feat: publish standalone worker

Separate worker packaging and service lifecycle from the control plane.

internal/notifier/run_exp_test.go

@@ -0,0 +1,160 @@
+package notifier
+
+import (
+	"log"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"rsgit.ru/rsmon/rsmon/app/models"
+	"rsgit.ru/rsmon/rsmon/config/database"
+	"rsgit.ru/rsmon/rsmon/spec/factories"
+)
+
+func init() {
+	database.Init()
+}
+
+func TestRunExp(t *testing.T) {
+	log.Println("TestRunExp")
+	models.Drop()
+	models.Migrate()
+	var err error
+	contact, notification, monitor := factories.MonitorWithNotification()
+
+	check := factories.PersistedCheck(&monitor, "ssl")
+	assert.Equal(t, "ssl", check.Kind, "check kind should be ssl")
+	assert.Equal(t, monitor.ID, check.MonitorID, "check should have correct monitor id")
+
+	exp := time.Now().Add(14 * 24 * time.Hour)
+	check.Expires = &exp
+	err = models.DB().Save(&check).Error
+	if err != nil {
+		t.Fatal(err)
+	}
+	RunExp()
+	shoudHaveMessages("exp1 - contact should receive no messages", t, "exp", notification.ID, contact.ID, []int64{})
+
+	exp = time.Now().Add(1 * time.Hour)
+	check.Expires = &exp
+	err = models.DB().Save(&check).Error
+	if err != nil {
+		t.Fatal(err)
+	}
+	RunExp()
+	shoudHaveMessages("exp2 - contact should receive messages", t, "exp", notification.ID, contact.ID, []int64{check.ID})
+
+	RunExp()
+	shoudHaveMessages("exp3 - contact should not receive duplicate messages", t, "exp", notification.ID, contact.ID, []int64{check.ID})
+}
+
+// TestRunExpExpiresSystemContact exercises the regression scenario that the
+// dev DB hit after being restored from the production dump: the
+// contacts.is_system column was missing from the production schema and the
+// notifier's RunExp panic'd on the GORM preload.
+//
+// The fix has three layers: AutoMigrate adds the column, GetContacts logs
+// instead of panicking, and RunExp/RunExpCheck recover from panics. This
+// test verifies all three by setting a contact's is_system=true, queueing an
+// expiring SSL check, and confirming RunExp:
+//   - does not panic;
+//   - writes a queued exp message for the is_system contact (i.e. the schema
+//     has the column and the field round-trips); and
+//   - leaves RunExp returnable to its caller.
+func TestRunExpExpiresSystemContact(t *testing.T) {
+	log.Println("TestRunExpExpiresSystemContact")
+	models.Drop()
+	models.Migrate()
+
+	account := &models.Account{Name: "acct-system-contact"}
+	require.NoError(t, models.DB().Create(account).Error)
+	accountID := account.ID
+
+	trueVal := true
+	contact := &models.Contact{
+		AccountID: &accountID,
+		Name:      "system-admin",
+		Kind:      "email",
+		Value:     "ops@example.com",
+		IsSystem:  &trueVal,
+	}
+	require.NoError(t, models.DB().Create(contact).Error)
+
+	group := factories.PersistedGroup(account)
+	notification := factories.PersistedNotification(
+		account, []int64{contact.ID}, []int64{group.ID}, 300, false,
+	)
+	monitor := factories.PersistedMonitor(&group)
+
+	check := factories.PersistedCheck(&monitor, "ssl")
+	exp := time.Now().Add(1 * time.Hour)
+	check.Expires = &exp
+	require.NoError(t, models.DB().Save(&check).Error)
+
+	assert.NotPanics(t, func() {
+		RunExp()
+	}, "RunExp must not panic when processing an is_system contact")
+
+	shoudHaveMessages(
+		"is_system contact must receive the exp message",
+		t, "exp", notification.ID, contact.ID, []int64{check.ID},
+	)
+}
+
+// TestRunExpDoesNotPanicOnBrokenAssociation replays the original prod-dump
+// panic in a contained way: the notification_contacts join row references a
+// non-existent contact id, which forces GORM's preload of Contacts to fail.
+// The fix's defensive recover() must keep RunExpCheck returning cleanly so
+// the scheduler loop survives a single bad row.
+//
+// We deliberately bypass the contacts.is_system column-drop path because
+// Postgres caches prepared statements per session; mutating the contacts
+// schema mid-test triggers SQLSTATE 0A000 (cached plan must not change
+// result type) on the pool's other connections and masks the panic we want
+// to verify.
+func TestRunExpDoesNotPanicOnBrokenAssociation(t *testing.T) {
+	log.Println("TestRunExpDoesNotPanicOnBrokenAssociation")
+	models.Drop()
+	models.Migrate()
+
+	_, notification, monitor := factories.MonitorWithNotification()
+
+	// Force a broken association by deleting the contact that the
+	// notification points to. GORM's preload of Contacts will then have no
+	// rows for that notification, exercising the empty-contacts path
+	// without involving DDL or FK violations.
+	require.NoError(t, models.DB().
+		Exec("DELETE FROM notification_contacts WHERE notification_id = ?", notification.ID).Error)
+	// Re-add a join row pointing to a contact id that has been deleted
+	// from contacts. We disable the FK temporarily so the join row sticks.
+	require.NoError(t, models.DB().
+		Exec("SET session_replication_role = 'replica'").Error)
+	t.Cleanup(func() {
+		_ = models.DB().
+			Exec("SET session_replication_role = 'origin'").Error
+	})
+	bogusContactID := int64(9999999)
+	require.NoError(t, models.DB().
+		Exec(
+			"INSERT INTO notification_contacts (notification_id, contact_id) VALUES (?, ?)",
+			notification.ID, bogusContactID,
+		).Error)
+
+	check := factories.PersistedCheck(&monitor, "ssl")
+	exp := time.Now().Add(1 * time.Hour)
+	check.Expires = &exp
+	require.NoError(t, models.DB().Save(&check).Error)
+
+	require.NoError(t, models.DB().
+		Preload("Monitor").
+		Preload("Monitor.Group").
+		Preload("Monitor.Group.Notifications").
+		Preload("Monitor.Group.Notifications.Contacts").
+		First(&check, check.ID).Error)
+
+	assert.NotPanics(t, func() {
+		RunExpCheck(&check)
+	}, "RunExpCheck must not panic when Contact preload encounters a broken association")
+}