go/sliding-sync
go/sliding-sync
1
0
Fork 0
mirror of https://github.com/matrix-org/sliding-sync.git synced 2025-03-10 13:37:11 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #441 from matrix-org/s7evink/startup-perf

Browse Source 
Optimize getting the latest events by type per room
This commit is contained in:
Kegan Dougal 2024-05-21 18:13:06 +01:00 committed by GitHub
commit 98b12cba6a
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 23 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

35
state/event_table.go
View File

@ -441,18 +441,29 @@ func (t *EventTable) SelectLatestEventsBetween(txn *sqlx.Tx, roomID string, lowe
func (t *EventTable) selectLatestEventByTypeInAllRooms(txn *sqlx.Tx) ([]Event, error) { func (t *EventTable) selectLatestEventByTypeInAllRooms(txn *sqlx.Tx) ([]Event, error) {
result := []Event{} result := []Event{}
// TODO: this query ends up doing a sequential scan on the events table. We have // What the following query does:
// an index on (event_type, room_id, event_nid) so I'm a little surprised that PG // 1. Gets all event types from a recursive CTE as the `event_types` CTE
// decides to do so. Can we do something better here? Ideas: // 2. Gets all rooms as the `room_ids` CTE
// - Find a better query for selecting the newest event of each type in a room. // 3. Gets the latest event_nid for each event_type and room as the `max_by_ev_type` CTE
// - At present we only care about the _timestamps_ of these events. Perhaps we // 4. Queries the required data using the event_nids provided by the `max_by_ev_type` CTE
// could store those in the DB (and even in an index) as a column and select rows, err := txn.Query(`
// those, to avoid having to parse the event bodies. WITH event_types AS (
// - We could have the application maintain a `latest_events` table so that the WITH RECURSIVE t AS (
// rows can be directly read. Assuming a mostly-static set of event types, reads (SELECT event_type FROM syncv3_events ORDER BY event_type LIMIT 1) -- parentheses required
// are then linear in the number of rooms. UNION ALL
rows, err := txn.Query( SELECT (SELECT event_type FROM syncv3_events WHERE event_type > t.event_type ORDER BY event_type LIMIT 1)
`SELECT room_id, event_nid, event FROM syncv3_events WHERE event_nid in (SELECT MAX(event_nid) FROM syncv3_events GROUP BY room_id, event_type)`, FROM t
WHERE t.event_type IS NOT NULL
)
SELECT event_type FROM t WHERE event_type IS NOT NULL
), room_ids AS (
SELECT DISTINCT room_id FROM syncv3_rooms
), max_by_ev_type AS (
SELECT m.max FROM event_types, room_ids,
LATERAL ( SELECT max(event_nid) as max FROM syncv3_events e WHERE e.room_id = room_ids.room_id AND e.event_type = event_types.event_type ) AS m
)
SELECT room_id, event_nid, event FROM syncv3_events, max_by_ev_type WHERE event_nid = max_by_ev_type.max
`,
) )
if err != nil { if err != nil {
return nil, err return nil, err