != null || !planSteps.isEmpty() || currentPlanStep != null) { Map plan = new LinkedHashMap<>(); if (planId != null) plan.put("planId", planId); if (!planSteps.isEmpty()) plan.put("steps", planSteps); if (currentPlanStep != null) plan.put("currentStep", currentPlanStep); if (planStepResults.stream().anyMatch(java.util.Objects::nonNull)) { plan.put("stepResults", planStepResults); } metadata.put("plan", plan); } if (pendingApproval != null && !pendingApproval.isEmpty()) { metadata.put("pendingApproval", pendingApproval); } if (!browserActions.isEmpty()) { metadata.put("browserActions", browserActions); } if (!warnings.isEmpty()) { metadata.put("warnings", warnings); } if (!directToolNames.isEmpty()) { // RFC-052 §3.3: only the tool names go into metadata — // the full content already lives in mate_message.content // (assembled by FinalAnswerNode). UI uses this to badge // historical messages as "data returned directly by tool". metadata.put("directToolNames", directToolNames); } if (!finishReason.isEmpty()) { // Surface graph FinishReason so MemorySummarizationGate and // any other downstream consumer can branch on a structured // status (e.g. skip INCOMPLETE / STOPPED / ERROR_FALLBACK // turns from long-term memory promotion) instead of doing // brittle text matching on the assistant content. metadata.put("finishReason", finishReason); } if (feedbackEvent != null && !feedbackEvent.isEmpty()) { // Persist the recovery-affordance payload so the // retry/regenerate/report card survives page reload. // Stored as-is (errorType, errorMessage, actions, // timestamp) — frontend MessageBubble reads // metadata.feedbackEvent and renders one button per // entry in `actions`. metadata.put("feedbackEvent", feedbackEvent); } if (routingDecision != null && !routingDecision.isEmpty()) { metadata.put("routing", routingDecision); } return objectMapper.writeValueAsString(metadata); } catch (Exception e) { log.warn("Failed to serialize metadata: {}", e.getMessage()); return "{}"; } } } private static Long parseLongOrNull(String s) { if (s == null || s.isBlank()) return null; try { return Long.parseLong(s.trim()); } catch (NumberFormatException e) { return null; } } }