orchestra/internal/operations/phase_request_test.go

package operations

import (
	"encoding/json"
	"errors"
	"testing"

	"orchestra/internal/domain"
	"orchestra/internal/registry"
	"orchestra/internal/store"
)

// epochOf returns the epoch of the task's current lease, which fences every
// worker-driven call.
func epochOf(t *testing.T, s *store.Store, id string) string {
	t.Helper()
	task, ok := s.Task(id)
	if !ok || task.Lease == nil {
		t.Fatal("task has no lease")
	}
	return task.Lease.Epoch
}

func TestRequestWorkPhaseAdvancesAndSealsEachArtifact(t *testing.T) {
	s, id := phaseStore(t)
	project := registry.Project{ID: "p"}
	lease(t, s, id)
	epoch := epochOf(t, s, id)

	// frame -> research seals nothing: framing produces no artifact.
	if _, err := RequestWorkPhase(s, project, id, epoch, "op-1", domain.WorkPhaseFrame, domain.WorkPhaseResearch, nil); err != nil {
		t.Fatal(err)
	}
	if got, _ := s.Task(id); got.WorkPhase != domain.WorkPhaseResearch {
		t.Fatalf("phase = %q", got.WorkPhase)
	}

	// research -> plan is refused until the research is sealed.
	if _, err := RequestWorkPhase(s, project, id, epoch, "op-2", domain.WorkPhaseResearch, domain.WorkPhasePlan, nil); !errors.Is(err, domain.ErrInvalid) {
		t.Fatalf("leaving research unsealed must fail, got %v", err)
	}
	if _, err := RequestWorkPhase(s, project, id, epoch, "op-3", domain.WorkPhaseResearch, domain.WorkPhasePlan, sealed(t, research)); err != nil {
		t.Fatal(err)
	}
	got, _ := s.Task(id)
	if got.WorkPhase != domain.WorkPhasePlan || got.ResearchRef == "" {
		t.Fatalf("task = %+v", got)
	}

	// plan -> implement is refused until the plan is sealed.
	if _, err := RequestWorkPhase(s, project, id, epoch, "op-4", domain.WorkPhasePlan, domain.WorkPhaseImplement, nil); !errors.Is(err, domain.ErrInvalid) {
		t.Fatalf("leaving plan unsealed must fail, got %v", err)
	}
	if _, err := RequestWorkPhase(s, project, id, epoch, "op-5", domain.WorkPhasePlan, domain.WorkPhaseImplement, sealed(t, plan)); err != nil {
		t.Fatal(err)
	}
	got, _ = s.Task(id)
	if got.WorkPhase != domain.WorkPhaseImplement || got.PlanRef == "" || got.ResearchRef == "" {
		t.Fatalf("task = %+v", got)
	}
}

func TestRequestWorkPhaseRefusesASkippedPhase(t *testing.T) {
	s, id := phaseStore(t)
	lease(t, s, id)
	// frame -> implement is a legal domain transition, but not the next step
	// on this project's declared path. The agent is refused rather than
	// silently corrected.
	_, err := RequestWorkPhase(s, registry.Project{ID: "p"}, id, epochOf(t, s, id), "op-1", domain.WorkPhaseFrame, domain.WorkPhaseImplement, nil)
	if !errors.Is(err, ErrPhaseRequest) {
		t.Fatalf("err = %v", err)
	}
	if got, _ := s.Task(id); got.WorkPhase != "" {
		t.Fatalf("phase moved to %q", got.WorkPhase)
	}
}

func TestRequestWorkPhaseRefusesAStalePhaseBelief(t *testing.T) {
	s, id := phaseStore(t)
	lease(t, s, id)
	epoch := epochOf(t, s, id)
	if _, err := RequestWorkPhase(s, registry.Project{ID: "p"}, id, epoch, "op-1", domain.WorkPhaseFrame, domain.WorkPhaseResearch, nil); err != nil {
		t.Fatal(err)
	}
	// The agent still believes it is framing. Acting on this would advance a
	// phase it never ran.
	_, err := RequestWorkPhase(s, registry.Project{ID: "p"}, id, epoch, "op-2", domain.WorkPhaseFrame, domain.WorkPhaseResearch, nil)
	if !errors.Is(err, ErrPhaseRequest) {
		t.Fatalf("err = %v", err)
	}
}

func TestRequestWorkPhaseRefusesAStaleLeaseEpoch(t *testing.T) {
	s, id := phaseStore(t)
	lease(t, s, id)
	_, err := RequestWorkPhase(s, registry.Project{ID: "p"}, id, "not-the-epoch", "op-1", domain.WorkPhaseFrame, domain.WorkPhaseResearch, nil)
	if !errors.Is(err, domain.ErrConflict) {
		t.Fatalf("err = %v", err)
	}
	if got, _ := s.Task(id); got.WorkPhase != "" {
		t.Fatalf("phase moved to %q", got.WorkPhase)
	}
}

func TestRequestWorkPhaseRequiresAnOperationID(t *testing.T) {
	s, id := phaseStore(t)
	lease(t, s, id)
	_, err := RequestWorkPhase(s, registry.Project{ID: "p"}, id, epochOf(t, s, id), "", domain.WorkPhaseFrame, domain.WorkPhaseResearch, nil)
	if !errors.Is(err, domain.ErrInvalid) {
		t.Fatalf("err = %v", err)
	}
}

// A lost response is the ordinary case, not the exotic one: the worker resends
// the same request and must not advance the phase a second time.
func TestRequestWorkPhaseIsIdempotentPerOperationID(t *testing.T) {
	s, id := phaseStore(t)
	project := registry.Project{ID: "p"}
	lease(t, s, id)
	epoch := epochOf(t, s, id)

	first, err := RequestWorkPhase(s, project, id, epoch, "op-1", domain.WorkPhaseFrame, domain.WorkPhaseResearch, nil)
	if err != nil {
		t.Fatal(err)
	}
	second, err := RequestWorkPhase(s, project, id, epoch, "op-1", domain.WorkPhaseFrame, domain.WorkPhaseResearch, nil)
	if err != nil {
		t.Fatal(err)
	}
	if second.ID != first.ID {
		t.Fatalf("redelivery produced a new event: %s then %s", first.ID, second.ID)
	}
	if got, _ := s.Task(id); got.WorkPhase != domain.WorkPhaseResearch {
		t.Fatalf("phase = %q", got.WorkPhase)
	}
	var changes int
	for _, e := range s.Events(0) {
		if e.TaskID == id && e.Type == domain.EventWorkPhaseChanged {
			changes++
		}
	}
	if changes != 1 {
		t.Fatalf("recorded %d phase changes, want 1", changes)
	}
}

// The operation id is what makes redelivery safe, so it has to survive into
// the event the next redelivery reads.
func TestRequestWorkPhaseRecordsTheOperationID(t *testing.T) {
	s, id := phaseStore(t)
	lease(t, s, id)
	e, err := RequestWorkPhase(s, registry.Project{ID: "p"}, id, epochOf(t, s, id), "op-1", domain.WorkPhaseFrame, domain.WorkPhaseResearch, nil)
	if err != nil {
		t.Fatal(err)
	}
	var p struct {
		OperationID string `json:"operation_id"`
		From        string `json:"from"`
		Phase       string `json:"phase"`
	}
	if err := json.Unmarshal(e.Payload, &p); err != nil {
		t.Fatal(err)
	}
	if p.OperationID != "op-1" || p.From != "frame" || p.Phase != "research" {
		t.Fatalf("payload = %+v", p)
	}
}