feat: enhance workflow resume functionality with interrupted run reuse and error handling

This commit is contained in:
mlogclub
2026-06-24 11:02:16 +08:00
parent 46bcc55658
commit 26578705ad
11 changed files with 333 additions and 40 deletions
+63 -40
View File
@@ -9,6 +9,7 @@ import (
"agent-desk/internal/ai/runtime/executor"
workflowexecutor "agent-desk/internal/ai/runtime/workflow"
"agent-desk/internal/models"
"agent-desk/internal/pkg/errorsx"
"agent-desk/internal/pkg/utils"
"agent-desk/internal/repositories"
@@ -69,34 +70,40 @@ func (s *Service) Resume(ctx context.Context, req ResumeRequest) (*Summary, erro
return nil, err
}
req.AIAgent = aiAgent
if interrupt := repositories.ConversationInterruptRepository.GetByCheckPointID(sqls.DB(), req.CheckPointID); interrupt != nil && strings.TrimSpace(interrupt.RequestData) != "" {
workflowResult, err := workflowexecutor.NewExecutor().Resume(ctx, workflowexecutor.Input{
Definition: workflow.Definition,
Conversation: req.Conversation,
AIAgent: req.AIAgent,
AIConfig: req.AIConfig,
}, interrupt.RequestData, firstWorkflowResumeText(req.ResumeData))
if err != nil {
if workflowResult != nil {
_, _ = writeWorkflowRun(Request{
Conversation: req.Conversation,
UserMessage: req.UserMessage,
AIAgent: req.AIAgent,
AIConfig: req.AIConfig,
}, workflow, workflowResult, err.Error())
if interrupt := repositories.ConversationInterruptRepository.GetByCheckPointID(sqls.DB(), req.CheckPointID); interrupt != nil {
if strings.TrimSpace(interrupt.RequestData) == "" {
if interrupt.WorkflowRunID > 0 || strings.HasPrefix(strings.TrimSpace(req.CheckPointID), "workflow:") {
return nil, errorsx.InvalidParam("workflow checkpoint data is required")
}
return nil, err
} else {
workflowResult, err := workflowexecutor.NewExecutor().Resume(ctx, workflowexecutor.Input{
Definition: workflow.Definition,
Conversation: req.Conversation,
AIAgent: req.AIAgent,
AIConfig: req.AIConfig,
}, interrupt.RequestData, firstWorkflowResumeText(req.ResumeData))
if err != nil {
if workflowResult != nil {
_, _ = writeWorkflowRunWithExistingID(Request{
Conversation: req.Conversation,
UserMessage: req.UserMessage,
AIAgent: req.AIAgent,
AIConfig: req.AIConfig,
}, workflow, workflowResult, err.Error(), interrupt.WorkflowRunID)
}
return nil, err
}
workflowRunID, err := writeWorkflowRunWithExistingID(Request{
Conversation: req.Conversation,
UserMessage: req.UserMessage,
AIAgent: req.AIAgent,
AIConfig: req.AIConfig,
}, workflow, workflowResult, "", interrupt.WorkflowRunID)
if err != nil {
return nil, err
}
return toWorkflowSummary(workflowResult, req.AIConfig.ModelName, workflow, workflowRunID), nil
}
workflowRunID, err := writeWorkflowRun(Request{
Conversation: req.Conversation,
UserMessage: req.UserMessage,
AIAgent: req.AIAgent,
AIConfig: req.AIConfig,
}, workflow, workflowResult, "")
if err != nil {
return nil, err
}
return toWorkflowSummary(workflowResult, req.AIConfig.ModelName, workflow, workflowRunID), nil
}
toolSet, err := s.prepare.prepareToolsForResume(req)
if err != nil {
@@ -173,6 +180,10 @@ func toWorkflowInterruptSummaries(items []workflowexecutor.InterruptSummary) []I
}
func writeWorkflowRun(req Request, workflow resolvedWorkflow, result *workflowexecutor.Result, errorMessage string) (int64, error) {
return writeWorkflowRunWithExistingID(req, workflow, result, errorMessage, 0)
}
func writeWorkflowRunWithExistingID(req Request, workflow resolvedWorkflow, result *workflowexecutor.Result, errorMessage string, existingRunID int64) (int64, error) {
if result == nil {
return 0, nil
}
@@ -185,20 +196,32 @@ func writeWorkflowRun(req Request, workflow resolvedWorkflow, result *workflowex
runStatus := workflowRunStatus(result.Status, errorMessage)
var runID int64
err := sqls.WithTransaction(func(ctx *sqls.TxContext) error {
run := &models.AIWorkflowRun{
WorkflowID: workflow.WorkflowID,
WorkflowVersionID: workflow.VersionID,
ConversationID: req.Conversation.ID,
AIAgentID: req.AIAgent.ID,
MessageID: req.UserMessage.ID,
Status: runStatus,
StartedAt: now,
EndedAt: &endedAt,
InterruptType: firstWorkflowInterruptType(result),
InterruptNodeID: firstWorkflowInterruptNodeID(result),
ErrorMessage: errorMessage,
}
if err := repositories.AIWorkflowRunRepository.Create(ctx.Tx, run); err != nil {
run := repositories.AIWorkflowRunRepository.Get(ctx.Tx, existingRunID)
if run == nil {
run = &models.AIWorkflowRun{
WorkflowID: workflow.WorkflowID,
WorkflowVersionID: workflow.VersionID,
ConversationID: req.Conversation.ID,
AIAgentID: req.AIAgent.ID,
MessageID: req.UserMessage.ID,
Status: runStatus,
StartedAt: now,
EndedAt: &endedAt,
InterruptType: firstWorkflowInterruptType(result),
InterruptNodeID: firstWorkflowInterruptNodeID(result),
ErrorMessage: errorMessage,
}
if err := repositories.AIWorkflowRunRepository.Create(ctx.Tx, run); err != nil {
return err
}
} else if err := repositories.AIWorkflowRunRepository.Updates(ctx.Tx, run.ID, map[string]any{
"status": runStatus,
"ended_at": &endedAt,
"interrupt_type": firstWorkflowInterruptType(result),
"interrupt_node_id": firstWorkflowInterruptNodeID(result),
"error_message": errorMessage,
"updated_at": now,
}); err != nil {
return err
}
runID = run.ID
@@ -101,6 +101,87 @@ func TestServiceResumeUsesWorkflowCheckpointData(t *testing.T) {
}
}
func TestServiceResumeReusesInterruptedWorkflowRun(t *testing.T) {
db := setupWorkflowResumeTestDB(t)
def := runtimeHumanConfirmDefinition()
version := models.AIWorkflowVersion{
WorkflowID: 1,
Version: 1,
Status: enums.StatusOk,
Definition: mustMarshalDefinition(t, def),
}
if err := db.Create(&version).Error; err != nil {
t.Fatalf("create workflow version: %v", err)
}
interruptedRun := models.AIWorkflowRun{
WorkflowID: version.WorkflowID,
WorkflowVersionID: version.ID,
ConversationID: 1,
AIAgentID: 1,
MessageID: 2,
Status: workflowRunStatusInterrupted,
InterruptType: "human_confirm",
InterruptNodeID: "confirm_1",
}
if err := db.Create(&interruptedRun).Error; err != nil {
t.Fatalf("create interrupted workflow run: %v", err)
}
if err := db.Create(&models.ConversationInterrupt{
ConversationID: 1,
AIAgentID: 1,
CheckPointID: "workflow:1:2:confirm_1",
InterruptID: "confirm_1",
InterruptType: "human_confirm",
WorkflowRunID: interruptedRun.ID,
WorkflowNodeID: "confirm_1",
RequestData: mustMarshalWorkflowCheckpoint(t, def),
Status: "pending",
}).Error; err != nil {
t.Fatalf("create interrupt: %v", err)
}
summary, err := NewService().Resume(context.Background(), ResumeRequest{
Conversation: models.Conversation{ID: 1},
UserMessage: models.Message{ID: 3, Content: "确认"},
AIAgent: models.AIAgent{
ID: 1,
WorkflowVersionID: version.ID,
},
AIConfig: models.AIConfig{ModelName: "test-model"},
CheckPointID: "workflow:1:2:confirm_1",
ResumeData: map[string]string{
"confirm_1": "确认",
},
})
if err != nil {
t.Fatalf("resume workflow: %v", err)
}
if summary.WorkflowRunID != interruptedRun.ID {
t.Fatalf("expected resume to reuse workflow run %d, got %d", interruptedRun.ID, summary.WorkflowRunID)
}
var runCount int64
if err := db.Model(&models.AIWorkflowRun{}).Count(&runCount).Error; err != nil {
t.Fatalf("count workflow runs: %v", err)
}
if runCount != 1 {
t.Fatalf("expected one workflow run after resume, got %d", runCount)
}
var updated models.AIWorkflowRun
if err := db.First(&updated, interruptedRun.ID).Error; err != nil {
t.Fatalf("find updated workflow run: %v", err)
}
if updated.Status != workflowRunStatusCompleted || updated.ErrorMessage != "" {
t.Fatalf("unexpected updated workflow run: %#v", updated)
}
var nodeCount int64
if err := db.Model(&models.AIWorkflowNodeRun{}).Where("workflow_run_id = ?", interruptedRun.ID).Count(&nodeCount).Error; err != nil {
t.Fatalf("count node runs: %v", err)
}
if nodeCount == 0 {
t.Fatalf("expected resumed node traces to be appended to original workflow run")
}
}
func TestServiceRunWritesFailedWorkflowRun(t *testing.T) {
db := setupWorkflowResumeTestDB(t)
def := dsl.Definition{
@@ -47,6 +47,7 @@ func TestBuildConversationInterruptStoresWorkflowCheckpointData(t *testing.T) {
item := buildConversationInterrupt(testConversation(1), testMessage(2), testAIAgent(3), &applicationruntime.Summary{
CheckPointData: `{"confirmNodeId":"confirm_1"}`,
Interrupted: true,
WorkflowRunID: 99,
Interrupts: []applicationruntime.InterruptContextSummary{
{Type: "human_confirm", ID: "confirm_1", InfoPreview: `{"message":"请确认"}`},
},
@@ -57,6 +58,9 @@ func TestBuildConversationInterruptStoresWorkflowCheckpointData(t *testing.T) {
if item.RequestData != `{"confirmNodeId":"confirm_1"}` {
t.Fatalf("unexpected request data: %q", item.RequestData)
}
if item.WorkflowRunID != 99 || item.WorkflowNodeID != "confirm_1" {
t.Fatalf("unexpected workflow interrupt identity: run=%d node=%q", item.WorkflowRunID, item.WorkflowNodeID)
}
}
type fakeErr string
@@ -32,6 +32,8 @@ func buildConversationInterrupt(conversation models.Conversation, message models
item.SourceMessageID = message.ID
item.InterruptID = firstInterruptID(summary)
item.InterruptType = firstInterruptType(summary)
item.WorkflowRunID = summary.WorkflowRunID
item.WorkflowNodeID = firstInterruptID(summary)
item.Status = "pending"
item.PromptText = resolveInterruptPrompt(summary)
item.RequestData = strings.TrimSpace(summary.CheckPointData)
+12
View File
@@ -420,6 +420,18 @@ func (e *Executor) executeAnalyzeConversation(ctx context.Context, state *runSta
}
func (e *Executor) executeHandoffToHuman(state *runState, node dsl.Node) error {
if _, hasConfirmedInput := node.Inputs["confirmed"]; hasConfirmedInput && !truthy(state.resolveInput(node, "confirmed")) {
state.setNodeVars(node.ID, map[string]any{
"handoffId": int64(0),
"reason": strings.TrimSpace(toString(state.resolveInput(node, "reason"))),
"decision": "cancelled",
"teamId": int64(0),
"assigneeId": int64(0),
"message": "",
"skipped": true,
})
return nil
}
reason := strings.TrimSpace(toString(state.resolveInput(node, "reason")))
result, err := services.ConversationHumanDispatchService.HandoffByAIWithRequestID(
state.input.Conversation.ID,
@@ -92,6 +92,50 @@ func TestExecutorHandoffToHumanRunsRealDispatchAction(t *testing.T) {
}
}
func TestExecutorResumeSkipsHandoffWhenConfirmationCancelled(t *testing.T) {
db := setupWorkflowExecutorHandoffDB(t)
aiAgent := createWorkflowExecutorHandoffAIAgent(t, db, "1")
createWorkflowExecutorHandoffTeam(t, db, 1, "售后支持组")
createWorkflowExecutorHandoffActiveSchedule(t, db, 1)
createWorkflowExecutorHandoffAgentProfile(t, db, 101, 1)
conversation := createWorkflowExecutorHandoffConversation(t, db, aiAgent.ID)
userMessage := createWorkflowExecutorCustomerMessage(t, db, conversation.ID, "需要人工处理")
input := Input{
Definition: handoffAfterConfirmationWorkflowDefinition(),
Conversation: conversation,
UserMessage: userMessage,
AIAgent: aiAgent,
}
interrupted, err := NewExecutor().Execute(context.Background(), input)
if err != nil {
t.Fatalf("execute workflow: %v", err)
}
if !interrupted.Interrupted {
t.Fatalf("expected workflow to interrupt before handoff")
}
result, err := NewExecutor().Resume(context.Background(), input, interrupted.CheckPointData, "取消")
if err != nil {
t.Fatalf("resume workflow: %v", err)
}
if result.Interrupted {
t.Fatalf("expected cancelled resume to complete")
}
assertPath(t, result.NodePath, []string{"handoff_1", "end_1"})
current := services.ConversationService.Get(conversation.ID)
if current.Status != enums.IMConversationStatusAIServing {
t.Fatalf("expected conversation to remain ai serving, got status=%d", current.Status)
}
if current.CurrentAssigneeID != 0 || current.CurrentTeamID != 0 || current.HandoffAt != nil {
t.Fatalf("expected no handoff side effect, got assignee=%d team=%d handoffAt=%v", current.CurrentAssigneeID, current.CurrentTeamID, current.HandoffAt)
}
if count := services.MessageService.Count(sqls.NewCnd().Eq("conversation_id", conversation.ID).Eq("sender_type", enums.IMSenderTypeAI)); count != 0 {
t.Fatalf("expected no handoff notice message, got %d", count)
}
}
func TestExecutorAnalyzeConversationOutputsBranchVariables(t *testing.T) {
db := setupWorkflowExecutorHandoffDB(t)
aiAgent := createWorkflowExecutorHandoffAIAgent(t, db, "1")
@@ -427,6 +471,31 @@ func handoffWorkflowDefinition() dsl.Definition {
}
}
func handoffAfterConfirmationWorkflowDefinition() dsl.Definition {
return dsl.Definition{
SchemaVersion: 1,
EntryNodeID: "start_1",
Nodes: []dsl.Node{
{ID: "start_1", Type: workflowregistry.NodeTypeStart, Name: "Start"},
{ID: "prompt_1", Type: workflowregistry.NodeTypeLLMReply, Name: "Prompt", Config: []byte(`{"staticReply":"请确认转人工"}`)},
{ID: "confirm_1", Type: workflowregistry.NodeTypeHumanConfirm, Name: "Confirm", Inputs: map[string]dsl.VariableSelector{
"prompt": {NodeID: "prompt_1", Field: "replyText"},
}},
{ID: "handoff_1", Type: workflowregistry.NodeTypeHandoffToHuman, Name: "Handoff", Inputs: map[string]dsl.VariableSelector{
"reason": {NodeID: "start_1", Field: "userMessage"},
"confirmed": {NodeID: "confirm_1", Field: "confirmed"},
}},
{ID: "end_1", Type: workflowregistry.NodeTypeEnd, Name: "End"},
},
Edges: []dsl.Edge{
{ID: "edge_start_prompt", Source: "start_1", Target: "prompt_1"},
{ID: "edge_prompt_confirm", Source: "prompt_1", Target: "confirm_1"},
{ID: "edge_confirm_handoff", Source: "confirm_1", Target: "handoff_1"},
{ID: "edge_handoff_end", Source: "handoff_1", Target: "end_1"},
},
}
}
func setupWorkflowExecutorHandoffDB(t *testing.T) *gorm.DB {
t.Helper()
dbName := strings.NewReplacer("/", "_", " ", "_").Replace(t.Name())