fix(actions): refresh reusable caller status when children are skipped (#39589)

services/actions/job_emitter.go

@@ -362,15 +362,50 @@ func checkJobsOfCurrentRunAttempt(ctx context.Context, run *actions_model.Action
 	}
 
 	result.UpdatedJobs = append(result.UpdatedJobs, resolver.matrixUpdatedJobs...)
-	// Caller and matrix expansion insert Pending or Blocked jobs and a deferred gate leaves a job Blocked, only a follow-up pass resolves them.
+	hasFinishedCaller, err := reloadCallersFinishedByChildren(ctx, run.ID, resolver.jobMap, result.UpdatedJobs)
+	if err != nil {
+		return nil, err
+	}
+	// Only a follow-up pass resolves:
+	//   - the children a caller expansion inserted, or the dependents of a caller that failed to expand
+	//   - the siblings a matrix expansion inserted and the expanded job's dependents, or the dependents of a placeholder that failed to expand
+	//   - a job the deferred gate left Blocked
+	//   - the dependents of a caller finished by its children, which was still unfinished to the resolver
 	// Like the caller's children, matrix siblings are left out of result.Jobs and picked up there.
-	if expandedAnyCaller || resolver.matrixChanged || resolver.gateDeferred {
+	if expandedAnyCaller || resolver.matrixChanged || resolver.gateDeferred || hasFinishedCaller {
 		result.RunIDsToReEmit = append(result.RunIDsToReEmit, run.ID)
 	}
 	result.CancelledJobs = append(result.CancelledJobs, resolver.cancelledJobs...)
 	return result, nil
 }
 
+// reloadCallersFinishedByChildren reloads the callers the children's cascade finished in this pass,
+// so this pass's commit statuses and run notification see the callers finished.
+// They are kept out of result.UpdatedJobs, so no workflow_job webhook is sent for them, as for callers finished through a runner.
+func reloadCallersFinishedByChildren(ctx context.Context, runID int64, jobMap map[int64]*actions_model.ActionRunJob, updatedJobs []*actions_model.ActionRunJob) (bool, error) {
+	hasFinishedCaller := false
+	checkedCallers := make(container.Set[int64])
+	finishedJobs := slices.Clone(updatedJobs)
+	for i := 0; i < len(finishedJobs); i++ {
+		child := finishedJobs[i]
+		caller := jobMap[child.ParentJobID]
+		if caller == nil || !child.Status.IsDone() || caller.Status.IsDone() || !checkedCallers.Add(caller.ID) {
+			continue
+		}
+		freshCaller, err := actions_model.GetRunJobByRunAndID(ctx, runID, caller.ID)
+		if err != nil {
+			return false, fmt.Errorf("reloadCallersFinishedByChildren: reload caller %d: %w", caller.ID, err)
+		}
+		if !freshCaller.Status.IsDone() {
+			continue
+		}
+		caller.Status, caller.Started, caller.Stopped = freshCaller.Status, freshCaller.Started, freshCaller.Stopped
+		finishedJobs = append(finishedJobs, caller)
+		hasFinishedCaller = true
+	}
+	return hasFinishedCaller, nil
+}
+
 func cancelFailedMatrixSiblings(ctx context.Context, jobs actions_model.ActionJobList) ([]*actions_model.ActionRunJob, error) {
 	var toCancel []*actions_model.ActionRunJob
 	for _, failed := range jobs {

services/actions/job_emitter_test.go

@@ -5,6 +5,7 @@ package actions
 
 import (
 	"fmt"
+	"slices"
 	"testing"
 
 	actions_model "gitea.dev/models/actions"
@@ -670,6 +671,65 @@ func Test_checkJobsOfCurrentRunAttempt_SkippedCallerIsUpdated(t *testing.T) {
 	require.Len(t, result.UpdatedJobs, 1)
 	assert.Equal(t, caller.ID, result.UpdatedJobs[0].ID)
 	assert.Equal(t, actions_model.StatusSkipped, unittest.AssertExistsAndLoadBean(t, &actions_model.ActionRunJob{ID: caller.ID}).Status)
+
+	// A child skipped in this pass finishes its caller and the outer caller, both are reported Success and the outer one's dependent is resolved by the re-emit.
+	run2 := &actions_model.ActionRun{
+		RepoID: 4, OwnerID: 1, TriggerUserID: 1,
+		WorkflowID: "test.yml", Index: 9915, Ref: "refs/heads/main", Status: actions_model.StatusRunning,
+	}
+	assert.NoError(t, db.Insert(ctx, run2))
+	attempt2 := &actions_model.ActionRunAttempt{RepoID: 4, RunID: run2.ID, Attempt: 1, Status: actions_model.StatusRunning}
+	assert.NoError(t, db.Insert(ctx, attempt2))
+	_, err = db.Exec(ctx, "UPDATE `action_run` SET latest_attempt_id = ? WHERE id = ?", attempt2.ID, run2.ID)
+	assert.NoError(t, err)
+	run2.LatestAttemptID = attempt2.ID
+	caller2 := &actions_model.ActionRunJob{
+		RunID: run2.ID, RunAttemptID: attempt2.ID, RepoID: 4, OwnerID: 1,
+		JobID: "deploy", Name: "deploy", Status: actions_model.StatusRunning, IsReusableCaller: true, IsExpanded: true,
+		WorkflowPayload:         []byte("jobs: {deploy: {uses: ./.gitea/workflows/mid.yml}}"),
+		ReusableWorkflowContent: []byte("on: {workflow_call: {}}\njobs: {inner: {uses: ./.gitea/workflows/leaf.yml}}"),
+	}
+	assert.NoError(t, db.Insert(ctx, caller2))
+	inner := &actions_model.ActionRunJob{
+		RunID: run2.ID, RunAttemptID: attempt2.ID, RepoID: 4, OwnerID: 1, ParentJobID: caller2.ID,
+		JobID: "inner", Name: "inner", Status: actions_model.StatusRunning, IsReusableCaller: true, IsExpanded: true,
+		WorkflowPayload: []byte("jobs: {inner: {uses: ./.gitea/workflows/leaf.yml}}"),
+	}
+	assert.NoError(t, db.Insert(ctx, inner))
+	assert.NoError(t, db.Insert(ctx, &actions_model.ActionRunJob{
+		RunID: run2.ID, RunAttemptID: attempt2.ID, RepoID: 4, OwnerID: 1, ParentJobID: inner.ID,
+		JobID: "work", Status: actions_model.StatusSuccess, WorkflowPayload: minimalWorkflowPayload("work"),
+	}))
+	assert.NoError(t, db.Insert(ctx, &actions_model.ActionRunJob{
+		RunID: run2.ID, RunAttemptID: attempt2.ID, RepoID: 4, OwnerID: 1, ParentJobID: inner.ID,
+		JobID: "alert", Status: actions_model.StatusBlocked, Needs: []string{"work"},
+		WorkflowPayload: []byte("jobs: {alert: {if: false, needs: [work], runs-on: ubuntu-latest, steps: [{run: echo}]}}"),
+	}))
+	after := &actions_model.ActionRunJob{
+		RunID: run2.ID, RunAttemptID: attempt2.ID, RepoID: 4, OwnerID: 1,
+		JobID: "after", Status: actions_model.StatusPending, Needs: []string{"deploy"},
+		WorkflowPayload: []byte("jobs: {after: {if: \"github.ref == 'refs/heads/main'\", needs: [deploy], runs-on: ubuntu-latest, steps: [{run: echo}]}}"),
+	}
+	assert.NoError(t, db.Insert(ctx, after))
+
+	result, err = checkJobsOfCurrentRunAttempt(ctx, run2)
+	assert.NoError(t, err)
+	assert.Contains(t, result.RunIDsToReEmit, run2.ID)
+	assert.Equal(t, actions_model.StatusPending, unittest.AssertExistsAndLoadBean(t, &actions_model.ActionRunJob{ID: after.ID}).Status)
+	updatedJobIDs := make([]string, 0, len(result.UpdatedJobs))
+	for _, job := range result.UpdatedJobs {
+		updatedJobIDs = append(updatedJobIDs, job.JobID)
+	}
+	assert.Equal(t, []string{"alert"}, updatedJobIDs)
+	for _, callerID := range []int64{inner.ID, caller2.ID} {
+		idx := slices.IndexFunc(result.Jobs, func(job *actions_model.ActionRunJob) bool { return job.ID == callerID })
+		require.NotEqual(t, -1, idx)
+		assert.Equal(t, actions_model.StatusSuccess, result.Jobs[idx].Status)
+	}
+
+	_, err = checkJobsOfCurrentRunAttempt(ctx, run2)
+	assert.NoError(t, err)
+	assert.Equal(t, actions_model.StatusWaiting, unittest.AssertExistsAndLoadBean(t, &actions_model.ActionRunJob{ID: after.ID}).Status)
 }
 
 // Test_checkRunConcurrency_HeldGroupDoesNotWake verifies that only an unoccupied concurrency group can wake up a blocked run/job.