tiny-forge/internal/scheduler/scheduler_test.go

package scheduler

import (
	"context"
	"testing"
	"time"

	"github.com/alexei/tinyforge/internal/store"
	"github.com/alexei/tinyforge/internal/workload/plugin"
)

// newTestStore opens an in-memory SQLite store. Each test gets its own
// DSN so parallel runs do not collide on shared cache databases.
func newTestStore(t *testing.T) *store.Store {
	t.Helper()
	st, err := store.New(":memory:")
	if err != nil {
		t.Fatalf("open store: %v", err)
	}
	t.Cleanup(func() { _ = st.Close() })
	return st
}

func seedScheduleTrigger(t *testing.T, st *store.Store, name, interval, lastFired string) store.Trigger {
	t.Helper()
	trg, err := st.CreateTrigger(store.Trigger{
		Kind:        "schedule",
		Name:        name,
		Config:      `{"interval":"` + interval + `"}`,
		LastFiredAt: lastFired,
	})
	if err != nil {
		t.Fatalf("CreateTrigger: %v", err)
	}
	return trg
}

func TestShouldFire(t *testing.T) {
	st := newTestStore(t)
	now := time.Date(2026, 5, 16, 12, 0, 0, 0, time.UTC)
	s := New(st, func(context.Context, store.Trigger, plugin.InboundEvent) error { return nil }, 0)

	cases := []struct {
		name      string
		interval  string
		lastFired string
		want      bool
	}{
		{"never fired fires", "1h", "", true},
		{"window not yet elapsed", "1h", now.Add(-30 * time.Minute).Format(time.RFC3339), false},
		{"window exactly elapsed fires", "1h", now.Add(-1 * time.Hour).Format(time.RFC3339), true},
		{"window long elapsed fires", "24h", now.Add(-48 * time.Hour).Format(time.RFC3339), true},
		{"bad interval suppressed", "banana", "", false},
		{"bad last_fired_at treated as never", "1h", "not-a-timestamp", true},
	}
	for _, tc := range cases {
		t.Run(tc.name, func(t *testing.T) {
			trg := store.Trigger{
				Config:      `{"interval":"` + tc.interval + `"}`,
				LastFiredAt: tc.lastFired,
			}
			got := s.shouldFire(trg, now)
			if got != tc.want {
				t.Fatalf("shouldFire = %v, want %v", got, tc.want)
			}
		})
	}
}

func TestTickOnce_FiresOverdueTriggers(t *testing.T) {
	st := newTestStore(t)
	now := time.Date(2026, 5, 16, 12, 0, 0, 0, time.UTC)

	// Three triggers: one overdue, one not yet due, one never-fired.
	overdue := seedScheduleTrigger(t, st, "overdue", "1h", now.Add(-2*time.Hour).Format(time.RFC3339))
	notDue := seedScheduleTrigger(t, st, "notdue", "1h", now.Add(-30*time.Minute).Format(time.RFC3339))
	never := seedScheduleTrigger(t, st, "never", "1h", "")

	fired := make(map[string]int)
	s := New(st, func(_ context.Context, trg store.Trigger, _ plugin.InboundEvent) error {
		fired[trg.Name]++
		return nil
	}, 0)
	s.clock = func() time.Time { return now }

	s.TickOnce(context.Background())

	if fired["overdue"] != 1 {
		t.Errorf("overdue should fire once, got %d", fired["overdue"])
	}
	if fired["notdue"] != 0 {
		t.Errorf("notdue should not fire, got %d", fired["notdue"])
	}
	if fired["never"] != 1 {
		t.Errorf("never should fire once on first tick, got %d", fired["never"])
	}

	// last_fired_at must advance for everyone we dispatched.
	for _, id := range []string{overdue.ID, never.ID} {
		row, err := st.GetTriggerByID(id)
		if err != nil {
			t.Fatalf("GetTriggerByID(%s): %v", id, err)
		}
		if row.LastFiredAt == "" {
			t.Errorf("last_fired_at not persisted for %s", row.Name)
		}
	}
	// not-due trigger's last_fired_at must NOT have changed.
	row, err := st.GetTriggerByID(notDue.ID)
	if err != nil {
		t.Fatalf("GetTriggerByID(notdue): %v", err)
	}
	if row.LastFiredAt != notDue.LastFiredAt {
		t.Errorf("notdue last_fired_at changed: was %q now %q", notDue.LastFiredAt, row.LastFiredAt)
	}
}

func TestTickOnce_DispatchErrorDoesNotWedgeOthers(t *testing.T) {
	st := newTestStore(t)
	now := time.Date(2026, 5, 16, 12, 0, 0, 0, time.UTC)

	broken := seedScheduleTrigger(t, st, "broken", "1h", "")
	seedScheduleTrigger(t, st, "healthy", "1h", "")

	fired := map[string]int{}
	s := New(st, func(_ context.Context, trg store.Trigger, _ plugin.InboundEvent) error {
		fired[trg.Name]++
		if trg.Name == "broken" {
			return context.Canceled
		}
		return nil
	}, 0)
	s.clock = func() time.Time { return now }

	s.TickOnce(context.Background())

	if fired["broken"] != 1 {
		t.Errorf("broken should be attempted once, got %d", fired["broken"])
	}
	if fired["healthy"] != 1 {
		t.Errorf("healthy should fire once, got %d", fired["healthy"])
	}

	// Core persist-before-dispatch invariant: even though the broken
	// trigger's dispatcher returned an error, last_fired_at must have
	// advanced. Otherwise the scheduler would re-fire it on every tick.
	row, err := st.GetTriggerByID(broken.ID)
	if err != nil {
		t.Fatalf("GetTriggerByID(broken): %v", err)
	}
	if row.LastFiredAt == "" {
		t.Fatalf("broken trigger last_fired_at must advance even on dispatch error")
	}

	// And: a second TickOnce at the same `now` must not re-fire broken.
	s.TickOnce(context.Background())
	if fired["broken"] != 1 {
		t.Errorf("broken refired after persist; got %d (want 1)", fired["broken"])
	}
}

func TestTickOnce_PersistsLastFiredBeforeDispatch(t *testing.T) {
	// Documented behavior: last_fired_at is persisted before the
	// dispatcher runs so a panicking match cannot wedge a tight loop.
	st := newTestStore(t)
	now := time.Date(2026, 5, 16, 12, 0, 0, 0, time.UTC)
	trg := seedScheduleTrigger(t, st, "tick", "1h", "")

	dispatched := false
	s := New(st, func(_ context.Context, t store.Trigger, _ plugin.InboundEvent) error {
		// At dispatch time the column must already be set.
		row, err := st.GetTriggerByID(t.ID)
		if err != nil {
			return err
		}
		dispatched = row.LastFiredAt != ""
		return nil
	}, 0)
	s.clock = func() time.Time { return now }

	s.TickOnce(context.Background())

	if !dispatched {
		t.Fatalf("last_fired_at must be persisted before dispatcher runs")
	}
	row, err := st.GetTriggerByID(trg.ID)
	if err != nil {
		t.Fatalf("get: %v", err)
	}
	if row.LastFiredAt != now.Format(time.RFC3339) {
		t.Errorf("last_fired_at = %q, want %q", row.LastFiredAt, now.Format(time.RFC3339))
	}
}

func TestLifecycle_StartStopIdempotent(t *testing.T) {
	// Start + Stop are wrapped in sync.Once. A second call must be a
	// no-op (no panic on double-cancel, no goroutine leak from double-
	// Start). This guards the shutdown path that runs Stop from both
	// defer and the signal-handler block in cmd/server/main.go.
	st := newTestStore(t)
	noop := func(context.Context, store.Trigger, plugin.InboundEvent) error { return nil }
	s := New(st, noop, 100*time.Millisecond)

	s.Start(context.Background())
	s.Start(context.Background()) // second call: no goroutine spawned

	s.Stop()
	s.Stop() // second call: no panic on closing already-cancelled context
}

func TestNew_ClampsInterval(t *testing.T) {
	st := newTestStore(t)
	noop := func(context.Context, store.Trigger, plugin.InboundEvent) error { return nil }
	if got := New(st, noop, 0).tickInterval; got != 30*time.Second {
		t.Errorf("default = %s, want 30s", got)
	}
	if got := New(st, noop, 1*time.Hour).tickInterval; got != 5*time.Minute {
		t.Errorf("clamped = %s, want 5m", got)
	}
	if got := New(st, noop, 2*time.Minute).tickInterval; got != 2*time.Minute {
		t.Errorf("passthrough = %s, want 2m", got)
	}
}