perf: workflow code (#5548)

* perf: workflow code * add tool call limit

perf: workflow code (#5548)
* perf: workflow code * add tool call limit
c4799df3 · Archer · GitHub · 610634e1 · c4799df3 · c4799df3
Commit c4799df3 authored Aug 27, 2025 by Archer Committed by GitHub Aug 27, 2025
11 changed files
--- a/packages/global/core/workflow/runtime/type.d.ts
+++ b/packages/global/core/workflow/runtime/type.d.ts
@@ -72,9 +72,10 @@ export type ChatDispatchProps = {
  maxRunTimes: number;
  isToolCall?: boolean;
  workflowStreamResponse?: WorkflowResponseType;
-  workflowDispatchDeep?: number;
  version?: 'v1' | 'v2';
+  workflowDispatchDeep: number;
  responseAllData?: boolean;
  responseDetail?: boolean;
 };

--- a/packages/service/core/ai/llm/promptToolCall.ts
+++ b/packages/service/core/ai/llm/promptToolCall.ts
@@ -66,11 +66,12 @@ export const promptToolCallMessageRewrite = (
  return cloneMessages;
 };
-const ERROR_TEXT = 'Tool run error';
+const ERROR_TEXT = 'Tool call error';
 export const parsePromptToolCall = (
  str: string
 ): {
  answer: string;
+  streamAnswer?: string;
  toolCalls?: ChatCompletionMessageToolCall[];
 } => {
  str = str.trim();
@@ -99,11 +100,13 @@ export const parsePromptToolCall = (
    } catch (error) {
      if (prefixReg.test(str)) {
        return {
-          answer: ERROR_TEXT
+          answer: `${ERROR_TEXT}: ${str}`,
+          streamAnswer: `${ERROR_TEXT}: ${str}`
        };
      } else {
        return {
-          answer: str
+          answer: str,
+          streamAnswer: str
        };
      }
    }

--- a/packages/service/core/ai/llm/request.ts
+++ b/packages/service/core/ai/llm/request.ts
@@ -324,7 +324,11 @@ export const createStreamResponse = async ({
      }
      const { reasoningContent, content, finish_reason, usage } = getResponseData();
-      const { answer: llmAnswer, toolCalls } = parsePromptToolCall(content);
+      const { answer: llmAnswer, streamAnswer, toolCalls } = parsePromptToolCall(content);
+      if (streamAnswer) {
+        onStreaming?.({ text: streamAnswer });
+      }
      toolCalls?.forEach((call) => {
        onToolCall?.({ call });

--- a/packages/service/core/workflow/dispatch/abandoned/runApp.ts
+++ b/packages/service/core/workflow/dispatch/abandoned/runApp.ts
@@ -2,7 +2,7 @@
 import type { ChatItemType } from '@fastgpt/global/core/chat/type.d';
 import type { ModuleDispatchProps } from '@fastgpt/global/core/workflow/runtime/type';
 import { type SelectAppItemType } from '@fastgpt/global/core/workflow/template/system/abandoned/runApp/type';
-import { dispatchWorkFlow } from '../index';
+import { runWorkflow } from '../index';
 import { ChatRoleEnum } from '@fastgpt/global/core/chat/constants';
 import { SseResponseEventEnum } from '@fastgpt/global/core/workflow/runtime/constants';
 import {
@@ -59,27 +59,25 @@ export const dispatchAppRequest = async (props: Props): Promise<Response> => {
  const chatHistories = getHistories(history, histories);
  const { files } = chatValue2RuntimePrompt(query);
-  const { flowResponses, flowUsages, assistantResponses, system_memories } = await dispatchWorkFlow(
+  const { flowResponses, flowUsages, assistantResponses, system_memories } = await runWorkflow({
-    {
+    ...props,
-      ...props,
+    runningAppInfo: {
-      runningAppInfo: {
+      id: String(appData._id),
-        id: String(appData._id),
+      teamId: String(appData.teamId),
-        teamId: String(appData.teamId),
+      tmbId: String(appData.tmbId)
-        tmbId: String(appData.tmbId)
+    },
-      },
+    runtimeNodes: storeNodes2RuntimeNodes(
-      runtimeNodes: storeNodes2RuntimeNodes(
+      appData.modules,
-        appData.modules,
+      getWorkflowEntryNodeIds(appData.modules)
-        getWorkflowEntryNodeIds(appData.modules)
+    ),
-      ),
+    runtimeEdges: storeEdges2RuntimeEdges(appData.edges),
-      runtimeEdges: storeEdges2RuntimeEdges(appData.edges),
+    histories: chatHistories,
-      histories: chatHistories,
+    query: runtimePrompt2ChatsValue({
-      query: runtimePrompt2ChatsValue({
+      files,
-        files,
+      text: userChatInput
-        text: userChatInput
+    }),
-      }),
+    variables: props.variables
-      variables: props.variables
+  });
-    }
-  );
  const completeMessages = chatHistories.concat([
    {

--- a/packages/service/core/workflow/dispatch/ai/agent/index.ts
+++ b/packages/service/core/workflow/dispatch/ai/agent/index.ts
@@ -201,7 +201,7 @@ export const dispatchRunTools = async (props: DispatchToolModuleProps): Promise<
      return runToolCall({
        ...props,
        ...requestParams,
-        maxRunToolTimes: 30
+        maxRunToolTimes: 100
      });
    })();

--- a/packages/service/core/workflow/dispatch/ai/agent/toolCall.ts
+++ b/packages/service/core/workflow/dispatch/ai/agent/toolCall.ts
@@ -8,7 +8,7 @@ import { responseWriteController } from '../../../../../common/response';
 import { SseResponseEventEnum } from '@fastgpt/global/core/workflow/runtime/constants';
 import { textAdaptGptResponse } from '@fastgpt/global/core/workflow/runtime/utils';
 import { ChatCompletionRequestMessageRoleEnum } from '@fastgpt/global/core/ai/constants';
-import { dispatchWorkFlow } from '../../index';
+import { runWorkflow } from '../../index';
 import type { DispatchToolModuleProps, RunToolResponse, ToolNodeItemType } from './type';
 import json5 from 'json5';
 import type { DispatchFlowResponse } from '../../type';
@@ -110,7 +110,7 @@ export const runToolCall = async (
    initToolCallEdges(runtimeEdges, interactiveEntryToolParams.entryNodeIds);
    // Run entry tool
-    const toolRunResponse = await dispatchWorkFlow({
+    const toolRunResponse = await runWorkflow({
      ...workflowProps,
      isToolCall: true
    });
@@ -383,7 +383,7 @@ export const runToolCall = async (
      })();
      initToolNodes(runtimeNodes, [toolNode.nodeId], startParams);
-      const toolRunResponse = await dispatchWorkFlow({
+      const toolRunResponse = await runWorkflow({
        ...workflowProps,
        isToolCall: true
      });

--- a/packages/service/core/workflow/dispatch/child/runApp.ts
+++ b/packages/service/core/workflow/dispatch/child/runApp.ts
 import type { ChatItemType } from '@fastgpt/global/core/chat/type.d';
 import type { ModuleDispatchProps } from '@fastgpt/global/core/workflow/runtime/type';
-import { dispatchWorkFlow } from '../index';
+import { runWorkflow } from '../index';
 import { ChatRoleEnum } from '@fastgpt/global/core/chat/constants';
 import { SseResponseEventEnum } from '@fastgpt/global/core/workflow/runtime/constants';
 import {
@@ -132,7 +132,7 @@ export const dispatchRunAppNode = async (props: Props): Promise<Response> => {
      runTimes,
      workflowInteractiveResponse,
      system_memories
-    } = await dispatchWorkFlow({
+    } = await runWorkflow({
      ...props,
      lastInteractive: childrenInteractive,
      // Rewrite stream mode

--- a/packages/service/core/workflow/dispatch/index.ts
+++ b/packages/service/core/workflow/dispatch/index.ts
@@ -47,7 +47,7 @@ import { removeSystemVariable, rewriteRuntimeWorkFlow } from './utils';
 import { getHandleId } from '@fastgpt/global/core/workflow/utils';
 import { callbackMap } from './constants';
-type Props = ChatDispatchProps & {
+type Props = Omit<ChatDispatchProps, 'workflowDispatchDeep'> & {
  runtimeNodes: RuntimeNodeItemType[];
  runtimeEdges: RuntimeEdgeItemType[];
 };
@@ -58,8 +58,62 @@ type NodeResponseCompleteType = Omit<NodeResponseType, 'responseData'> & {
  [DispatchNodeResponseKeyEnum.nodeResponse]?: ChatHistoryItemResType;
 };
-/* running */
+// Run workflow
 export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowResponse> {
+  const { res, stream, externalProvider } = data;
+  let streamCheckTimer: NodeJS.Timeout | null = null;
+  // set sse response headers
+  if (res) {
+    res.setHeader('Connection', 'keep-alive'); // Set keepalive for long connection
+    if (stream) {
+      res.on('close', () => res.end());
+      res.on('error', () => {
+        addLog.error('Request error');
+        res.end();
+      });
+      res.setHeader('Content-Type', 'text/event-stream;charset=utf-8');
+      res.setHeader('Access-Control-Allow-Origin', '*');
+      res.setHeader('X-Accel-Buffering', 'no');
+      res.setHeader('Cache-Control', 'no-cache, no-transform');
+      // 10s sends a message to prevent the browser from thinking that the connection is disconnected
+      streamCheckTimer = setInterval(() => {
+        data?.workflowStreamResponse?.({
+          event: SseResponseEventEnum.answer,
+          data: textAdaptGptResponse({
+            text: ''
+          })
+        });
+      }, 10000);
+    }
+  }
+  // Get default variables
+  const defaultVariables = {
+    ...externalProvider.externalWorkflowVariables,
+    ...getSystemVariables(data)
+  };
+  // Init some props
+  return runWorkflow({
+    ...data,
+    variables: defaultVariables,
+    workflowDispatchDeep: 0
+  }).finally(() => {
+    if (streamCheckTimer) {
+      clearInterval(streamCheckTimer);
+    }
+  });
+}
+type RunWorkflowProps = ChatDispatchProps & {
+  runtimeNodes: RuntimeNodeItemType[];
+  runtimeEdges: RuntimeEdgeItemType[];
+};
+export const runWorkflow = async (data: RunWorkflowProps): Promise<DispatchFlowResponse> => {
  let {
    res,
    runtimeNodes = [],
@@ -67,28 +121,15 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
    histories = [],
    variables = {},
    externalProvider,
-    stream = false,
    retainDatasetCite = true,
    version = 'v1',
    responseDetail = true,
    responseAllData = true
  } = data;
-  const startTime = Date.now();
-  await rewriteRuntimeWorkFlow({ nodes: runtimeNodes, edges: runtimeEdges, lang: data.lang });
-  // 初始化深度和自动增加深度，避免无限嵌套
-  if (!data.workflowDispatchDeep) {
-    data.workflowDispatchDeep = 1;
-  } else {
-    data.workflowDispatchDeep += 1;
-  }
-  const isRootRuntime = data.workflowDispatchDeep === 1;
-  // 初始化 runtimeNodesMap
-  const runtimeNodesMap = new Map(runtimeNodes.map((item) => [item.nodeId, item]));
  // Over max depth
+  data.workflowDispatchDeep++;
+  const isRootRuntime = data.workflowDispatchDeep === 1;
  if (data.workflowDispatchDeep > 20) {
    return {
      flowResponses: [],
@@ -106,42 +147,9 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
    };
  }
-  let workflowRunTimes = 0;
+  const startTime = Date.now();
-  let streamCheckTimer: NodeJS.Timeout | null = null;
-  // Init
-  if (isRootRuntime) {
-    // set sse response headers
-    res?.setHeader('Connection', 'keep-alive'); // Set keepalive for long connection
-    if (stream && res) {
-      res.on('close', () => res.end());
-      res.on('error', () => {
-        addLog.error('Request error');
-        res.end();
-      });
-      res.setHeader('Content-Type', 'text/event-stream;charset=utf-8');
-      res.setHeader('Access-Control-Allow-Origin', '*');
-      res.setHeader('X-Accel-Buffering', 'no');
-      res.setHeader('Cache-Control', 'no-cache, no-transform');
-      // 10s sends a message to prevent the browser from thinking that the connection is disconnected
-      streamCheckTimer = setInterval(() => {
-        data?.workflowStreamResponse?.({
-          event: SseResponseEventEnum.answer,
-          data: textAdaptGptResponse({
-            text: ''
-          })
-        });
-      }, 10000);
-    }
-    // Get default variables
+  await rewriteRuntimeWorkFlow({ nodes: runtimeNodes, edges: runtimeEdges, lang: data.lang });
-    variables = {
-      ...externalProvider.externalWorkflowVariables,
-      ...getSystemVariables(data)
-    };
-  }
  /* 
    工作流队列控制
@@ -161,7 +169,9 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
      - 触发交互节点后，需要跳过所有 skip 节点，避免后续执行了 skipNode。
  */
  class WorkflowQueue {
+    runtimeNodesMap = new Map(runtimeNodes.map((item) => [item.nodeId, item]));
    // Workflow variables
+    workflowRunTimes = 0;
    chatResponses: ChatHistoryItemResType[] = []; // response request and save to database
    chatAssistantResponse: AIChatItemValueItemType[] = []; // The value will be returned to the user
    chatNodeUsages: ChatNodeUsageType[] = [];
@@ -221,7 +231,7 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
      }
      const nodeId = this.activeRunQueue.keys().next().value;
-      const node = nodeId ? runtimeNodesMap.get(nodeId) : undefined;
+      const node = nodeId ? this.runtimeNodesMap.get(nodeId) : undefined;
      if (nodeId) {
        this.activeRunQueue.delete(nodeId);
@@ -501,7 +511,7 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
        system_memories: newMemories
      }: NodeResponseCompleteType) => {
        // Add run times
-        workflowRunTimes += runTimes;
+        this.workflowRunTimes += runTimes;
        data.maxRunTimes -= runTimes;
        if (newMemories) {
@@ -650,7 +660,7 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
      // Get node run status by edges
      const status = checkNodeRunStatus({
-        nodesMap: runtimeNodesMap,
+        nodesMap: this.runtimeNodesMap,
        node,
        runtimeEdges
      });
@@ -820,10 +830,6 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
    });
  }
-  if (streamCheckTimer) {
-    clearInterval(streamCheckTimer);
-  }
  return {
    flowResponses: workflowQueue.chatResponses,
    flowUsages: workflowQueue.chatNodeUsages,
@@ -833,7 +839,7 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
      nextStepRunNodes: workflowQueue.debugNextStepRunNodes
    },
    workflowInteractiveResponse: interactiveResult,
-    [DispatchNodeResponseKeyEnum.runTimes]: workflowRunTimes,
+    [DispatchNodeResponseKeyEnum.runTimes]: workflowQueue.workflowRunTimes,
    [DispatchNodeResponseKeyEnum.assistantResponses]: mergeAssistantResponseAnswerText(
      workflowQueue.chatAssistantResponse
    ),
@@ -848,7 +854,7 @@ export async function dispatchWorkFlow(data: Props): Promise<DispatchFlowRespons
        : undefined,
    durationSeconds
  };
-}
+};
 /* get system variable */
 const getSystemVariables = ({

--- a/packages/service/core/workflow/dispatch/loop/runLoop.ts
+++ b/packages/service/core/workflow/dispatch/loop/runLoop.ts
@@ -4,7 +4,7 @@ import {
  type DispatchNodeResultType,
  type ModuleDispatchProps
 } from '@fastgpt/global/core/workflow/runtime/type';
-import { dispatchWorkFlow } from '..';
+import { runWorkflow } from '..';
 import { DispatchNodeResponseKeyEnum } from '@fastgpt/global/core/workflow/runtime/constants';
 import {
  type AIChatItemValueItemType,
@@ -93,7 +93,7 @@ export const dispatchLoop = async (props: Props): Promise<Response> => {
    index++;
-    const response = await dispatchWorkFlow({
+    const response = await runWorkflow({
      ...props,
      lastInteractive: interactiveData?.childrenResponse,
      variables: newVariables,

--- a/packages/service/core/workflow/dispatch/plugin/run.ts
+++ b/packages/service/core/workflow/dispatch/plugin/run.ts
@@ -20,7 +20,7 @@ import { filterSystemVariables, getNodeErrResponse } from '../utils';
 import { getPluginRunUserQuery } from '@fastgpt/global/core/workflow/utils';
 import type { NodeInputKeyEnum, NodeOutputKeyEnum } from '@fastgpt/global/core/workflow/constants';
 import { getChildAppRuntimeById } from '../../../app/plugin/controller';
-import { dispatchWorkFlow } from '../index';
+import { runWorkflow } from '../index';
 import { getUserChatInfoAndAuthTeamPoints } from '../../../../support/permission/auth/team';
 import { dispatchRunTool } from '../child/runTool';
 import type { PluginRuntimeType } from '@fastgpt/global/core/app/plugin/type';
@@ -118,7 +118,7 @@ export const dispatchRunPlugin = async (props: RunPluginProps): Promise<RunPlugi
      ...(externalProvider ? externalProvider.externalWorkflowVariables : {})
    };
    const { flowResponses, flowUsages, assistantResponses, runTimes, system_memories } =
-      await dispatchWorkFlow({
+      await runWorkflow({
        ...props,
        // Rewrite stream mode
        ...(system_forbid_stream

--- a/test/cases/service/core/ai/llm/toolCall.test.ts
+++ b/test/cases/service/core/ai/llm/toolCall.test.ts
@@ -141,18 +141,20 @@ describe('parsePromptToolCall function tests', () => {
      const input = '1: {"name": "tool", "arguments": invalid json}';
      const result = parsePromptToolCall(input);
-      expect(result).toEqual({
+      expect(result.answer).toEqual(
-        answer: 'Tool run error'
+        'Tool call error: 1: {"name": "tool", "arguments": invalid json}'
-      });
+      );
+      expect(result.streamAnswer).toEqual(
+        'Tool call error: 1: {"name": "tool", "arguments": invalid json}'
+      );
    });
    it('should return error message for incomplete JSON with 1:', () => {
      const input = '1: {"name": "tool"';
      const result = parsePromptToolCall(input);
-      expect(result).toEqual({
+      expect(result.answer).toEqual('Tool call error: 1: {"name": "tool"');
-        answer: 'Tool run error'
+      expect(result.streamAnswer).toEqual('Tool call error: 1: {"name": "tool"');
-      });
    });
    it('should handle empty JSON object with 1: (creates tool call with undefined properties)', () => {
@@ -187,18 +189,16 @@ describe('parsePromptToolCall function tests', () => {
      const input = '1:';
      const result = parsePromptToolCall(input);
-      expect(result).toEqual({
+      expect(result.answer).toEqual('Tool call error: 1:');
-        answer: 'Tool run error'
+      expect(result.streamAnswer).toEqual('Tool call error: 1:');
-      });
    });
    it('should handle input with only prefix and whitespace', () => {
      const input = '1:   ';
      const result = parsePromptToolCall(input);
-      expect(result).toEqual({
+      expect(result.answer).toEqual('Tool call error: 1:');
-        answer: 'Tool run error'
+      expect(result.streamAnswer).toEqual('Tool call error: 1:');
-      });
    });
    it('should handle JSON5 syntax in tool call', () => {
@@ -244,9 +244,12 @@ describe('parsePromptToolCall function tests', () => {
      const result = parsePromptToolCall(input);
      // The sliceJsonStr function can't properly extract JSON when there's extra text after
-      expect(result).toEqual({
+      expect(result.answer).toEqual(
-        answer: 'Tool run error'
+        'Tool call error: Text 1: {"name": "tool1", "arguments": {"param": "value"}} more text 1: {"name": "tool2", "arguments": {}}'
-      });
+      );
+      expect(result.streamAnswer).toEqual(
+        'Tool call error: Text 1: {"name": "tool1", "arguments": {"param": "value"}} more text 1: {"name": "tool2", "arguments": {}}'
+      );
    });
    it('should handle tool name with underscores and numbers', () => {