State storage refactor (#1839)

* Hash-deduplicated state storage (and migrations) for PostgreSQL and SQLite

* Refactor droomserver database setup for migrations

* Fix conflict statements

* Update migration names

* Set a boundary for old to new block/snapshot IDs so we don't rewrite them more than once accidentally

* Create sequence if not exists

* Fix boundary queries

* Fix boundary queries

* Use Query

* Break out queries a bit

* More sequence tweaks

* Query parameters are not playing the game

* Injection escaping may not work for CREATE SEQUENCE after all

* Fix snapshot sequence name

* Use boundaried IDs in SQLite too

* Use IFNULL for SQLite

* Use COALESCE in PostgreSQL

* Review comments @Kegsay
This commit is contained in:
Neil Alexander 2021-04-26 13:25:57 +01:00 committed by GitHub
parent d6e9b7b307
commit 5ce1fe80de
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
39 changed files with 1076 additions and 554 deletions

View file

@ -27,19 +27,34 @@ import (
"github.com/matrix-org/dendrite/roomserver/storage/shared"
"github.com/matrix-org/dendrite/roomserver/storage/tables"
"github.com/matrix-org/dendrite/roomserver/types"
"github.com/matrix-org/util"
)
const stateSnapshotSchema = `
CREATE TABLE IF NOT EXISTS roomserver_state_snapshots (
-- The state snapshot NID that identifies this snapshot.
state_snapshot_nid INTEGER PRIMARY KEY AUTOINCREMENT,
-- 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 BLOB UNIQUE,
-- The room NID that the snapshot belongs to.
room_nid INTEGER NOT NULL,
-- The state blocks contained within this snapshot, encoded as JSON.
state_block_nids TEXT NOT NULL DEFAULT '[]'
);
`
// 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 (room_nid, state_block_nids)
VALUES ($1, $2);`
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
RETURNING state_snapshot_nid
`
// Bulk state data NID lookup.
// Sorting by state_snapshot_nid means we can use binary search over the result
@ -54,14 +69,15 @@ type stateSnapshotStatements struct {
bulkSelectStateBlockNIDsStmt *sql.Stmt
}
func NewSqliteStateSnapshotTable(db *sql.DB) (tables.StateSnapshot, error) {
func createStateSnapshotTable(db *sql.DB) error {
_, err := db.Exec(stateSnapshotSchema)
return err
}
func prepareStateSnapshotTable(db *sql.DB) (tables.StateSnapshot, error) {
s := &stateSnapshotStatements{
db: db,
}
_, err := db.Exec(stateSnapshotSchema)
if err != nil {
return nil, err
}
return s, shared.StatementList{
{&s.insertStateStmt, insertStateSQL},
@ -70,22 +86,20 @@ func NewSqliteStateSnapshotTable(db *sql.DB) (tables.StateSnapshot, error) {
}
func (s *stateSnapshotStatements) InsertState(
ctx context.Context, txn *sql.Tx, roomNID types.RoomNID, stateBlockNIDs []types.StateBlockNID,
ctx context.Context, txn *sql.Tx, roomNID types.RoomNID, stateBlockNIDs types.StateBlockNIDs,
) (stateNID types.StateSnapshotNID, err error) {
stateBlockNIDs = stateBlockNIDs[:util.SortAndUnique(stateBlockNIDs)]
stateBlockNIDsJSON, err := json.Marshal(stateBlockNIDs)
if err != nil {
return
}
insertStmt := sqlutil.TxStmt(txn, s.insertStateStmt)
res, err := insertStmt.ExecContext(ctx, int64(roomNID), string(stateBlockNIDsJSON))
var id int64
err = insertStmt.QueryRowContext(ctx, stateBlockNIDs.Hash(), int64(roomNID), string(stateBlockNIDsJSON)).Scan(&id)
if err != nil {
return 0, err
}
lastRowID, err := res.LastInsertId()
if err != nil {
return 0, err
}
stateNID = types.StateSnapshotNID(lastRowID)
stateNID = types.StateSnapshotNID(id)
return
}