ai-agent/workflow/service/flow/exec_record.go

package flow

import (
	"context"
	"errors"
	"fmt"
	"time"

	"gitea.redpowerfuture.com/red-future/common/oss"
	"github.com/cloudwego/eino/compose"
	"github.com/gogf/gf/v2/frame/g"
	"github.com/gogf/gf/v2/os/glog"
	"github.com/gogf/gf/v2/util/gconv"

	"ai-agent/gateway"
	"ai-agent/workflow/consts/flow"
	"ai-agent/workflow/consts/node"
	nodeDao "ai-agent/workflow/dao/node"
	sessionDao "ai-agent/workflow/dao/session"
	flowDto "ai-agent/workflow/model/dto/flow"
	nodeDto "ai-agent/workflow/model/dto/node"
	"ai-agent/workflow/model/entity"
)

// ====================== 执行记录落库 ======================

// recordExecutionFailure 记录一次失败状态。有 execId 直接更新该记录；拿不到 execId
// （executeOrResume 在创建记录后、返回前 panic，或查询/创建执行记录失败）时，
// 兜底按会话+工作流查最近一条仍处于"运行中"的记录标记为失败，避免前端已报错但记录卡在 Running。
// 最近记录已是成功/失败状态则不处理（可能是上一次执行的结果，不应误改）。
func recordExecutionFailure(ctx context.Context, sessionId string, flowId int64, execId int64, runErr error) {
	// 兜底失败路径同样携带 retryable 分类（与 handleExecute 一致）：
	// 用户取消=0；其余（程序关停中断/程序报错）可重试=1，保证任意 status=3 写库都带分类
	var retryable, retryCnt *int
	if runErr != nil {
		if errors.Is(runErr, context.Canceled) && !IsShuttingDown() {
			retryable, retryCnt = intptr(0), intptr(0)
		} else {
			retryable, retryCnt = intptr(1), intptr(0)
		}
	}
	if !g.IsEmpty(execId) {
		recordWorkflow(ctx, execId, 0, runErr, retryable, retryCnt)
		return
	}
	lastExec, err := sessionDao.ExecWorkflowDao.GetLatestBySessionAndFlow(ctx, sessionId, flowId)
	if err != nil || lastExec == nil {
		glog.Errorf(ctx, "兜底标记失败状态失败: sessionId=%s flowId=%d err=%v", sessionId, flowId, err)
		return
	}
	if lastExec.Status == nil || *lastExec.Status != *flow.FlowExecutionStatusRunning.Code() {
		return
	}
	recordWorkflow(ctx, lastExec.Id, 0, runErr, retryable, retryCnt)
}

// intptr 取 int 值指针（recordWorkflow 的 retryable/retryCount 参数：nil=不改动）
func intptr(v int) *int { return &v }

// recordWorkflow 把一次工作流执行写入 exec_workflow/exec_workflow_result：运行记录 + 输出文件结果。
// retryable/retryCount 非 nil 时随终态同语句原子落库，避免"先 UpdateRetry 再 Update"两步写部分生效
// 导致 retryable 与 status/error_message 不一致（关停中断场景曾出现 retryable=0 但 message=程序关停中断）
func recordWorkflow(ctx context.Context, id int64, duration time.Duration, runErr error, retryable, retryCount *int) {
	// exec_workflow 状态沿用 1-运行中,2-成功,3-失败；前端结果卡片也只识别 1/2/3
	// （4 会误显示为"运行中"），故取消同样记为失败，错误信息写"用户已终止执行"
	// error_message 存友好提示，error 存原始错误明细
	status := flow.FlowExecutionStatusSuccess
	var errorMessage, errorDetail string
	if runErr != nil {
		status = flow.FlowExecutionStatusFailed
		switch {
		case errors.Is(runErr, context.Canceled):
			errorMessage = errWorkflowTerminated
		case errors.Is(runErr, errInterruptedByShutdown):
			errorMessage = "程序关停中断"
		default:
			errorMessage = "工作流执行失败"
			errorDetail = runErr.Error()
		}
	}
	data := map[string]any{
		entity.ExecWorkflowCol.Status: *status.Code(),
	}
	if d := int64(duration.Seconds()); d != 0 {
		data[entity.ExecWorkflowCol.Duration] = d
	}
	if errorMessage != "" {
		data[entity.ExecWorkflowCol.ErrorMessage] = errorMessage
	}
	if errorDetail != "" {
		data[entity.ExecWorkflowCol.Error] = errorDetail
	}
	if retryable != nil {
		data[entity.ExecWorkflowCol.Retryable] = *retryable
		data[entity.ExecWorkflowCol.RetryCount] = *retryCount
	}
	if err := sessionDao.ExecWorkflowDao.UpdateMap(ctx, id, data); err != nil {
		glog.Errorf(ctx, "exec_workflow 终态落库失败 execId=%d: %v", id, err)
		return
	}
	// 执行成功：重新执行复用了同一条记录，需显式清空，避免上一次失败的报错残留
	if runErr == nil {
		if _, err := sessionDao.ExecWorkflowDao.ClearError(ctx, id); err != nil {
			glog.Errorf(ctx, "exec_workflow 报错信息清空失败: %v", err)
		}
	}
	// 工作流计费：终态结算（成功→Settle/用户取消→Cancel/永久失败→Fail/可恢复→跳过）。
	// recordWorkflow 汇聚全部路径（WS/恢复/panic），此处一处接线全覆盖；计费错误仅记日志不拖垮落库
	settleBilling(ctx, id, runErr, retryable, retryCount)
}

// workflowResultFileUrls 查询指定工作流执行保存的结果文件路径（带文件前缀，与 session/get 返回一致）
func workflowResultFileUrls(ctx context.Context, execId int64) []string {
	results, err := sessionDao.ExecWorkflowResultDao.ListByExecId(ctx, execId)
	if err != nil {
		glog.Errorf(ctx, "查询工作流结果路径失败: %v", err)
		return nil
	}
	prefix, _ := oss.GetFileAddressPrefix(ctx)
	urls := make([]string, 0, len(results))
	for _, r := range results {
		if r.ResultFileUrl != "" {
			urls = append(urls, prefix+r.ResultFileUrl)
		}
	}
	return urls
}

// ====================== 节点执行记录 ======================

// BuildNodeExecutionInput 构建节点执行入参，包含中断恢复逻辑
func BuildNodeExecutionInput(ctx context.Context, input any, flowNode entity.FlowNode) (*flowDto.FlowExecutionInput, *flowDto.NodeExecutionInput, error) {
	execInput := new(flowDto.FlowExecutionInput)

	wasInterrupted, _, _ := compose.GetInterruptState[any](ctx)
	if wasInterrupted {
		if err := compose.ProcessState(ctx, func(_ context.Context, s *flowDto.NodeExecutionState) error {
			execInput = s.SavedFlowInput
			return nil
		}); err != nil {
			return nil, nil, fmt.Errorf("节点：%v 进程状态读取失败: %v", flowNode.Name, err)
		}
		// 兼容旧 checkpoint（无 SavedFlowInput 时，降级使用 input 参数）
		if execInput == nil {
			var ok bool
			execInput, ok = input.(*flowDto.FlowExecutionInput)
			if !ok {
				return nil, nil, fmt.Errorf("节点：%v 进程状态为空，节点入参类型不匹配", flowNode.Name)
			}
			if g.IsEmpty(execInput) {
				return nil, nil, fmt.Errorf("节点：%v 进程状态为空，节点入参参数为空", flowNode.Name)
			}
		}
		// 续跑必定非全新执行：checkpoint 恢复的 SavedFlowInput 里 ForceNewRun 是上次（fresh）执行留下的 true，
		// 不清则 ModelLambda 误走"清段重生成"而非复用已成功段
		execInput.ForceNewRun = false
	} else {
		var ok bool
		execInput, ok = input.(*flowDto.FlowExecutionInput)
		if !ok {
			return nil, nil, fmt.Errorf("节点：%v 入参类型不匹配", flowNode.Name)
		}
		if g.IsEmpty(execInput) {
			return nil, nil, fmt.Errorf("节点：%v 入参参数为空", flowNode.Name)
		}
	}

	configMap := execInput.ConfigMap
	currentConfig := configMap[flowNode.Id]
	if currentConfig == nil {
		return nil, nil, fmt.Errorf("节点：%v 节点信息为空", flowNode.Name)
	}

	// 构建节点执行入参
	realInput := &flowDto.NodeExecutionInput{
		Config: currentConfig,
		Global: execInput,
	}

	return execInput, realInput, nil
}

// HandleSuccessfulNodeExecution 处理节点执行成功的后续操作
func HandleSuccessfulNodeExecution(ctx context.Context, execInput *flowDto.FlowExecutionInput, realInput *flowDto.NodeExecutionInput, nodeExecutionId int64, flowNode entity.FlowNode, durationMs int64) error {
	// 上传输出到OSS
	ossResult, err := gateway.Upload(ctx, fmt.Sprintf("nodeInput:%v.txt", time.Now().UnixMilli()), gconv.Bytes(gconv.String(realInput)))
	if err != nil {
		return fmt.Errorf("节点：%v 上传OSS失败: %v", realInput.Config.Name, err)
	}

	// 更新执行记录为成功
	if err := UpdateNodeExecutionRecord(ctx, nodeExecutionId, durationMs, node.NodeExecutionStatusSuccess.Code(), ossResult, ""); err != nil {
		return fmt.Errorf("节点：%v 更新成功状态错误: %v", flowNode.Name, err)
	}

	// 记录成功到已执行列表
	RecordExecutionResult(execInput, flowNode.Id, node.NodeExecutionStatusSuccess.Code())
	return nil
}

// HandleFailedNodeExecution 处理节点执行失败的后续操作
func HandleFailedNodeExecution(ctx context.Context, execInput *flowDto.FlowExecutionInput, nodeExecutionId int64, flowNode entity.FlowNode, err error, durationMs int64) error {
	// 保存状态用于续跑
	if stateErr := compose.ProcessState(ctx, func(_ context.Context, s *flowDto.NodeExecutionState) error {
		s.CompletedNodes = append(s.CompletedNodes, flowNode.Name)
		s.SavedFlowInput = execInput
		s.ExecutionCount++
		return nil
	}); stateErr != nil {
		fmt.Printf("节点：%v 进程状态保存失败: %v", flowNode.Name, stateErr)
	}

	if !g.IsEmpty(nodeExecutionId) {
		// 更新执行记录为失败
		if updateErr := UpdateNodeExecutionRecord(ctx, nodeExecutionId, durationMs, node.NodeExecutionStatusFailed.Code(), "", err.Error()); updateErr != nil {
			fmt.Printf("节点：%v 更新失败状态错误: %v", flowNode.Name, updateErr)
		}
	}

	// 触发中断
	return compose.Interrupt(ctx, map[string]string{
		"node":  flowNode.Name,
		"error": err.Error(),
	})
}

// RecordExecutionResult 将节点执行结果写入 Global.ExecutedNodes
func RecordExecutionResult(execInput *flowDto.FlowExecutionInput, nodeId string, status node.NodeExecutionStatus) {
	execInput.ExecutedNodes = append(execInput.ExecutedNodes, flowDto.ExecutedNode{
		NodeId: nodeId,
		Status: status,
	})
}

// UpdateNodeExecutionRecord 更新节点执行记录
func UpdateNodeExecutionRecord(ctx context.Context, nodeExecutionId int64, durationMs int64, status node.NodeExecutionStatus, outputParamsPath string, errMsg string) error {
	if _, err := nodeDao.NodeExecutionDao.Update(ctx, &nodeDto.UpdateNodeExecutionReq{
		Id:               nodeExecutionId,
		DurationMs:       durationMs,
		Status:           status,
		OutputParamsPath: outputParamsPath,
		ErrorMessage:     errMsg,
	}); err != nil {
		return err
	}
	return nil
}

// CreateNodeExecutionRecord 创建节点执行记录，返回记录ID
func CreateNodeExecutionRecord(ctx context.Context, execInput *flowDto.FlowExecutionInput, flowNode entity.FlowNode, inputOssUrl string) (int64, error) {
	id, err := nodeDao.NodeExecutionDao.Insert(ctx, &nodeDto.CreateNodeExecutionReq{
		FlowExecutionId: execInput.ExecutionId,
		NodeId:          flowNode.Id,
		NodeName:        flowNode.Name,
		NodeGroupId:     execInput.NodeGroupId,
		InputParamsPath: inputOssUrl,
		Status:          node.NodeExecutionStatusRunning.Code(),
	})
	if err != nil {
		return 0, fmt.Errorf("节点：%v 创建节点执行记录失败: %v", flowNode.Name, err)
	}
	return id, nil
}