feat: sync multilayer td integration

This commit is contained in:
A 2026-07-15 13:32:06 +08:00
parent 20a310f6ca
commit 766c5db992
491 changed files with 26235 additions and 35340 deletions

View file

@ -8,7 +8,6 @@ import (
"time"
"github.com/jackc/pgx/v5"
"github.com/jackc/pgx/v5/pgconn"
"github.com/jackc/pgx/v5/pgtype"
"telesrv/internal/store"
@ -56,24 +55,26 @@ WHERE auth_keys.body = EXCLUDED.body
// 注册进 SessionManager”的窗口被后台清理。
func (s *AuthKeyStore) Get(ctx context.Context, id [8]byte) (store.AuthKeyData, bool, error) {
var (
body []byte
serverSalt int64
expiresAt int
createdAt pgtype.Timestamptz
layer int
deviceModel string
platform string
systemVersion string
apiID int
appVersion string
body []byte
serverSalt int64
expiresAt int
createdAt pgtype.Timestamptz
layer int
layerObservationID int64
deviceModel string
platform string
systemVersion string
apiID int
appVersion string
)
err := s.db.QueryRow(ctx, `
UPDATE auth_keys
SET last_used_at = now()
WHERE auth_key_id = $1
RETURNING auth_key_id, body, server_salt, created_at,
expires_at, layer, device_model, platform, system_version, api_id, app_version
`, authKeyIDToInt64(id)).Scan(new(int64), &body, &serverSalt, &createdAt, &expiresAt, &layer, &deviceModel, &platform, &systemVersion, &apiID, &appVersion)
expires_at, layer, layer_observation_id,
device_model, platform, system_version, api_id, app_version
`, authKeyIDToInt64(id)).Scan(new(int64), &body, &serverSalt, &createdAt, &expiresAt, &layer, &layerObservationID, &deviceModel, &platform, &systemVersion, &apiID, &appVersion)
if err != nil {
if errors.Is(err, pgx.ErrNoRows) {
return store.AuthKeyData{}, false, nil
@ -84,15 +85,16 @@ RETURNING auth_key_id, body, server_salt, created_at,
return store.AuthKeyData{}, false, fmt.Errorf("auth key body length = %d, want 256", len(body))
}
data := store.AuthKeyData{
ID: id,
ServerSalt: serverSalt,
ExpiresAt: expiresAt,
Layer: layer,
DeviceModel: deviceModel,
Platform: platform,
SystemVersion: systemVersion,
APIID: apiID,
AppVersion: appVersion,
ID: id,
ServerSalt: serverSalt,
ExpiresAt: expiresAt,
Layer: layer,
LayerObservationID: layerObservationID,
DeviceModel: deviceModel,
Platform: platform,
SystemVersion: systemVersion,
APIID: apiID,
AppVersion: appVersion,
}
copy(data.Value[:], body)
if createdAt.Valid {
@ -148,7 +150,8 @@ WHERE auth_key_id = ANY($1::bigint[])`, batch)
}
func (s *AuthKeyStore) UpdateClientInfo(ctx context.Context, id [8]byte, info store.AuthKeyClientInfo) error {
if _, err := s.db.Exec(ctx, `
var updated int
err := s.db.QueryRow(ctx, `
UPDATE auth_keys
SET layer = CASE WHEN $2::integer > 0 THEN $2 ELSE layer END,
device_model = CASE WHEN $3::text <> '' THEN $3 ELSE device_model END,
@ -157,7 +160,31 @@ SET layer = CASE WHEN $2::integer > 0 THEN $2 ELSE layer END,
api_id = CASE WHEN $6::integer <> 0 THEN $6 ELSE api_id END,
app_version = CASE WHEN $7::text <> '' THEN $7 ELSE app_version END
WHERE auth_key_id = $1
`, authKeyIDToInt64(id), info.Layer, info.DeviceModel, info.Platform, info.SystemVersion, info.APIID, info.AppVersion); err != nil {
AND ($2::integer <= 0 OR layer_observation_id = 0 OR layer = $2::integer)
RETURNING 1
`, authKeyIDToInt64(id), info.Layer, info.DeviceModel, info.Platform, info.SystemVersion, info.APIID, info.AppVersion).Scan(&updated)
if err != nil {
if errors.Is(err, pgx.ErrNoRows) {
var (
currentLayer int
observation int64
)
lookupErr := s.db.QueryRow(ctx, `
SELECT layer, layer_observation_id
FROM auth_keys
WHERE auth_key_id = $1
`, authKeyIDToInt64(id)).Scan(&currentLayer, &observation)
switch {
case errors.Is(lookupErr, pgx.ErrNoRows):
return store.ErrAuthKeyNotFound
case lookupErr != nil:
return fmt.Errorf("classify auth key client info update: %w", lookupErr)
case info.Layer > 0 && observation > 0 && currentLayer != info.Layer:
return store.ErrAuthKeySessionLayerConflict
default:
return fmt.Errorf("update auth key client info: guarded update affected no row")
}
}
return fmt.Errorf("update auth key client info: %w", err)
}
return nil
@ -171,25 +198,20 @@ WHERE auth_key_id = $1
// RESTRICT FK 防止悬空,因此被踢/销毁 perm key 时必须先把关联 temp key 一并删掉。否则 Web/上传连接用
// raw temp key 重连时仍能进入 RPC 层,只得到 AUTH_KEY_UNREGISTERED而不是连接层 404。
func (s *AuthKeyStore) Delete(ctx context.Context, id [8]byte) error {
for attempt := 0; attempt < 3; attempt++ {
err := s.deleteAuthKeyOnce(ctx, id)
if err == nil {
return nil
}
if !isPermAuthKeyDeleteRace(err) {
return err
}
if _, inTx := s.db.(pgx.Tx); inTx {
return err
}
}
return fmt.Errorf("delete auth key: permanent-key binding changed during all retries")
return withAuthIdentityTx(ctx, s.db, "delete auth key", func(tx pgx.Tx) error {
return deleteAuthKeyTx(ctx, tx, authKeyIDToInt64(id))
})
}
func (s *AuthKeyStore) deleteAuthKeyOnce(ctx context.Context, id [8]byte) error {
keyID := authKeyIDToInt64(id)
func deleteAuthKeyTx(ctx context.Context, tx pgx.Tx, keyID int64) error {
if _, _, _, err := lockRawAuthKeyInIdentityOrder(ctx, tx, keyID); err != nil {
if errors.Is(err, store.ErrAuthKeyNotFound) {
return nil
}
return err
}
var touched int
if err := s.db.QueryRow(ctx, `
if err := tx.QueryRow(ctx, `
WITH doomed_temp AS MATERIALIZED (
SELECT temp_auth_key_id
FROM temp_auth_key_bindings
@ -225,13 +247,6 @@ SELECT
const tempAuthKeyPermFKConstraint = "temp_auth_key_bindings_perm_auth_key_id_fkey"
func isPermAuthKeyDeleteRace(err error) bool {
var pgErr *pgconn.PgError
return errors.As(err, &pgErr) &&
pgErr.Code == "23503" &&
pgErr.ConstraintName == tempAuthKeyPermFKConstraint
}
// DeleteOrphaned 回收握手已落库、但从未形成 authorization/temp binding 且当前没有
// 活跃物理连接的旧 auth key。last_used_at 与 Get 的 UPDATE ... RETURNING 行锁配对,封住
// active-key 快照之后新连接开始使用旧 key 的竞态;所有引用条件仍在最终 DELETE 中复核。
@ -248,51 +263,140 @@ func (s *AuthKeyStore) DeleteOrphaned(ctx context.Context, olderThan time.Durati
protectedIDs = append(protectedIDs, authKeyIDToInt64(id))
}
var deleted int
err := s.db.QueryRow(ctx, `
WITH candidates AS MATERIALIZED (
err := withAuthIdentityTx(ctx, s.db, "delete orphaned auth keys", func(tx pgx.Tx) error {
var err error
deleted, err = deleteOrphanedAuthKeysTx(ctx, tx, olderThan, limit, protectedIDs)
return err
})
if err != nil {
return 0, fmt.Errorf("delete orphaned auth keys: %w", err)
}
return deleted, nil
}
func deleteOrphanedAuthKeysTx(
ctx context.Context,
tx pgx.Tx,
olderThan time.Duration,
limit int,
protectedIDs []int64,
) (int, error) {
// Phase 1 is only a bounded hint. It must not lock rows before the complete
// permanent-identity advisory set has been derived and acquired.
rows, err := tx.Query(ctx, `
/* orphan_identity_candidates */
SELECT k.auth_key_id, k.expires_at
FROM auth_keys AS k
WHERE k.last_used_at < now() - make_interval(secs => $1::double precision)
AND NOT (k.auth_key_id = ANY($2::bigint[]))
AND NOT EXISTS (
SELECT 1 FROM authorizations AS a WHERE a.auth_key_id = k.auth_key_id
)
AND NOT EXISTS (
SELECT 1
FROM temp_auth_key_bindings AS b
WHERE b.temp_auth_key_id = k.auth_key_id OR b.perm_auth_key_id = k.auth_key_id
)
ORDER BY k.last_used_at, k.auth_key_id
LIMIT $3`, olderThan.Seconds(), protectedIDs, limit)
if err != nil {
return 0, fmt.Errorf("select orphan auth key candidates: %w", err)
}
candidates := make([]int64, 0, limit)
permanentCandidates := make([]int64, 0, limit)
for rows.Next() {
var (
id int64
expiresAt int
)
if err := rows.Scan(&id, &expiresAt); err != nil {
rows.Close()
return 0, fmt.Errorf("scan orphan auth key candidate: %w", err)
}
candidates = append(candidates, id)
if expiresAt == 0 {
permanentCandidates = append(permanentCandidates, id)
}
}
if err := rows.Err(); err != nil {
rows.Close()
return 0, fmt.Errorf("iterate orphan auth key candidates: %w", err)
}
rows.Close()
if len(candidates) == 0 {
return 0, nil
}
if err := lockPermanentAuthIdentities(ctx, tx, permanentCandidates); err != nil {
return 0, err
}
// Phase 2 locks only the hinted raw rows, in real-ID order, after every P
// gate is held. A temp bind already holding its raw row is skipped; if it
// committed immediately before this lock, phase 3's new READ COMMITTED
// statement sees the binding and excludes it.
lockRows, err := tx.Query(ctx, `
SELECT auth_key_id
FROM auth_keys
WHERE auth_key_id = ANY($1::bigint[])
ORDER BY auth_key_id
FOR UPDATE SKIP LOCKED`, candidates)
if err != nil {
return 0, fmt.Errorf("lock orphan auth key candidates: %w", err)
}
lockedIDs := make([]int64, 0, len(candidates))
for lockRows.Next() {
var id int64
if err := lockRows.Scan(&id); err != nil {
lockRows.Close()
return 0, fmt.Errorf("scan locked orphan auth key: %w", err)
}
lockedIDs = append(lockedIDs, id)
}
if err := lockRows.Err(); err != nil {
lockRows.Close()
return 0, fmt.Errorf("iterate locked orphan auth keys: %w", err)
}
lockRows.Close()
if len(lockedIDs) == 0 {
return 0, nil
}
// Phase 3 is a separate statement snapshot and repeats every ownership,
// activity and protection predicate. Never delete from the phase-1 hint.
var deleted int
err = tx.QueryRow(ctx, `
WITH still_orphaned AS MATERIALIZED (
SELECT k.auth_key_id
FROM auth_keys k
WHERE k.last_used_at < now() - make_interval(secs => $1::double precision)
AND NOT (k.auth_key_id = ANY($2::bigint[]))
AND NOT EXISTS (
SELECT 1 FROM authorizations a WHERE a.auth_key_id = k.auth_key_id
)
AND NOT EXISTS (
SELECT 1
FROM temp_auth_key_bindings b
WHERE b.temp_auth_key_id = k.auth_key_id OR b.perm_auth_key_id = k.auth_key_id
)
ORDER BY k.last_used_at ASC, k.auth_key_id ASC
LIMIT $3
FOR UPDATE OF k SKIP LOCKED
), deleted_update_states AS (
-- Historical authorization-only deletion could leave a cursor without an
-- auth_keys FK. GC owns that stale row once the raw key is proven orphaned.
DELETE FROM update_states s
USING candidates c
WHERE s.auth_key_id = c.auth_key_id
RETURNING s.auth_key_id
), deleted_keys AS (
DELETE FROM auth_keys k
USING candidates c
WHERE k.auth_key_id = c.auth_key_id
FROM auth_keys AS k
WHERE k.auth_key_id = ANY($3::bigint[])
AND k.last_used_at < now() - make_interval(secs => $1::double precision)
AND NOT (k.auth_key_id = ANY($2::bigint[]))
AND NOT EXISTS (
SELECT 1 FROM authorizations a WHERE a.auth_key_id = k.auth_key_id
SELECT 1 FROM authorizations AS a WHERE a.auth_key_id = k.auth_key_id
)
AND NOT EXISTS (
SELECT 1
FROM temp_auth_key_bindings b
FROM temp_auth_key_bindings AS b
WHERE b.temp_auth_key_id = k.auth_key_id OR b.perm_auth_key_id = k.auth_key_id
)
RETURNING k.auth_key_id
), deleted_update_states AS (
DELETE FROM update_states AS state
USING still_orphaned AS orphan
WHERE state.auth_key_id = orphan.auth_key_id
RETURNING state.auth_key_id
), deleted_keys AS (
DELETE FROM auth_keys AS key
USING still_orphaned AS orphan
WHERE key.auth_key_id = orphan.auth_key_id
RETURNING key.auth_key_id
)
SELECT count(*)::int
SELECT count(*)::integer
FROM deleted_keys
CROSS JOIN LATERAL (SELECT count(*) FROM deleted_update_states) AS touched`, olderThan.Seconds(), protectedIDs, limit).Scan(&deleted)
CROSS JOIN LATERAL (SELECT count(*) FROM deleted_update_states) AS touched`,
olderThan.Seconds(), protectedIDs, lockedIDs,
).Scan(&deleted)
if err != nil {
return 0, fmt.Errorf("delete orphaned auth keys: %w", err)
return 0, fmt.Errorf("delete revalidated orphan auth keys: %w", err)
}
return deleted, nil
}