处理出错的 DDL 语句

    目前,TiDB 并不完全兼容所有的 MySQL 语法(详见 DDL 的限制)。当使用 DM 从 MySQL 迁移数据到 TiDB 时,如果 TiDB 不支持对应的 DDL 语句,可能会造成错误并中断迁移任务。在这种情况下,DM 提供 命令来恢复迁移。

    如果业务不能接受下游 TiDB 跳过异常的 DDL 语句,也不接受使用其他 DDL 语句作为替代,则不适合使用此方式进行处理。

    比如:DROP PRIMARY KEY,这种情况下,只能在下游重建一个(DDL 执行完后的)新表结构对应的表,并将原表的全部数据重新导入该新表。

    迁移过程中,上游执行了 TiDB 不支持的 DDL 语句并迁移到了 DM,造成迁移任务中断。

    • 如果业务能接受下游 TiDB 不执行该 DDL 语句,则使用 handle-error <task-name> skip 跳过对该 DDL 语句的迁移以恢复迁移任务。
    • 如果业务能接受下游 TiDB 执行其他 DDL 语句来作为替代,则使用 handle-error <task-name> replace 替代该 DDL 的迁移以恢复迁移任务。

    使用 dmctl 手动处理出错的 DDL 语句时,主要使用的命令包括 query-statushandle-error

    query-status 命令用于查询当前 MySQL 实例内子任务及 relay 单元等的状态和错误信息,详见。

    handle-error

    handle-error 命令用于处理错误的 DDL 语句。

    1. Usage:
    2. dmctl handle-error <task-name | task-file> [-s source ...] [-b binlog-pos] <skip/replace/revert> [replace-sql1;replace-sql2;] [flags]
    3. Flags:
    4. -b, --binlog-pos string position used to match binlog event if matched the handler-error operation will be applied. The format like "mysql-bin|000001.000003:3270"
    5. -h, --help help for handle-error
    6. Global Flags:
    7. -s, --source strings MySQL Source ID

    参数解释

    • task-name

      • 非 flag 参数,string,必选;
      • 指定预设的操作将生效的任务。
    • source

      • flag 参数,string,--source
      • source 指定预设操作将生效的 MySQL 实例。
    • skip:跳过该错误

    • replace:替代错误的 DDL 语句

    • revert:重置该错误先前的 skip/replace 操作, 仅在先前的操作没有最终生效前执行

    • binlog-pos

      • flag 参数,string,--binlog-pos
      • 若不指定,DM 会自动处理当前出错的 DDL 语句
      • 在指定时表示操作将在 binlog-pos 与 binlog event 的 position 匹配时生效,格式为 binlog-filename:binlog-pos,如 mysql-bin|000001.000003:3270
      • 在迁移执行出错后,binlog position 可直接从 query-status 返回的 startLocation 中的 position 获得;在迁移执行出错前,binlog position 可在上游 MySQL 中使用 获得。

    非合库合表场景

    假设现在需要将上游的 db1.tbl1 表迁移到下游 TiDB,初始时表结构为:

    1. SHOW CREATE TABLE db1.tbl1;
    1. +-------+--------------------------------------------------+
    2. | Table | Create Table |
    3. +-------+--------------------------------------------------+
    4. | tbl1 | CREATE TABLE `tbl1` (
    5. `c1` int(11) NOT NULL,
    6. `c2` decimal(11,3) DEFAULT NULL,
    7. PRIMARY KEY (`c1`)
    8. ) ENGINE=InnoDB DEFAULT CHARSET=latin1 |
    9. +-------+--------------------------------------------------+

    此时,上游执行以下 DDL 操作修改表结构(将列的 DECIMAL(11, 3) 修改为 DECIMAL(10, 3)):

    1. ALTER TABLE db1.tbl1 CHANGE c2 c2 DECIMAL (10, 3);
    1. ERROR 8200 (HY000): Unsupported modify column: can't change decimal column precision

    假设业务上可以接受下游 TiDB 不执行此 DDL 语句(即继续保持原有的表结构),则可以通过使用 handle-error <task-name> skip 命令跳过该 DDL 语句以恢复迁移任务。操作步骤如下:

    1. 使用 handle-error <task-name> skip 跳过当前错误的 DDL 语句

      1. » handle-error test skip
      1. {
      2. "result": true,
      3. "msg": "",
      4. "sources": [
      5. {
      6. "result": true,
      7. "msg": "",
      8. "source": "mysql-replica-01",
      9. "worker": "worker1"
      10. }
      11. ]
      12. }
    2. 使用 query-status <task-name> 查看任务状态

      1. » query-status test
      执行结果 { "result": true, "msg": "", "sources": [ { "result": true, "msg": "", "sourceStatus": { "source": "mysql-replica-01", "worker": "worker1", "result": null, "relayStatus": null }, "subTaskStatus": [ { "name": "test", "stage": "Running", "unit": "Sync", "result": null, "unresolvedDDLLockID": "", "sync": { "totalEvents": "4", "totalTps": "0", "recentTps": "0", "masterBinlog": "(DESKTOP-T561TSO-bin.000001, 2388)", "masterBinlogGtid": "143bdef3-dd4a-11ea-8b00-00155de45f57:1-10", "syncerBinlog": "(DESKTOP-T561TSO-bin.000001, 2388)", "syncerBinlogGtid": "143bdef3-dd4a-11ea-8b00-00155de45f57:1-4", "blockingDDLs": [ ], "unresolvedGroups": [ ], "synced": true, "binlogType": "remote" } } ] } ] }

      可以看到任务运行正常,错误的 DDL 被跳过。

    合库合表场景

    假设现在存在如下四个上游表需要合并迁移到下游的同一个表 `shard_db`.`shard_table` ,任务模式为悲观协调模式:

    • MySQL 实例 1 内有 shard_db_1 库,包括 shard_table_1shard_table_2 两个表。
    • MySQL 实例 2 内有 shard_db_2 库,包括 shard_table_1shard_table_2 两个表。

    初始时表结构为:

    1. SHOW CREATE TABLE shard_db.shard_table;
    1. +-------+-----------------------------------------------------------------------------------------------------------+
    2. | Table | Create Table |
    3. +-------+-----------------------------------------------------------------------------------------------------------+
    4. | tb | CREATE TABLE `shard_table` (
    5. `id` int(11) DEFAULT NULL,
    6. PRIMARY KEY (`id`)
    7. ) ENGINE=InnoDB DEFAULT CHARSET=latin1 COLLATE=latin1_bin |
    8. +-------+-----------------------------------------------------------------------------------------------------------+

    此时,在上游所有分表上都执行以下 DDL 操作修改表字符集

      则会由于 TiDB 不支持该 DDL 语句而导致 DM 迁移任务中断,使用 query-status 命令可以看到 MySQL 实例 1 的 shard_db_1.shard_table_1 表和 MySQL 实例 2 的 shard_db_2.shard_table_1 表报错:

      1. {
      2. "RawCause": "[ddl:8200]Unsupported modify charset from latin1 to utf8"
      3. }
      1. {
      2. "Message": "cannot track DDL: ALTER TABLE `shard_db_2`.`shard_table_1` CHARACTER SET UTF8 COLLATE UTF8_UNICODE_CI",
      3. "RawCause": "[ddl:8200]Unsupported modify charset from latin1 to utf8"
      4. }

      假设业务上可以接受下游 TiDB 不执行此 DDL 语句(即继续保持原有的表结构),则可以通过使用 handle-error <task-name> skip 命令跳过该 DDL 语句以恢复迁移任务。操作步骤如下:

      1. 使用 handle-error <task-name> skip 跳过 MySQL 实例 1 和实例 2 当前错误的 DDL 语句

        1. {
        2. "result": true,
        3. "msg": "",
        4. "sources": [
        5. {
        6. "result": true,
        7. "msg": "",
        8. "source": "mysql-replica-01",
        9. "worker": "worker1"
        10. },
        11. {
        12. "result": true,
        13. "msg": "",
        14. "source": "mysql-replica-02",
        15. "worker": "worker2"
        16. }
        17. ]
        18. }
      2. 使用 query-status <task-name> 查看任务状态,可以看到 MySQL 实例 1 的 shard_db_1.shard_table_2 表和 MySQL 实例 2 的 shard_db_2.shard_table_2 表报错:

        1. {
        2. "Message": "cannot track DDL: ALTER TABLE `shard_db_1`.`shard_table_2` CHARACTER SET UTF8 COLLATE UTF8_UNICODE_CI",
        3. "RawCause": "[ddl:8200]Unsupported modify charset from latin1 to utf8"
        4. }
        1. {
        2. "Message": "cannot track DDL: ALTER TABLE `shard_db_2`.`shard_table_2` CHARACTER SET UTF8 COLLATE UTF8_UNICODE_CI",
        3. "RawCause": "[ddl:8200]Unsupported modify charset from latin1 to utf8"
        4. }
      3. 继续使用 handle-error <task-name> skip 跳过 MySQL 实例 1 和实例 2 当前错误的 DDL 语句

        1. » handle-error test skip
        1. {
        2. "result": true,
        3. "msg": "",
        4. "sources": [
        5. {
        6. "result": true,
        7. "msg": "",
        8. "source": "mysql-replica-01",
        9. "worker": "worker1"
        10. },
        11. {
        12. "result": true,
        13. "msg": "",
        14. "source": "mysql-replica-02",
        15. "worker": "worker2"
        16. }
        17. ]
        18. }
      4. 使用 query-status <task-name> 查看任务状态

        1. » query-status test
        执行结果 { "result": true, "msg": "", "sources": [ { "result": true, "msg": "", "sourceStatus": { "source": "mysql-replica-01", "worker": "worker1", "result": null, "relayStatus": null }, "subTaskStatus": [ { "name": "test", "stage": "Running", "unit": "Sync", "result": null, "unresolvedDDLLockID": "", "sync": { "totalEvents": "4", "totalTps": "0", "recentTps": "0", "masterBinlog": "(DESKTOP-T561TSO-bin.000001, 2388)", "masterBinlogGtid": "143bdef3-dd4a-11ea-8b00-00155de45f57:1-10", "syncerBinlog": "(DESKTOP-T561TSO-bin.000001, 2388)", "syncerBinlogGtid": "143bdef3-dd4a-11ea-8b00-00155de45f57:1-4", "blockingDDLs": [ ], "unresolvedGroups": [ ], "synced": true, "binlogType": "remote" } } ] }, { "result": true, "msg": "", "sourceStatus": { "source": "mysql-replica-02", "worker": "worker2", "result": null, "relayStatus": null }, "subTaskStatus": [ { "name": "test", "stage": "Running", "unit": "Sync", "result": null, "unresolvedDDLLockID": "", "sync": { "totalEvents": "4", "totalTps": "0", "recentTps": "0", "masterBinlog": "(DESKTOP-T561TSO-bin.000001, 2388)", "masterBinlogGtid": "143bdef3-dd4a-11ea-8b00-00155de45f57:1-10", "syncerBinlog": "(DESKTOP-T561TSO-bin.000001, 2388)", "syncerBinlogGtid": "143bdef3-dd4a-11ea-8b00-00155de45f57:1-4", "blockingDDLs": [ ], "unresolvedGroups": [ ], "synced": true, "binlogType": "remote" } } ] } ] }

        可以看到任务运行正常,无错误信息。四条 DDL 全部被跳过。

      迁移中断执行替代操作

      非合库合表场景

      假设现在需要将上游的 db1.tbl1 表迁移到下游 TiDB,初始时表结构为:

      1. SHOW CREATE TABLE db1.tbl1;
      1. +-------+-----------------------------------------------------------------------------------------------------------+
      2. | Table | Create Table |
      3. +-------+-----------------------------------------------------------------------------------------------------------+
      4. | tb | CREATE TABLE `tbl1` (
      5. `id` int(11) DEFAULT NULL,
      6. PRIMARY KEY (`id`)
      7. ) ENGINE=InnoDB DEFAULT CHARSET=latin1 COLLATE=latin1_bin |
      8. +-------+-----------------------------------------------------------------------------------------------------------+
      1. ALTER TABLE `db1`.`tbl1` ADD COLUMN new_col INT UNIQUE;

      则会由于 TiDB 不支持该 DDL 语句而导致 DM 迁移任务中断,使用 query-status 命令可看到如下错误:

      1. {
      2. "Message": "cannot track DDL: ALTER TABLE `db1`.`tbl1` ADD COLUMN `new_col` INT UNIQUE KEY",
      3. "RawCause": "[ddl:8200]unsupported add column 'new_col' constraint UNIQUE KEY when altering 'db1.tbl1'",
      4. }

      我们将该 DDL 替换成两条等价的 DDL。操作步骤如下:

      1. 使用如下命令替换错误的 DDL 语句

        1. » handle-error test replace "ALTER TABLE `db1`.`tbl1` ADD COLUMN `new_col` INT;ALTER TABLE `db1`.`tbl1` ADD UNIQUE(`new_col`)";
        1. {
        2. "result": true,
        3. "msg": "",
        4. "sources": [
        5. "result": true,
        6. "msg": "",
        7. "source": "mysql-replica-01",
        8. }
        9. ]
        10. }
      2. 使用 query-status <task-name> 查看任务状态

        1. » query-status test
        执行结果 { "result": true, "msg": "", "sources": [ { "result": true, "msg": "", "sourceStatus": { "source": "mysql-replica-01", "worker": "worker1", "result": null, "relayStatus": null }, "subTaskStatus": [ { "name": "test", "stage": "Running", "unit": "Sync", "result": null, "unresolvedDDLLockID": "", "sync": { "totalEvents": "4", "totalTps": "0", "recentTps": "0", "masterBinlog": "(DESKTOP-T561TSO-bin.000001, 2388)", "masterBinlogGtid": "143bdef3-dd4a-11ea-8b00-00155de45f57:1-10", "syncerBinlog": "(DESKTOP-T561TSO-bin.000001, 2388)", "syncerBinlogGtid": "143bdef3-dd4a-11ea-8b00-00155de45f57:1-4", "blockingDDLs": [ ], "unresolvedGroups": [ ], "synced": true, "binlogType": "remote" } } ] } ] }

        可以看到任务运行正常,错误的 DDL 已被替换且执行成功。

      合库合表场景

      假设现在存在如下四个上游表需要合并迁移到下游的同一个表 `shard_db`.`shard_table` ,任务模式为悲观协调模式:

      • MySQL 实例 1 内有 shard_db_1 库,包括 shard_table_1shard_table_2 两个表。
      • MySQL 实例 2 内有 shard_db_2 库,包括 shard_table_1shard_table_2 两个表。

      初始时表结构为:

      1. +-------+-----------------------------------------------------------------------------------------------------------+
      2. | Table | Create Table |
      3. +-------+-----------------------------------------------------------------------------------------------------------+
      4. | tb | CREATE TABLE `shard_table` (
      5. `id` int(11) DEFAULT NULL,
      6. PRIMARY KEY (`id`)
      7. ) ENGINE=InnoDB DEFAULT CHARSET=latin1 COLLATE=latin1_bin |
      8. +-------+-----------------------------------------------------------------------------------------------------------+

      此时,在上游所有分表上都执行以下 DDL 操作增加新列,并添加 UNIQUE 约束:

      1. ALTER TABLE `shard_db_*`.`shard_table_*` ADD COLUMN new_col INT UNIQUE;

      则会由于 TiDB 不支持该 DDL 语句而导致 DM 迁移任务中断,使用 query-status 命令可以看到 MySQL 实例 1 的 shard_db_1.shard_table_1 表和 MySQL 实例 2 的 shard_db_2.shard_table_1 表报错:

      1. {
      2. "Message": "cannot track DDL: ALTER TABLE `shard_db_1`.`shard_table_1` ADD COLUMN `new_col` INT UNIQUE KEY",
      3. "RawCause": "[ddl:8200]unsupported add column 'new_col' constraint UNIQUE KEY when altering 'shard_db_1.shard_table_1'",
      4. }
      1. {
      2. "Message": "cannot track DDL: ALTER TABLE `shard_db_2`.`shard_table_1` ADD COLUMN `new_col` INT UNIQUE KEY",
      3. "RawCause": "[ddl:8200]unsupported add column 'new_col' constraint UNIQUE KEY when altering 'shard_db_2.shard_table_1'",
      4. }

      我们将该 DDL 替换成两条等价的 DDL。操作步骤如下:

      1. 使用如下命令分别替换 MySQL 实例 1 和实例 2 中错误的 DDL 语句

        1. » handle-error test -s mysql-replica-01 replace "ALTER TABLE `shard_db_1`.`shard_table_1` ADD COLUMN `new_col` INT;ALTER TABLE `shard_db_1`.`shard_table_1` ADD UNIQUE(`new_col`)";
        1. {
        2. "result": true,
        3. "msg": "",
        4. "sources": [
        5. {
        6. "result": true,
        7. "msg": "",
        8. "source": "mysql-replica-01",
        9. "worker": "worker1"
        10. }
        11. ]
        12. }
        1. » handle-error test -s mysql-replica-02 replace "ALTER TABLE `shard_db_2`.`shard_table_1` ADD COLUMN `new_col` INT;ALTER TABLE `shard_db_2`.`shard_table_1` ADD UNIQUE(`new_col`)";
        1. {
        2. "result": true,
        3. "msg": "",
        4. "sources": [
        5. {
        6. "result": true,
        7. "msg": "",
        8. "source": "mysql-replica-02",
        9. "worker": "worker2"
        10. }
        11. ]
        12. }
      2. 使用 query-status <task-name> 查看任务状态,可以看到 MySQL 实例 1 的 shard_db_1.shard_table_2 表和 MySQL 实例 2 的 shard_db_2.shard_table_2 表报错:

        1. {
        2. "Message": "detect inconsistent DDL sequence from source ... ddls: [ALTER TABLE `shard_db`.`tb` ADD COLUMN `new_col` INT UNIQUE KEY] source: `shard_db_1`.`shard_table_2`], right DDL sequence should be ..."
        3. }
        1. {
        2. "Message": "detect inconsistent DDL sequence from source ... ddls: [ALTER TABLE `shard_db`.`tb` ADD COLUMN `new_col` INT UNIQUE KEY] source: `shard_db_2`.`shard_table_2`], right DDL sequence should be ..."
        3. }
      3. 使用如下命令继续分别替换 MySQL 实例 1 和实例 2 中错误的 DDL 语句

        1. » handle-error test -s mysql-replica-01 replace "ALTER TABLE `shard_db_1`.`shard_table_2` ADD COLUMN `new_col` INT;ALTER TABLE `shard_db_1`.`shard_table_2` ADD UNIQUE(`new_col`)";
        1. {
        2. "result": true,
        3. "msg": "",
        4. "sources": [
        5. {
        6. "result": true,
        7. "msg": "",
        8. "source": "mysql-replica-01",
        9. "worker": "worker1"
        10. }
        11. ]
        12. }
        1. » handle-error test -s mysql-replica-02 replace "ALTER TABLE `shard_db_2`.`shard_table_2` ADD COLUMN `new_col` INT;ALTER TABLE `shard_db_2`.`shard_table_2` ADD UNIQUE(`new_col`)";
      4. 使用 query-status <task-name> 查看任务状态

          执行结果 { "result": true, "msg": "", "sources": [ { "result": true, "msg": "", "sourceStatus": { "source": "mysql-replica-01", "worker": "worker1", "result": null, "relayStatus": null }, "subTaskStatus": [ { "name": "test", "stage": "Running", "unit": "Sync", "result": null, "unresolvedDDLLockID": "", "sync": { "totalEvents": "4", "totalTps": "0", "recentTps": "0", "masterBinlog": "(DESKTOP-T561TSO-bin.000001, 2388)", "masterBinlogGtid": "143bdef3-dd4a-11ea-8b00-00155de45f57:1-10", "syncerBinlog": "(DESKTOP-T561TSO-bin.000001, 2388)", "syncerBinlogGtid": "143bdef3-dd4a-11ea-8b00-00155de45f57:1-4", "blockingDDLs": [ ], "unresolvedGroups": [ ], "unresolvedGroups": [ ], "synced": true, "binlogType": "remote" } } ] }, { "result": true, "msg": "", "sourceStatus": { "source": "mysql-replica-02", "worker": "worker2", "result": null, "relayStatus": null }, "subTaskStatus": [ { "name": "test", "stage": "Running", "unit": "Sync", "result": null, "unresolvedDDLLockID": "", "sync": { "totalEvents": "4", "totalTps": "0", "recentTps": "0", "masterBinlog": "(DESKTOP-T561TSO-bin.000001, 2388)", "masterBinlogGtid": "143bdef3-dd4a-11ea-8b00-00155de45f57:1-10", "syncerBinlog": "(DESKTOP-T561TSO-bin.000001, 2388)", "syncerBinlogGtid": "143bdef3-dd4a-11ea-8b00-00155de45f57:1-4", "blockingDDLs": [ ], "unresolvedGroups": [ ], "unresolvedGroups": [ ], "synced": try, "binlogType": "remote" } } ] } ] }

          可以看到任务运行正常,无错误信息。四条 DDL 全部被替换。