Fix ordering when backfilling (#1000)

* Fix ordering when backfilling The problem was that we weren't sorting the returned events by depth when sending them back to the caller, instead we were sorting by prev_events which is not the same thing. * Fixup tests
author: Kegsay <kegan@matrix.org> 2020-05-01 16:41:13 +0100
committer: GitHub <noreply@github.com> 2020-05-01 16:41:13 +0100
commit: 36bbb255614d289a3417194d4d2ac129e339f531 (patch)
tree: be8ea170d1ae2e04740b95dedbe63d4892381109
parent: f7cfa758864cc9849ce8b0539895283417b75eb0 (diff)
3 files changed, 110 insertions, 10 deletions
diff --git a/syncapi/routing/messages.go b/syncapi/routing/messages.go
index d7e39704..270b0ee9 100644
--- a/syncapi/routing/messages.go
+++ b/syncapi/routing/messages.go
@@ -210,15 +210,16 @@ func (r *messagesReq) retrieveEvents() (
 	}
 
 	// Sort the events to ensure we send them in the right order.
-	events = gomatrixserverlib.HeaderedReverseTopologicalOrdering(
-		events,
-		gomatrixserverlib.TopologicalOrderByPrevEvents,
-	)
 	if r.backwardOrdering {
 		// This reverses the array from old->new to new->old
-		sort.SliceStable(events, func(i, j int) bool {
-			return true
-		})
+		reversed := func(in []gomatrixserverlib.HeaderedEvent) []gomatrixserverlib.HeaderedEvent {
+			out := make([]gomatrixserverlib.HeaderedEvent, len(in))
+			for i := 0; i < len(in); i++ {
+				out[i] = in[len(in)-i-1]
+			}
+			return out
+		}
+		events = reversed(events)
 	}
 
 	// Convert all of the events into client events.
@@ -259,6 +260,7 @@ func (r *messagesReq) retrieveEvents() (
 		// to them by the event on their left, therefore we need to decrement the
 		// end position we send in the response if we're going backward.
 		end.PDUPosition--
+		end.EDUTypingPosition += 1000
 	}
 
 	// The lowest token value is 1, therefore we need to manually set it to that
@@ -345,10 +347,23 @@ func (r *messagesReq) handleNonEmptyEventsSlice(streamEvents []types.StreamEvent
 
 	// Append the events ve previously retrieved locally.
 	events = append(events, r.db.StreamEventsToEvents(nil, streamEvents)...)
+	sort.Sort(eventsByDepth(events))
 
 	return
 }
 
+type eventsByDepth []gomatrixserverlib.HeaderedEvent
+
+func (e eventsByDepth) Len() int {
+	return len(e)
+}
+func (e eventsByDepth) Swap(i, j int) {
+	e[i], e[j] = e[j], e[i]
+}
+func (e eventsByDepth) Less(i, j int) bool {
+	return e[i].Depth() < e[j].Depth()
+}
+
 // backfill performs a backfill request over the federation on another
 // homeserver in the room.
 // See: https://matrix.org/docs/spec/server_server/latest#get-matrix-federation-v1-backfill-roomid
@@ -375,17 +390,24 @@ func (r *messagesReq) backfill(roomID string, fromEventIDs []string, limit int)
 	// Currently, this can race with live events for the room and cause problems. It's also just a bit unclear
 	// when you have multiple entry points to write events.
 
+	// we have to order these by depth, starting with the lowest because otherwise the topology tokens
+	// will skip over events that have the same depth but different stream positions due to the query which is:
+	//  - anything less than the depth OR
+	//  - anything with the same depth and a lower stream position.
+	sort.Sort(eventsByDepth(res.Events))
+
 	// Store the events in the database, while marking them as unfit to show
 	// up in responses to sync requests.
 	for i := range res.Events {
-		if _, err = r.db.WriteEvent(
+		_, err = r.db.WriteEvent(
 			r.ctx,
 			&res.Events[i],
 			[]gomatrixserverlib.HeaderedEvent{},
 			[]string{},
 			[]string{},
 			nil, true,
-		); err != nil {
+		)
+		if err != nil {
 			return nil, err
 		}
 	}
diff --git a/syncapi/storage/sqlite3/syncserver.go b/syncapi/storage/sqlite3/syncserver.go
index 425073a7..314ea2aa 100644
--- a/syncapi/storage/sqlite3/syncserver.go
+++ b/syncapi/storage/sqlite3/syncserver.go
@@ -434,6 +434,7 @@ func (d *SyncServerDatasource) syncPositionTx(
 	}
 	sp.PDUPosition = types.StreamPosition(maxEventID)
 	sp.EDUTypingPosition = types.StreamPosition(d.eduCache.GetLatestSyncPosition())
+	sp.Type = types.PaginationTokenTypeStream
 	return
 }
 
@@ -658,6 +659,7 @@ func (d *SyncServerDatasource) getResponseWithPDUsForCompleteSync(
 			backwardTopologyPos = types.StreamPosition(1)
 		} else {
 			backwardTopologyPos--
+			backwardTopologyStreamPos += 1000 // this has to be bigger than the number of events we backfill per request
 		}
 
 		// We don't include a device here as we don't need to send down
@@ -817,11 +819,13 @@ func (d *SyncServerDatasource) getBackwardTopologyPos(
 	if len(events) > 0 {
 		pos, spos, _ = d.topology.selectPositionInTopology(ctx, txn, events[0].EventID())
 	}
-	// TODO: I have no idea what this is doing.
+	// go to the previous position so we don't pull out the same event twice
+	// FIXME: This could be done more nicely by being explicit with inclusive/exclusive rules
 	if pos-1 <= 0 {
 		pos = types.StreamPosition(1)
 	} else {
 		pos = pos - 1
+		spos += 1000 // this has to be bigger than the number of events we backfill per request
 	}
 	return
 }
diff --git a/syncapi/storage/storage_test.go b/syncapi/storage/storage_test.go
index 378c1fe3..b951efa4 100644
--- a/syncapi/storage/storage_test.go
+++ b/syncapi/storage/storage_test.go
@@ -118,6 +118,7 @@ func MustWriteEvents(t *testing.T, db storage.Database, events []gomatrixserverl
 		if err != nil {
 			t.Fatalf("WriteEvent failed: %s", err)
 		}
+		fmt.Println("Event ID", ev.EventID(), " spos=", pos, "depth=", ev.Depth())
 		positions = append(positions, pos)
 	}
 	return
@@ -407,6 +408,64 @@ func TestGetEventsInRangeWithEventsSameDepth(t *testing.T) {
 	}
 }
 
+// The purpose of this test is to make sure that events are returned in the right *order* when they have been inserted in a manner similar to
+// how any kind of backfill operation will insert the events. This test inserts the SimpleRoom events in a manner similar to how backfill over
+// federation would:
+// - First inserts join event of test user C
+// - Inserts chunks of history in strata e.g (25-30, 20-25, 15-20, 10-15, 5-10, 0-5).
+// The test then does a backfill to ensure that the response is ordered correctly according to depth.
+func TestGetEventsInRangeWithEventsInsertedLikeBackfill(t *testing.T) {
+	t.Parallel()
+	db := MustCreateDatabase(t)
+	events, _ := SimpleRoom(t, testRoomID, testUserIDA, testUserIDB)
+
+	// "federation" join
+	userC := fmt.Sprintf("@radiance:%s", testOrigin)
+	joinEvent := MustCreateEvent(t, testRoomID, []gomatrixserverlib.HeaderedEvent{events[len(events)-1]}, &gomatrixserverlib.EventBuilder{
+		Content:  []byte(fmt.Sprintf(`{"membership":"join"}`)),
+		Type:     "m.room.member",
+		StateKey: &userC,
+		Sender:   userC,
+		Depth:    int64(len(events) + 1),
+	})
+	MustWriteEvents(t, db, []gomatrixserverlib.HeaderedEvent{joinEvent})
+
+	// Sync will return this for the prev_batch
+	from := topologyTokenBefore(t, db, joinEvent.EventID())
+
+	// inject events in batches as if they were from backfill
+	// e.g [1,2,3,4,5,6] => [4,5,6] , [1,2,3]
+	chunkSize := 5
+	for i := len(events); i >= 0; i -= chunkSize {
+		start := i - chunkSize
+		if start < 0 {
+			start = 0
+		}
+		backfill := events[start:i]
+		MustWriteEvents(t, db, backfill)
+	}
+
+	// head towards the beginning of time
+	to := types.NewPaginationTokenFromTypeAndPosition(types.PaginationTokenTypeTopology, 0, 0)
+
+	// starting at `from`, backpaginate to the beginning of time, asserting as we go.
+	chunkSize = 3
+	events = reversed(events)
+	for i := 0; i < len(events); i += chunkSize {
+		paginatedEvents, err := db.GetEventsInRange(ctx, from, to, testRoomID, chunkSize, true)
+		if err != nil {
+			t.Fatalf("GetEventsInRange returned an error: %s", err)
+		}
+		gots := gomatrixserverlib.HeaderedToClientEvents(db.StreamEventsToEvents(&testUserDeviceA, paginatedEvents), gomatrixserverlib.FormatAll)
+		endi := i + chunkSize
+		if endi > len(events) {
+			endi = len(events)
+		}
+		assertEventsEqual(t, from.String(), true, gots, events[i:endi])
+		from = topologyTokenBefore(t, db, paginatedEvents[len(paginatedEvents)-1].EventID())
+	}
+}
+
 func assertEventsEqual(t *testing.T, msg string, checkRoomID bool, gots []gomatrixserverlib.ClientEvent, wants []gomatrixserverlib.HeaderedEvent) {
 	if len(gots) != len(wants) {
 		t.Fatalf("%s response returned %d events, want %d", msg, len(gots), len(wants))
@@ -447,6 +506,21 @@ func assertEventsEqual(t *testing.T, msg string, checkRoomID bool, gots []gomatr
 	}
 }
 
+func topologyTokenBefore(t *testing.T, db storage.Database, eventID string) *types.PaginationToken {
+	pos, spos, err := db.EventPositionInTopology(ctx, eventID)
+	if err != nil {
+		t.Fatalf("failed to get EventPositionInTopology: %s", err)
+	}
+
+	if pos-1 <= 0 {
+		pos = types.StreamPosition(1)
+	} else {
+		pos = pos - 1
+		spos += 1000 // this has to be bigger than the chunk limit
+	}
+	return types.NewPaginationTokenFromTypeAndPosition(types.PaginationTokenTypeTopology, pos, spos)
+}
+
 func reversed(in []gomatrixserverlib.HeaderedEvent) []gomatrixserverlib.HeaderedEvent {
 	out := make([]gomatrixserverlib.HeaderedEvent, len(in))
 	for i := 0; i < len(in); i++ {
author	Kegsay <kegan@matrix.org>	2020-05-01 16:41:13 +0100
committer	GitHub <noreply@github.com>	2020-05-01 16:41:13 +0100
commit	36bbb255614d289a3417194d4d2ac129e339f531 (patch)
tree	be8ea170d1ae2e04740b95dedbe63d4892381109
parent	f7cfa758864cc9849ce8b0539895283417b75eb0 (diff)