refactor(mysql): 适配PXC严格模式，替换全局锁方案

重构MySQL迁移逻辑，新增acquireMigrationLock适配Percona XtraDB Cluster的pxc_strict_mode限制，将原有的GET_LOCK迁移锁替换为更兼容的方案，同时为各迁移流程添加快速检查路径减少锁竞争

server/db/mysql/migration_lock.go

@@ -0,0 +1,36 @@
+//go:build mysql
+// +build mysql
+
+package mysql
+
+import (
+	"context"
+	"database/sql"
+	"errors"
+
+	"github.com/jmoiron/sqlx"
+)
+
+// acquireMigrationLock wraps SELECT GET_LOCK so that schema migrations keep
+// working on Percona XtraDB Cluster where pxc_strict_mode=ENFORCING prohibits
+// named locks: instead of failing, locked=false is returned and the caller
+// proceeds without the lock. This is safe because Galera's total order
+// isolation serializes DDL across the cluster, and every migration using this
+// helper is idempotent: steps are guarded by existence checks, so concurrent
+// runs at worst fail with duplicate-key errors and are retried on the next
+// startup.
+func acquireMigrationLock(ctx context.Context, conn *sqlx.Conn, lockExpr string) (locked bool, unlock func(), err error) {
+	var held sql.NullInt64
+	if err = conn.QueryRowContext(ctx, "SELECT GET_LOCK("+lockExpr+", 60)").Scan(&held); err != nil {
+		if isPXCStrictModeNamedLockError(err) {
+			return false, nil, nil
+		}
+		return false, nil, err
+	}
+	if !held.Valid || held.Int64 != 1 {
+		return false, nil, errors.New("mysql: timed out waiting for schema migration lock")
+	}
+	return true, func() {
+		conn.ExecContext(ctx, "SELECT RELEASE_LOCK("+lockExpr+")")
+	}, nil
+}

server/db/mysql/snowflake_primary_keys.go

@@ -58,15 +58,16 @@ func ensureSnowflakePrimaryKeys(db *sqlx.DB) error {
 	}
 	defer conn.Close()
 
-	var locked sql.NullInt64
+	// PXC strict-mode clusters prohibit GET_LOCK; acquireMigrationLock falls back
-	if err = conn.QueryRowContext(ctx,
+	// to lock-free execution in that case. migrateSnowflakePrimaryKey checks the
-		"SELECT GET_LOCK("+snowflakeMigrationLock+", 60)").Scan(&locked); err != nil {
+	// current state of every table before touching it, so it is idempotent.
+	_, unlock, err := acquireMigrationLock(ctx, conn, snowflakeMigrationLock)
+	if err != nil {
 		return err
 	}
-	if !locked.Valid || locked.Int64 != 1 {
+	if unlock != nil {
-		return errors.New("mysql: timed out waiting for snowflake primary-key migration lock")
+		defer unlock()
 	}
-	defer conn.ExecContext(ctx, "SELECT RELEASE_LOCK("+snowflakeMigrationLock+")")
 
 	for _, spec := range snowflakePrimaryKeySpecs {
 		if err = migrateSnowflakePrimaryKey(ctx, conn, spec); err != nil {

server/db/mysql/user_identity_schema.go

@@ -5,7 +5,6 @@ package mysql
 
 import (
 	"context"
-	"database/sql"
 	"errors"
 	"fmt"
 	"strings"
@@ -32,14 +31,35 @@ func ensureUserIdentitySchema(db *sqlx.DB) error {
 	}
 	defer conn.Close()
 
-	var locked sql.NullInt64
+	// Fast path: the last two steps of this migration drop the legacy im_usertags
-	if err = conn.QueryRowContext(ctx, "SELECT GET_LOCK("+userIdentityMigrationLock+", 60)").Scan(&locked); err != nil {
+	// table and the im_users.tags column. If both are already gone, the migration
+	// has fully run (or the database was created fresh) and no lock is needed.
+	stale := false
+	if exists, err := identityTableExists(ctx, conn, "im_usertags"); err != nil {
+		return err
+	} else if exists {
+		stale = true
+	}
+	if !stale {
+		if exists, err := identityColumnExists(ctx, conn, "im_users", "tags"); err != nil {
+			return err
+		} else if exists {
+			stale = true
+		}
+	}
+	if !stale {
+		return nil
+	}
+
+	// PXC strict-mode clusters prohibit GET_LOCK; acquireMigrationLock falls back
+	// to lock-free execution in that case. The migration body is idempotent.
+	_, unlock, err := acquireMigrationLock(ctx, conn, userIdentityMigrationLock)
+	if err != nil {
 		return err
 	}
-	if !locked.Valid || locked.Int64 != 1 {
+	if unlock != nil {
-		return errors.New("mysql: timed out waiting for user identity migration lock")
+		defer unlock()
 	}
-	defer conn.ExecContext(ctx, "SELECT RELEASE_LOCK("+userIdentityMigrationLock+")")
 
 	usersExist, err := identityTableExists(ctx, conn, "im_users")
 	if err != nil || !usersExist {

server/db/mysql/user_privacy_schema.go

@@ -5,8 +5,6 @@ package mysql
 
 import (
 	"context"
-	"database/sql"
-	"errors"
 
 	"github.com/jmoiron/sqlx"
 )
@@ -26,22 +24,40 @@ func ensureUserPrivacySchema(db *sqlx.DB) error {
 	}
 	defer conn.Close()
 
-	var locked sql.NullInt64
+	// Fast path: check for missing columns without taking the migration lock.
-	if err = conn.QueryRowContext(ctx, "SELECT GET_LOCK("+userPrivacyMigrationLock+", 60)").Scan(&locked); err != nil {
+	// Databases created by createIMUserProductTables already have every column,
+	// so this is the common case and avoids GET_LOCK entirely (prohibited on
+	// Percona XtraDB Cluster with pxc_strict_mode=ENFORCING).
+	added, err := ensureUserPrivacyColumns(ctx, conn)
+	if err != nil || added {
 		return err
 	}
-	if !locked.Valid || locked.Int64 != 1 {
-		return errors.New("mysql: timed out waiting for user privacy migration lock")
-	}
-	defer conn.ExecContext(ctx, "SELECT RELEASE_LOCK("+userPrivacyMigrationLock+")")
 
+	_, unlock, err := acquireMigrationLock(ctx, conn, userPrivacyMigrationLock)
+	if err != nil {
+		return err
+	}
+	if unlock != nil {
+		defer unlock()
+	}
+
+	// Double-check under the lock: another instance may have migrated while we
+	// were waiting. Without a lock (PXC fallback) this simply runs idempotently.
+	_, err = ensureUserPrivacyColumns(ctx, conn)
+	return err
+}
+
+// ensureUserPrivacyColumns adds any columns of im_user_privacy_settings that
+// are missing and reports whether nothing was missing. A missing table is
+// skipped: it is created by the fresh-install path and there is nothing to
+// upgrade yet. Column definitions mirror createIMUserProductTables; the SELECT
+// in userPrivacyByUID references every one of them.
+func ensureUserPrivacyColumns(ctx context.Context, conn *sqlx.Conn) (bool, error) {
 	tableExists, err := identityTableExists(ctx, conn, "im_user_privacy_settings")
 	if err != nil || !tableExists {
-		return err
+		return true, err
 	}
 
-	// Column definitions mirror createIMUserProductTables; the SELECT in
-	// userPrivacyByUID references every one of them.
 	columns := []struct{ name, ddl string }{
 		{"profile_visibility", "VARCHAR(32) NOT NULL DEFAULT 'friends'"},
 		{"last_seen_visibility", "VARCHAR(32) NOT NULL DEFAULT 'friends'"},
@@ -49,16 +65,19 @@ func ensureUserPrivacySchema(db *sqlx.DB) error {
 		{"friend_request_verification", "TINYINT NOT NULL DEFAULT 1"},
 		{"group_invite_verification", "TINYINT NOT NULL DEFAULT 0"},
 	}
+	missing := false
 	for _, col := range columns {
 		exists, err := identityColumnExists(ctx, conn, "im_user_privacy_settings", col.name)
 		if err != nil {
-			return err
+			return false, err
 		}
-		if !exists {
+		if exists {
-			if _, err = conn.ExecContext(ctx, "ALTER TABLE im_user_privacy_settings ADD COLUMN "+col.name+" "+col.ddl); err != nil {
+			continue
-				return err
+		}
-			}
+		missing = true
+		if _, err = conn.ExecContext(ctx, "ALTER TABLE im_user_privacy_settings ADD COLUMN "+col.name+" "+col.ddl); err != nil {
+			return false, err
 		}
 	}
-	return nil
+	return !missing, nil
 }