feat(workflow): exec_workflow 增加重试/心跳三列与可恢复查询 DAO

workflow/dao/session/exec_workflow_dao.go

@@ -1,10 +1,13 @@
 package session
 
 import (
+	flow "ai-agent/workflow/consts/flow"
 	"ai-agent/workflow/consts/public"
 	sessionDto "ai-agent/workflow/model/dto/session"
 	"ai-agent/workflow/model/entity"
 	"context"
+	"fmt"
+	"time"
 
 	"gitea.redpowerfuture.com/red-future/common/beans"
 	"gitea.redpowerfuture.com/red-future/common/db/gfdb"
@@ -117,3 +120,55 @@ func (d *execWorkflowDao) ListBySession(ctx context.Context, sessionId string) (
 	err = r.Structs(&res)
 	return
 }
+
+// ResetRunning 置为运行中并刷新心跳（execute 复用失败记录 / reExecute / 恢复例程共用；
+// 用 map 更新避免 OmitEmpty 省略 0 值，同时写 status、last_heartbeat、node_group_id）
+func (d *execWorkflowDao) ResetRunning(ctx context.Context, id int64, nodeGroupId string) error {
+	_, err := gfdb.DB(ctx, public.DbNameBlackDeacon).Model(ctx, public.TableNameExecWorkflow).
+		Where(entity.ExecWorkflowCol.Id, id).
+		Data(map[string]any{
+			entity.ExecWorkflowCol.Status:        gconv.Int8(*flow.FlowExecutionStatusRunning.Code()),
+			entity.ExecWorkflowCol.LastHeartbeat: time.Now().UnixMilli(),
+			entity.ExecWorkflowCol.NodeGroupId:   nodeGroupId,
+		}).
+		Update()
+	return err
+}
+
+// TouchHeartbeat 更新执行心跳（毫秒时间戳），供后台心跳 goroutine 每 30s 调用一次
+func (d *execWorkflowDao) TouchHeartbeat(ctx context.Context, id int64) error {
+	_, err := gfdb.DB(ctx, public.DbNameBlackDeacon).Model(ctx, public.TableNameExecWorkflow).
+		Where(entity.ExecWorkflowCol.Id, id).
+		Data(map[string]any{entity.ExecWorkflowCol.LastHeartbeat: time.Now().UnixMilli()}).
+		Update()
+	return err
+}
+
+// UpdateRetry 更新重试标记与已重试次数（map 更新，retryable=0 也需写入）
+func (d *execWorkflowDao) UpdateRetry(ctx context.Context, id int64, retryable int, retryCount int) error {
+	_, err := gfdb.DB(ctx, public.DbNameBlackDeacon).Model(ctx, public.TableNameExecWorkflow).
+		Where(entity.ExecWorkflowCol.Id, id).
+		Data(map[string]any{
+			entity.ExecWorkflowCol.Retryable:  retryable,
+			entity.ExecWorkflowCol.RetryCount: retryCount,
+		}).
+		Update()
+	return err
+}
+
+// ListRecoverable 返回可恢复执行：僵尸运行中（status=1 且心跳陈旧）或可重试失败（status=3 且 retryable=1 且未耗尽）
+func (d *execWorkflowDao) ListRecoverable(ctx context.Context, now int64, staleBefore int64, maxRetry int) (res []*entity.ExecWorkflow, err error) {
+	r, err := gfdb.DB(ctx, public.DbNameBlackDeacon).Model(ctx, public.TableNameExecWorkflow).
+		Where(fmt.Sprintf("(%s = ? AND %s < ?) OR (%s = ? AND %s = 1 AND %s < ?)",
+			entity.ExecWorkflowCol.Status, entity.ExecWorkflowCol.LastHeartbeat,
+			entity.ExecWorkflowCol.Status, entity.ExecWorkflowCol.Retryable, entity.ExecWorkflowCol.RetryCount),
+			gconv.Int8(*flow.FlowExecutionStatusRunning.Code()), staleBefore,
+			gconv.Int8(*flow.FlowExecutionStatusFailed.Code()), maxRetry).
+		All()
+	if err != nil {
+		return
+	}
+	// 用 All+Structs 而非 Scan：Scan 生成的列清单会丢嵌入 SQLBaseDO 的 id 等基础列，恢复例程需要 id
+	err = r.Structs(&res)
+	return
+}