mirror of
https://github.com/1f349/dendrite.git
synced 2024-11-29 23:21:36 +00:00
eb29a31550
Should fix the following issues or make a lot less worse when using Postgres: The main issue behind #2911: The client gives up after a certain time, causing a cascade of context errors, because the response couldn't be built up fast enough. This mostly happens on accounts with many rooms, due to the inefficient way we're getting recent events and current state For #2777: The queries for getting the membership events for history visibility were being executed for each room (I think 185?), resulting in a whooping 2k queries for membership events. (Getting the statesnapshot -> block nids -> actual wanted membership event) Both should now be better by: - Using a LATERAL join to get all recent events for all joined rooms in one go (TODO: maybe do the same for room summary and current state etc) - If we're lazy loading on initial syncs, we're now not getting the whole current state, just to drop the majority of it because we're lazy loading members - we add a filter to exclude membership events on the first call to `CurrentState`. - Using an optimized query to get the membership events needed to calculate history visibility --------- Co-authored-by: kegsay <kegan@matrix.org>
247 lines
9.9 KiB
Go
247 lines
9.9 KiB
Go
// Copyright 2017-2018 New Vector Ltd
|
|
// Copyright 2019-2020 The Matrix.org Foundation C.I.C.
|
|
//
|
|
// Licensed under the Apache License, Version 2.0 (the "License");
|
|
// you may not use this file except in compliance with the License.
|
|
// You may obtain a copy of the License at
|
|
//
|
|
// http://www.apache.org/licenses/LICENSE-2.0
|
|
//
|
|
// Unless required by applicable law or agreed to in writing, software
|
|
// distributed under the License is distributed on an "AS IS" BASIS,
|
|
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
// See the License for the specific language governing permissions and
|
|
// limitations under the License.
|
|
|
|
package postgres
|
|
|
|
import (
|
|
"context"
|
|
"database/sql"
|
|
"fmt"
|
|
|
|
"github.com/lib/pq"
|
|
"github.com/matrix-org/gomatrixserverlib"
|
|
"github.com/matrix-org/util"
|
|
|
|
"github.com/matrix-org/dendrite/internal/sqlutil"
|
|
"github.com/matrix-org/dendrite/roomserver/types"
|
|
)
|
|
|
|
const stateSnapshotSchema = `
|
|
-- The state of a room before an event.
|
|
-- Stored as a list of state_block entries stored in a separate table.
|
|
-- The actual state is constructed by combining all the state_block entries
|
|
-- referenced by state_block_nids together. If the same state key tuple appears
|
|
-- multiple times then the entry from the later state_block clobbers the earlier
|
|
-- entries.
|
|
-- This encoding format allows us to implement a delta encoding which is useful
|
|
-- because room state tends to accumulate small changes over time. Although if
|
|
-- the list of deltas becomes too long it becomes more efficient to encode
|
|
-- the full state under single state_block_nid.
|
|
CREATE SEQUENCE IF NOT EXISTS roomserver_state_snapshot_nid_seq;
|
|
CREATE TABLE IF NOT EXISTS roomserver_state_snapshots (
|
|
-- The state snapshot NID that identifies this snapshot.
|
|
state_snapshot_nid bigint PRIMARY KEY DEFAULT nextval('roomserver_state_snapshot_nid_seq'),
|
|
-- The hash of the state snapshot, which is used to enforce uniqueness. The hash is
|
|
-- generated in Dendrite and passed through to the database, as a btree index over
|
|
-- this column is cheap and fits within the maximum index size.
|
|
state_snapshot_hash BYTEA UNIQUE,
|
|
-- The room NID that the snapshot belongs to.
|
|
room_nid bigint NOT NULL,
|
|
-- The state blocks contained within this snapshot.
|
|
state_block_nids bigint[] NOT NULL
|
|
);
|
|
`
|
|
|
|
// Insert a new state snapshot. If we conflict on the hash column then
|
|
// we must perform an update so that the RETURNING statement returns the
|
|
// ID of the row that we conflicted with, so that we can then refer to
|
|
// the original snapshot.
|
|
const insertStateSQL = "" +
|
|
"INSERT INTO roomserver_state_snapshots (state_snapshot_hash, room_nid, state_block_nids)" +
|
|
" VALUES ($1, $2, $3)" +
|
|
" ON CONFLICT (state_snapshot_hash) DO UPDATE SET room_nid=$2" +
|
|
// Performing an update, above, ensures that the RETURNING statement
|
|
// below will always return a valid state snapshot ID
|
|
" RETURNING state_snapshot_nid"
|
|
|
|
// Bulk state data NID lookup.
|
|
// Sorting by state_snapshot_nid means we can use binary search over the result
|
|
// to lookup the state data NIDs for a state snapshot NID.
|
|
const bulkSelectStateBlockNIDsSQL = "" +
|
|
"SELECT state_snapshot_nid, state_block_nids FROM roomserver_state_snapshots" +
|
|
" WHERE state_snapshot_nid = ANY($1) ORDER BY state_snapshot_nid ASC"
|
|
|
|
// Looks up both the history visibility event and relevant membership events from
|
|
// a given domain name from a given state snapshot. This is used to optimise the
|
|
// helpers.CheckServerAllowedToSeeEvent function.
|
|
// TODO: There's a sequence scan here because of the hash join strategy, which is
|
|
// probably O(n) on state key entries, so there must be a way to avoid that somehow.
|
|
// Event type NIDs are:
|
|
// - 5: m.room.member as per https://github.com/matrix-org/dendrite/blob/c7f7aec4d07d59120d37d5b16a900f6d608a75c4/roomserver/storage/postgres/event_types_table.go#L40
|
|
// - 7: m.room.history_visibility as per https://github.com/matrix-org/dendrite/blob/c7f7aec4d07d59120d37d5b16a900f6d608a75c4/roomserver/storage/postgres/event_types_table.go#L42
|
|
const bulkSelectStateForHistoryVisibilitySQL = `
|
|
SELECT event_nid FROM (
|
|
SELECT event_nid, event_type_nid, event_state_key_nid FROM roomserver_events
|
|
WHERE (event_type_nid = 5 OR event_type_nid = 7)
|
|
AND event_nid = ANY(
|
|
SELECT UNNEST(event_nids) FROM roomserver_state_block
|
|
WHERE state_block_nid = ANY(
|
|
SELECT UNNEST(state_block_nids) FROM roomserver_state_snapshots
|
|
WHERE state_snapshot_nid = $1
|
|
)
|
|
)
|
|
ORDER BY depth ASC
|
|
) AS roomserver_events
|
|
INNER JOIN roomserver_event_state_keys
|
|
ON roomserver_events.event_state_key_nid = roomserver_event_state_keys.event_state_key_nid
|
|
AND (event_type_nid = 7 OR event_state_key LIKE '%:' || $2);
|
|
`
|
|
|
|
// bulkSelectMembershipForHistoryVisibilitySQL is an optimization to get membership events for a specific user for defined set of events.
|
|
// Returns the event_id of the event we want the membership event for, the event_id of the membership event and the membership event JSON.
|
|
const bulkSelectMembershipForHistoryVisibilitySQL = `
|
|
SELECT re.event_id, re2.event_id, rej.event_json
|
|
FROM roomserver_events re
|
|
LEFT JOIN roomserver_state_snapshots rss on re.state_snapshot_nid = rss.state_snapshot_nid
|
|
CROSS JOIN unnest(rss.state_block_nids) AS blocks(block_nid)
|
|
LEFT JOIN roomserver_state_block rsb ON rsb.state_block_nid = blocks.block_nid
|
|
CROSS JOIN unnest(rsb.event_nids) AS rsb2(event_nid)
|
|
JOIN roomserver_events re2 ON re2.room_nid = $3 AND re2.event_type_nid = 5 AND re2.event_nid = rsb2.event_nid AND re2.event_state_key_nid = $1
|
|
LEFT JOIN roomserver_event_json rej ON rej.event_nid = re2.event_nid
|
|
WHERE re.event_id = ANY($2)
|
|
|
|
`
|
|
|
|
type stateSnapshotStatements struct {
|
|
insertStateStmt *sql.Stmt
|
|
bulkSelectStateBlockNIDsStmt *sql.Stmt
|
|
bulkSelectStateForHistoryVisibilityStmt *sql.Stmt
|
|
bulktSelectMembershipForHistoryVisibilityStmt *sql.Stmt
|
|
}
|
|
|
|
func CreateStateSnapshotTable(db *sql.DB) error {
|
|
_, err := db.Exec(stateSnapshotSchema)
|
|
return err
|
|
}
|
|
|
|
func PrepareStateSnapshotTable(db *sql.DB) (*stateSnapshotStatements, error) {
|
|
s := &stateSnapshotStatements{}
|
|
|
|
return s, sqlutil.StatementList{
|
|
{&s.insertStateStmt, insertStateSQL},
|
|
{&s.bulkSelectStateBlockNIDsStmt, bulkSelectStateBlockNIDsSQL},
|
|
{&s.bulkSelectStateForHistoryVisibilityStmt, bulkSelectStateForHistoryVisibilitySQL},
|
|
{&s.bulktSelectMembershipForHistoryVisibilityStmt, bulkSelectMembershipForHistoryVisibilitySQL},
|
|
}.Prepare(db)
|
|
}
|
|
|
|
func (s *stateSnapshotStatements) InsertState(
|
|
ctx context.Context, txn *sql.Tx, roomNID types.RoomNID, nids types.StateBlockNIDs,
|
|
) (stateNID types.StateSnapshotNID, err error) {
|
|
nids = nids[:util.SortAndUnique(nids)]
|
|
err = sqlutil.TxStmt(txn, s.insertStateStmt).QueryRowContext(ctx, nids.Hash(), int64(roomNID), stateBlockNIDsAsArray(nids)).Scan(&stateNID)
|
|
if err != nil {
|
|
return 0, err
|
|
}
|
|
return
|
|
}
|
|
|
|
func (s *stateSnapshotStatements) BulkSelectStateBlockNIDs(
|
|
ctx context.Context, txn *sql.Tx, stateNIDs []types.StateSnapshotNID,
|
|
) ([]types.StateBlockNIDList, error) {
|
|
nids := make([]int64, len(stateNIDs))
|
|
for i := range stateNIDs {
|
|
nids[i] = int64(stateNIDs[i])
|
|
}
|
|
stmt := sqlutil.TxStmt(txn, s.bulkSelectStateBlockNIDsStmt)
|
|
rows, err := stmt.QueryContext(ctx, pq.Int64Array(nids))
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
defer rows.Close() // nolint: errcheck
|
|
results := make([]types.StateBlockNIDList, len(stateNIDs))
|
|
i := 0
|
|
var stateBlockNIDs pq.Int64Array
|
|
for ; rows.Next(); i++ {
|
|
result := &results[i]
|
|
if err = rows.Scan(&result.StateSnapshotNID, &stateBlockNIDs); err != nil {
|
|
return nil, err
|
|
}
|
|
result.StateBlockNIDs = make([]types.StateBlockNID, len(stateBlockNIDs))
|
|
for k := range stateBlockNIDs {
|
|
result.StateBlockNIDs[k] = types.StateBlockNID(stateBlockNIDs[k])
|
|
}
|
|
}
|
|
if err = rows.Err(); err != nil {
|
|
return nil, err
|
|
}
|
|
if i != len(stateNIDs) {
|
|
return nil, types.MissingStateError(fmt.Sprintf("storage: state NIDs missing from the database (%d != %d)", i, len(stateNIDs)))
|
|
}
|
|
return results, nil
|
|
}
|
|
|
|
func (s *stateSnapshotStatements) BulkSelectStateForHistoryVisibility(
|
|
ctx context.Context, txn *sql.Tx, stateSnapshotNID types.StateSnapshotNID, domain string,
|
|
) ([]types.EventNID, error) {
|
|
stmt := sqlutil.TxStmt(txn, s.bulkSelectStateForHistoryVisibilityStmt)
|
|
rows, err := stmt.QueryContext(ctx, stateSnapshotNID, domain)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
defer rows.Close() // nolint: errcheck
|
|
results := make([]types.EventNID, 0, 16)
|
|
for rows.Next() {
|
|
var eventNID types.EventNID
|
|
if err = rows.Scan(&eventNID); err != nil {
|
|
return nil, err
|
|
}
|
|
results = append(results, eventNID)
|
|
}
|
|
return results, rows.Err()
|
|
}
|
|
|
|
func (s *stateSnapshotStatements) BulkSelectMembershipForHistoryVisibility(
|
|
ctx context.Context, txn *sql.Tx, userNID types.EventStateKeyNID, roomInfo *types.RoomInfo, eventIDs ...string,
|
|
) (map[string]*gomatrixserverlib.HeaderedEvent, error) {
|
|
stmt := sqlutil.TxStmt(txn, s.bulktSelectMembershipForHistoryVisibilityStmt)
|
|
rows, err := stmt.QueryContext(ctx, userNID, pq.Array(eventIDs), roomInfo.RoomNID)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
defer rows.Close() // nolint: errcheck
|
|
result := make(map[string]*gomatrixserverlib.HeaderedEvent, len(eventIDs))
|
|
var evJson []byte
|
|
var eventID string
|
|
var membershipEventID string
|
|
|
|
knownEvents := make(map[string]*gomatrixserverlib.HeaderedEvent, len(eventIDs))
|
|
|
|
for rows.Next() {
|
|
if err = rows.Scan(&eventID, &membershipEventID, &evJson); err != nil {
|
|
return nil, err
|
|
}
|
|
if len(evJson) == 0 {
|
|
result[eventID] = &gomatrixserverlib.HeaderedEvent{}
|
|
continue
|
|
}
|
|
// If we already know this event, don't try to marshal the json again
|
|
if ev, ok := knownEvents[membershipEventID]; ok {
|
|
result[eventID] = ev
|
|
continue
|
|
}
|
|
event, err := gomatrixserverlib.NewEventFromTrustedJSON(evJson, false, roomInfo.RoomVersion)
|
|
if err != nil {
|
|
result[eventID] = &gomatrixserverlib.HeaderedEvent{}
|
|
// not fatal
|
|
continue
|
|
}
|
|
he := event.Headered(roomInfo.RoomVersion)
|
|
result[eventID] = he
|
|
knownEvents[membershipEventID] = he
|
|
}
|
|
return result, rows.Err()
|
|
}
|