Pass an error message to the failure node

flytepropeller/pkg/controller/executors/failure_node_lookup.go

@@ -3,13 +3,15 @@ package executors
 import (
 	"context"
 
+	"github.com/flyteorg/flyte/flyteidl/gen/pb-go/flyteidl/core"
 	"github.com/flyteorg/flyte/flytepropeller/pkg/apis/flyteworkflow/v1alpha1"
 )
 
 type FailureNodeLookup struct {
 	NodeLookup
 	FailureNode       v1alpha1.ExecutableNode
 	FailureNodeStatus v1alpha1.ExecutableNodeStatus
+	OriginalError     *core.ExecutionError
 }
 
 func (f FailureNodeLookup) GetNode(nodeID v1alpha1.NodeID) (v1alpha1.ExecutableNode, bool) {
@@ -35,10 +37,15 @@ func (f FailureNodeLookup) FromNode(id v1alpha1.NodeID) ([]v1alpha1.NodeID, erro
 	return nil, nil
 }
 
-func NewFailureNodeLookup(nodeLookup NodeLookup, failureNode v1alpha1.ExecutableNode, failureNodeStatus v1alpha1.ExecutableNodeStatus) NodeLookup {
+func (f FailureNodeLookup) GetOriginalError() (*core.ExecutionError, error) {
+	return f.OriginalError, nil
+}
+
+func NewFailureNodeLookup(nodeLookup NodeLookup, failureNode v1alpha1.ExecutableNode, failureNodeStatus v1alpha1.ExecutableNodeStatus, originalError *core.ExecutionError) NodeLookup {
 	return FailureNodeLookup{
 		NodeLookup:        nodeLookup,
 		FailureNode:       failureNode,
 		FailureNodeStatus: failureNodeStatus,
+		OriginalError:     originalError,
 	}
 }

flytepropeller/pkg/controller/executors/failure_node_lookup_test.go

@@ -6,6 +6,7 @@ import (
 
 	"github.com/stretchr/testify/assert"
 
+	"github.com/flyteorg/flyte/flyteidl/gen/pb-go/flyteidl/core"
 	"github.com/flyteorg/flyte/flytepropeller/pkg/apis/flyteworkflow/v1alpha1"
 	"github.com/flyteorg/flyte/flytepropeller/pkg/apis/flyteworkflow/v1alpha1/mocks"
 )
@@ -26,7 +27,10 @@ func TestNewFailureNodeLookup(t *testing.T) {
 	nl := nl{}
 	en := en{}
 	ns := ns{}
-	nodeLoopUp := NewFailureNodeLookup(nl, en, ns)
+	execErr := &core.ExecutionError{
+		Message: "node failure",
+	}
+	nodeLoopUp := NewFailureNodeLookup(nl, en, ns, execErr)
 	assert.NotNil(t, nl)
 	typed := nodeLoopUp.(FailureNodeLookup)
 	assert.Equal(t, nl, typed.NodeLookup)
@@ -38,14 +42,17 @@ func TestNewTestFailureNodeLookup(t *testing.T) {
 	n := &mocks.ExecutableNode{}
 	ns := &mocks.ExecutableNodeStatus{}
 	failureNodeID := "fn1"
+	originalErr := &core.ExecutionError{
+		Message: "node failure",
+	}
 	nl := NewTestNodeLookup(
 		map[string]v1alpha1.ExecutableNode{v1alpha1.StartNodeID: n, failureNodeID: n},
 		map[string]v1alpha1.ExecutableNodeStatus{v1alpha1.StartNodeID: ns, failureNodeID: ns},
 	)
 
 	assert.NotNil(t, nl)
 
-	failureNodeLookup := NewFailureNodeLookup(nl, n, ns)
+	failureNodeLookup := NewFailureNodeLookup(nl, n, ns, originalErr).(FailureNodeLookup)
 	r, ok := failureNodeLookup.GetNode(v1alpha1.StartNodeID)
 	assert.True(t, ok)
 	assert.Equal(t, n, r)
@@ -64,4 +71,9 @@ func TestNewTestFailureNodeLookup(t *testing.T) {
 	nodeIDs, err = failureNodeLookup.FromNode(failureNodeID)
 	assert.Nil(t, nodeIDs)
 	assert.Nil(t, err)
+
+	oe, err := failureNodeLookup.GetOriginalError()
+	assert.NotNil(t, oe)
+	assert.Equal(t, originalErr, oe)
+	assert.Nil(t, err)
 }

flytepropeller/pkg/controller/nodes/executor.go

@@ -765,6 +765,16 @@
 			}
 
 			if nodeInputs != nil {
+				// Resolve error input if current node is an on failure node
+				failureNodeLookup, ok := nCtx.ContextualNodeLookup().(executors.FailureNodeLookup)
+				if ok {
+					originalErr, err := failureNodeLookup.GetOriginalError()
+					if err != nil {
+						return handler.PhaseInfoFailure(core.ExecutionError_SYSTEM, "FailureNodeError", err.Error(), nil), nil
+					} else if originalErr != nil {
+						ResolveOnFailureNodeInput(ctx, nodeInputs, node.GetID(), originalErr)
+					}
+				}
 				p := common.CheckOffloadingCompat(ctx, nCtx, nodeInputs.GetLiterals(), node, c.literalOffloadingConfig)
 				if p != nil {
 					return *p, nil

flytepropeller/pkg/controller/nodes/resolve.go

@@ -105,3 +105,39 @@ func Resolve(ctx context.Context, outputResolver OutputResolver, nl executors.No
 		Literals: literalMap,
 	}, nil
 }
+
+func ResolveOnFailureNodeInput(ctx context.Context, nodeInputs *core.LiteralMap, nodeID v1alpha1.NodeID, execErr *core.ExecutionError) {
+	literals := nodeInputs.GetLiterals()
+	if literal, exists := literals["err"]; exists {
+		// make new Scalar for literal map
+		errorUnion := &core.Scalar_Union{
+			Union: &core.Union{
+				Value: &core.Literal{
+					Value: &core.Literal_Scalar{
+						Scalar: &core.Scalar{
+							Value: &core.Scalar_Error{
+								Error: &core.Error{
+									Message:      execErr.GetMessage(),
+									FailedNodeId: nodeID,
+								},
+							},
+						},
+					},
+				},
+				Type: &core.LiteralType{
+					Type: &core.LiteralType_Simple{
+						Simple: core.SimpleType_ERROR,
+					},
+					Structure: &core.TypeStructure{
+						Tag: "FlyteError",
+					},
+				},
+			},
+		}
+		literal.Value = &core.Literal_Scalar{
+			Scalar: &core.Scalar{
+				Value: errorUnion,
+			},
+		}
+	}
+}

flytepropeller/pkg/controller/nodes/resolve_test.go

@@ -467,3 +467,66 @@ func TestResolve(t *testing.T) {
 	})
 
 }
+
+func TestResolveOnFailureNodeInput(t *testing.T) {
+	ctx := context.Background()
+	t.Run("ResolveErrorInputs", func(t *testing.T) {
+		noneLiteral, _ := coreutils.MakeLiteral(nil)
+		inputLiterals := make(map[string]*core.Literal, 1)
+		inputLiterals["err"] = &core.Literal{
+			Value: &core.Literal_Scalar{
+				Scalar: &core.Scalar{
+					Value: &core.Scalar_Union{
+						Union: &core.Union{
+							Value: noneLiteral,
+							Type: &core.LiteralType{
+								Type: &core.LiteralType_Simple{
+									Simple: core.SimpleType_NONE,
+								},
+								Structure: &core.TypeStructure{
+									Tag: "none",
+								},
+							},
+						},
+					},
+				},
+			},
+		}
+		inputLiteralMap := &core.LiteralMap{
+			Literals: inputLiterals,
+		}
+		nID := "fn"
+		execErr := &core.ExecutionError{
+			Message: "node failure",
+		}
+		expectedLiterals := make(map[string]*core.Literal, 1)
+		errorLiteral, err := coreutils.MakeLiteral(&core.Error{Message: execErr.GetMessage(), FailedNodeId: nID})
+		assert.NoError(t, err)
+		expectedLiterals["err"] = &core.Literal{
+			Value: &core.Literal_Scalar{
+				Scalar: &core.Scalar{
+					Value: &core.Scalar_Union{
+						Union: &core.Union{
+							Value: errorLiteral,
+							Type: &core.LiteralType{
+								Type: &core.LiteralType_Simple{
+									Simple: core.SimpleType_ERROR,
+								},
+								Structure: &core.TypeStructure{
+									Tag: "FlyteError",
+								},
+							},
+						},
+					},
+				},
+			},
+		}
+		expectedLiteralMap := &core.LiteralMap{
+			Literals: expectedLiterals,
+		}
+		// Execute resolve
+		ResolveOnFailureNodeInput(ctx, inputLiteralMap, nID, execErr)
+		flyteassert.EqualLiteralMap(t, expectedLiteralMap, inputLiteralMap)
+	})
+
+}

flytepropeller/pkg/controller/nodes/subworkflow/subworkflow.go

@@ -169,7 +169,7 @@
 		status := nCtx.NodeStatus()
 		subworkflowNodeLookup := executors.NewNodeLookup(subworkflow, status, subworkflow)
 		failureNodeStatus := status.GetNodeExecutionStatus(ctx, failureNode.GetID())
-		failureNodeLookup := executors.NewFailureNodeLookup(subworkflowNodeLookup, failureNode, failureNodeStatus)
+		failureNodeLookup := executors.NewFailureNodeLookup(subworkflowNodeLookup, failureNode, failureNodeStatus, originalError)
 
 		state, err := s.nodeExecutor.RecursiveNodeHandler(ctx, execContext, failureNodeLookup, failureNodeLookup, failureNode)
 		if err != nil {

flytepropeller/pkg/controller/workflow/executor.go

@@ -203,7 +203,7 @@ func (c *workflowExecutor) handleFailureNode(ctx context.Context, w *v1alpha1.Fl
 	execcontext := executors.NewExecutionContext(w, w, w, nil, executors.InitializeControlFlow())
 
 	failureNodeStatus := w.GetExecutionStatus().GetNodeExecutionStatus(ctx, failureNode.GetID())
-	failureNodeLookup := executors.NewFailureNodeLookup(w, failureNode, failureNodeStatus)
+	failureNodeLookup := executors.NewFailureNodeLookup(w, failureNode, failureNodeStatus, execErr)
 	state, handlerErr := c.nodeExecutor.RecursiveNodeHandler(ctx, execcontext, failureNodeLookup, failureNodeLookup, failureNode)
 	c.updateExecutionStats(ctx, execcontext)
 

flytepropeller/pkg/utils/assert/literals.go

@@ -9,6 +9,24 @@
 	"github.com/flyteorg/flyte/flyteidl/gen/pb-go/flyteidl/core"
 )
 
+func EqualLiteralType(t *testing.T, lt1 *core.LiteralType, lt2 *core.LiteralType) {
+	if !assert.Equal(t, lt1 == nil, lt2 == nil) {
+		assert.FailNow(t, "One of the values is nil")
+	}
+	assert.Equal(t, reflect.TypeOf(lt1.GetType()), reflect.TypeOf(lt2.GetType()))
+	switch lt1.GetType().(type) {
+	case *core.LiteralType_Simple:
+		assert.Equal(t, lt1.GetType().(*core.LiteralType_Simple).Simple, lt2.GetType().(*core.LiteralType_Simple).Simple)
+	default:
+		assert.FailNow(t, "Not yet implemented for types %v", reflect.TypeOf(lt1.GetType()))
+	}
+	structure1 := lt1.GetStructure()
+	structure2 := lt2.GetStructure()
+	if structure1 != nil && structure2 != nil {
+		assert.Equal(t, structure1.GetTag(), structure2.GetTag())
+	}
+}
+
 func EqualPrimitive(t *testing.T, p1 *core.Primitive, p2 *core.Primitive) {
 	if !assert.Equal(t, p1 == nil, p2 == nil) {
 		assert.FailNow(t, "One of the values is nil")
@@ -27,6 +45,23 @@
 	}
 }
 
+func EqualError(t *testing.T, e1 *core.Error, e2 *core.Error) {
+	if !assert.Equal(t, e1 == nil, e2 == nil) {
+		assert.FailNow(t, "One of the values is nil")
+	}
+	assert.Equal(t, e1.GetMessage(), e2.GetMessage())
+	assert.Equal(t, e1.GetFailedNodeId(), e2.GetFailedNodeId())
+}
+
+func EqualUnion(t *testing.T, u1 *core.Union, u2 *core.Union) {
+	if !assert.Equal(t, u1 == nil, u2 == nil) {
+		assert.FailNow(t, "One of the values is nil")
+	}
+	assert.Equal(t, reflect.TypeOf(u1.GetValue()), reflect.TypeOf(u2.GetValue()))
+	EqualLiterals(t, u1.GetValue(), u2.GetValue())
+	EqualLiteralType(t, u1.GetType(), u2.GetType())
+}
+
 func EqualScalar(t *testing.T, p1 *core.Scalar, p2 *core.Scalar) {
 	if !assert.Equal(t, p1 == nil, p2 == nil) {
 		assert.FailNow(t, "One of the values is nil")
@@ -38,6 +73,10 @@
 	switch p1.GetValue().(type) {
 	case *core.Scalar_Primitive:
 		EqualPrimitive(t, p1.GetPrimitive(), p2.GetPrimitive())
+	case *core.Scalar_Error:
+		EqualError(t, p1.GetError(), p2.GetError())
+	case *core.Scalar_Union:
+		EqualUnion(t, p1.GetUnion(), p2.GetUnion())
 	default:
 		assert.FailNow(t, "Not yet implemented for types %v", reflect.TypeOf(p1.GetValue()))
 	}