Cleanup stale device lists for users we don't share a room with anymore (#2857)

The stale device lists table might contain entries for users we don't
share a room with anymore. This now asks the roomserver about left users
and removes those entries from the table.

Co-authored-by: Neil Alexander <neilalexander@users.noreply.github.com>

keyserver/internal/device_list_update.go

@@ -24,6 +24,8 @@ import (
 	"sync"
 	"time"
 
+	rsapi "github.com/matrix-org/dendrite/roomserver/api"
+
 	"github.com/matrix-org/gomatrix"
 	"github.com/matrix-org/gomatrixserverlib"
 	"github.com/matrix-org/util"
@@ -102,6 +104,7 @@ type DeviceListUpdater struct {
 	// block on or timeout via a select.
 	userIDToChan   map[string]chan bool
 	userIDToChanMu *sync.Mutex
+	rsAPI          rsapi.KeyserverRoomserverAPI
 }
 
 // DeviceListUpdaterDatabase is the subset of functionality from storage.Database required for the updater.
@@ -124,6 +127,8 @@ type DeviceListUpdaterDatabase interface {
 
 	// DeviceKeysJSON populates the KeyJSON for the given keys. If any proided `keys` have a `KeyJSON` or `StreamID` already then it will be replaced.
 	DeviceKeysJSON(ctx context.Context, keys []api.DeviceMessage) error
+
+	DeleteStaleDeviceLists(ctx context.Context, userIDs []string) error
 }
 
 type DeviceListUpdaterAPI interface {
@@ -140,7 +145,7 @@ func NewDeviceListUpdater(
 	process *process.ProcessContext, db DeviceListUpdaterDatabase,
 	api DeviceListUpdaterAPI, producer KeyChangeProducer,
 	fedClient fedsenderapi.KeyserverFederationAPI, numWorkers int,
-	thisServer gomatrixserverlib.ServerName,
+	rsAPI rsapi.KeyserverRoomserverAPI, thisServer gomatrixserverlib.ServerName,
 ) *DeviceListUpdater {
 	return &DeviceListUpdater{
 		process:        process,
@@ -154,6 +159,7 @@ func NewDeviceListUpdater(
 		workerChans:    make([]chan gomatrixserverlib.ServerName, numWorkers),
 		userIDToChan:   make(map[string]chan bool),
 		userIDToChanMu: &sync.Mutex{},
+		rsAPI:          rsAPI,
 	}
 }
 
@@ -168,7 +174,7 @@ func (u *DeviceListUpdater) Start() error {
 		go u.worker(ch)
 	}
 
-	staleLists, err := u.db.StaleDeviceLists(context.Background(), []gomatrixserverlib.ServerName{})
+	staleLists, err := u.db.StaleDeviceLists(u.process.Context(), []gomatrixserverlib.ServerName{})
 	if err != nil {
 		return err
 	}
@@ -186,6 +192,25 @@ func (u *DeviceListUpdater) Start() error {
 	return nil
 }
 
+// CleanUp removes stale device entries for users we don't share a room with anymore
+func (u *DeviceListUpdater) CleanUp() error {
+	staleUsers, err := u.db.StaleDeviceLists(u.process.Context(), []gomatrixserverlib.ServerName{})
+	if err != nil {
+		return err
+	}
+
+	res := rsapi.QueryLeftUsersResponse{}
+	if err = u.rsAPI.QueryLeftUsers(u.process.Context(), &rsapi.QueryLeftUsersRequest{StaleDeviceListUsers: staleUsers}, &res); err != nil {
+		return err
+	}
+
+	if len(res.LeftUsers) == 0 {
+		return nil
+	}
+	logrus.Debugf("Deleting %d stale device list entries", len(res.LeftUsers))
+	return u.db.DeleteStaleDeviceLists(u.process.Context(), res.LeftUsers)
+}
+
 func (u *DeviceListUpdater) mutex(userID string) *sync.Mutex {
 	u.mu.Lock()
 	defer u.mu.Unlock()

keyserver/internal/device_list_update_test.go

@@ -30,7 +30,12 @@ import (
 	"github.com/matrix-org/gomatrixserverlib"
 
 	"github.com/matrix-org/dendrite/keyserver/api"
+	"github.com/matrix-org/dendrite/keyserver/storage"
+	roomserver "github.com/matrix-org/dendrite/roomserver/api"
+	"github.com/matrix-org/dendrite/setup/config"
 	"github.com/matrix-org/dendrite/setup/process"
+	"github.com/matrix-org/dendrite/test"
+	"github.com/matrix-org/dendrite/test/testrig"
 )
 
 var (
@@ -53,6 +58,10 @@ type mockDeviceListUpdaterDatabase struct {
 	mu           sync.Mutex // protect staleUsers
 }
 
+func (d *mockDeviceListUpdaterDatabase) DeleteStaleDeviceLists(ctx context.Context, userIDs []string) error {
+	return nil
+}
+
 // StaleDeviceLists returns a list of user IDs ending with the domains provided who have stale device lists.
 // If no domains are given, all user IDs with stale device lists are returned.
 func (d *mockDeviceListUpdaterDatabase) StaleDeviceLists(ctx context.Context, domains []gomatrixserverlib.ServerName) ([]string, error) {
@@ -153,7 +162,7 @@ func TestUpdateHavePrevID(t *testing.T) {
 	}
 	ap := &mockDeviceListUpdaterAPI{}
 	producer := &mockKeyChangeProducer{}
-	updater := NewDeviceListUpdater(process.NewProcessContext(), db, ap, producer, nil, 1, "localhost")
+	updater := NewDeviceListUpdater(process.NewProcessContext(), db, ap, producer, nil, 1, nil, "localhost")
 	event := gomatrixserverlib.DeviceListUpdateEvent{
 		DeviceDisplayName: "Foo Bar",
 		Deleted:           false,
@@ -225,7 +234,7 @@ func TestUpdateNoPrevID(t *testing.T) {
 			`)),
 		}, nil
 	})
-	updater := NewDeviceListUpdater(process.NewProcessContext(), db, ap, producer, fedClient, 2, "example.test")
+	updater := NewDeviceListUpdater(process.NewProcessContext(), db, ap, producer, fedClient, 2, nil, "example.test")
 	if err := updater.Start(); err != nil {
 		t.Fatalf("failed to start updater: %s", err)
 	}
@@ -239,6 +248,7 @@ func TestUpdateNoPrevID(t *testing.T) {
 		UserID:            remoteUserID,
 	}
 	err := updater.Update(ctx, event)
+
 	if err != nil {
 		t.Fatalf("Update returned an error: %s", err)
 	}
@@ -294,7 +304,7 @@ func TestDebounce(t *testing.T) {
 		close(incomingFedReq)
 		return <-fedCh, nil
 	})
-	updater := NewDeviceListUpdater(process.NewProcessContext(), db, ap, producer, fedClient, 1, "localhost")
+	updater := NewDeviceListUpdater(process.NewProcessContext(), db, ap, producer, fedClient, 1, nil, "localhost")
 	if err := updater.Start(); err != nil {
 		t.Fatalf("failed to start updater: %s", err)
 	}
@@ -349,3 +359,73 @@ func TestDebounce(t *testing.T) {
 		t.Errorf("user %s is marked as stale", userID)
 	}
 }
+
+func mustCreateKeyserverDB(t *testing.T, dbType test.DBType) (storage.Database, func()) {
+	t.Helper()
+
+	base, _, _ := testrig.Base(nil)
+	connStr, clearDB := test.PrepareDBConnectionString(t, dbType)
+	db, err := storage.NewDatabase(base, &config.DatabaseOptions{ConnectionString: config.DataSource(connStr)})
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	return db, clearDB
+}
+
+type mockKeyserverRoomserverAPI struct {
+	leftUsers []string
+}
+
+func (m *mockKeyserverRoomserverAPI) QueryLeftUsers(ctx context.Context, req *roomserver.QueryLeftUsersRequest, res *roomserver.QueryLeftUsersResponse) error {
+	res.LeftUsers = m.leftUsers
+	return nil
+}
+
+func TestDeviceListUpdater_CleanUp(t *testing.T) {
+	processCtx := process.NewProcessContext()
+
+	alice := test.NewUser(t)
+	bob := test.NewUser(t)
+
+	// Bob is not joined to any of our rooms
+	rsAPI := &mockKeyserverRoomserverAPI{leftUsers: []string{bob.ID}}
+
+	test.WithAllDatabases(t, func(t *testing.T, dbType test.DBType) {
+		db, clearDB := mustCreateKeyserverDB(t, dbType)
+		defer clearDB()
+
+		// This should not get deleted
+		if err := db.MarkDeviceListStale(processCtx.Context(), alice.ID, true); err != nil {
+			t.Error(err)
+		}
+
+		// this one should get deleted
+		if err := db.MarkDeviceListStale(processCtx.Context(), bob.ID, true); err != nil {
+			t.Error(err)
+		}
+
+		updater := NewDeviceListUpdater(processCtx, db, nil,
+			nil, nil,
+			0, rsAPI, "test")
+		if err := updater.CleanUp(); err != nil {
+			t.Error(err)
+		}
+
+		// check that we still have Alice in our stale list
+		staleUsers, err := db.StaleDeviceLists(ctx, []gomatrixserverlib.ServerName{"test"})
+		if err != nil {
+			t.Error(err)
+		}
+
+		// There should only be Alice
+		wantCount := 1
+		if count := len(staleUsers); count != wantCount {
+			t.Fatalf("expected there to be %d stale device lists, got %d", wantCount, count)
+		}
+
+		if staleUsers[0] != alice.ID {
+			t.Fatalf("unexpected stale device list user: %s, want %s", staleUsers[0], alice.ID)
+		}
+	})
+}