Important

You are viewing the archived documentation of TiDB, which no longer receives updates. It is recommended that you use the latest LTS version of the TiDB database.

Key Features of TiDB Data Migration

This document describes the data migration features provided by TiDB Data Migration (DM) and introduces appropriate parameter configurations.

For different DM versions, pay attention to the different match rules of schema or table names in the table routing, block & allow lists, and binlog event filter features:

For DM v1.0.5 or later versions, all the above features support the wildcard match. For all versions of DM, note that there can be only one * in the wildcard expression, and * must be placed at the end.
For DM versions earlier than v1.0.5, table routing and binlog event filter support the wildcard but do not support the [...] and [!...] expressions. The block & allow lists only supports the regular expression.

It is recommended that you use the wildcard for matching in simple scenarios.

Table routing

The table routing feature enables DM to migrate a certain table of the upstream MySQL or MariaDB instance to the specified table in the downstream.

Note

Configuring multiple different routing rules for a single table is not supported.
The match rule of schema needs to be configured separately, which is used to migrate CREATE/DROP SCHEMA xx, as shown in rule-2 of the parameter configuration.

Parameter configuration

routes:
  rule-1:
    schema-pattern: "test_*"
    table-pattern: "t_*"
    target-schema: "test"
    target-table: "t"
  rule-2:
    schema-pattern: "test_*"
    target-schema: "test"

Parameter explanation

DM migrates the upstream MySQL or MariaDB instance table that matches the schema-pattern/table-pattern rule provided by Table selector to the downstream target-schema/target-table.

Usage examples

This section shows the usage examples in different scenarios.

Merge sharded schemas and tables

Assuming in the scenario of sharded schemas and tables, you want to migrate the test_{1,2,3...}.t_{1,2,3...} tables in two upstream MySQL instances to the test.t table in the downstream TiDB instance.

To migrate the upstream instances to the downstream test.t, you must create the following routing rules:

rule-1 is used to migrate DML or DDL statements of the table that matches schema-pattern: "test_*" and table-pattern: "t_*" to the downstream test.t.
rule-2 is used to migrate DDL statements of the schema that matches schema-pattern: "test_*", such as CREATE/DROP SCHEMA xx.

Note

If the downstream schema: test already exists and is not to be deleted, you can omit rule-2.
If the downstream schema: test does not exist and only rule-1 is configured, then it reports the schema test doesn't exist error during migration.

  rule-1:
    schema-pattern: "test_*"
    table-pattern: "t_*"
    target-schema: "test"
    target-table: "t"
  rule-2:
    schema-pattern: "test_*"
    target-schema: "test"

Merge sharded schemas

Assuming in the scenario of sharded schemas, you want to migrate the test_{1,2,3...}.t_{1,2,3...} tables in the two upstream MySQL instances to the test.t_{1,2,3...} tables in the downstream TiDB instance.

To migrate the upstream schemas to the downstream test.t_[1,2,3], you only need to create one routing rule.

  rule-1:
    schema-pattern: "test_*"
    target-schema: "test"

Incorrect table routing

Assuming that the following two routing rules are configured and test_1_bak.t_1_bak matches both rule-1 and rule-2, an error is reported because the table routing configuration violates the number limitation.

  rule-1:
    schema-pattern: "test_*"
    table-pattern: "t_*"
    target-schema: "test"
    target-table: "t"
  rule-2:
    schema-pattern: "test_1_bak"
    table-pattern: "t_1_bak"
    target-schema: "test"
    target-table: "t_bak"

Block and allow table lists

The block and allow lists filtering rule of the upstream database instance tables is similar to MySQL replication-rules-db/tables, which can be used to filter or only migrate all operations of some databases or some tables.

Parameter configuration

block-allow-list:             # Use black-white-list if the DM version is earlier than or equal to v2.0.0-beta.2.
  rule-1:
    do-dbs: ["test*"]         # Starting with characters other than "~" indicates that it is a wildcard;
                              # v1.0.5 or later versions support the regular expression rules.
    do-tables:
    - db-name: "test[123]"    # Matches test1, test2, and test3.
      tbl-name: "t[1-5]"      # Matches t1, t2, t3, t4, and t5.
    - db-name: "test"
      tbl-name: "t"
  rule-2:
    do-dbs: ["~^test.*"]      # Starting with "~" indicates that it is a regular expression.
    ignore-dbs: ["mysql"]
    do-tables:
    - db-name: "~^test.*"
      tbl-name: "~^t.*"
    - db-name: "test"
      tbl-name: "t"
    ignore-tables:
    - db-name: "test"
      tbl-name: "log"

Parameter explanation

do-dbs: allow lists of the schemas to be migrated, similar to replicate-do-db in MySQL
ignore-dbs: block lists of the schemas to be migrated, similar to replicate-ignore-db in MySQL
do-tables: allow lists of the tables to be migrated, similar to replicate-do-table in MySQL. Both db-name and tbl-name must be specified
ignore-tables: block lists of the tables to be migrated, similar to replicate-ignore-table in MySQL. Both db-name and tbl-name must be specified

If a value of the above parameters starts with the ~ character, the subsequent characters of this value are treated as a regular expression. You can use this parameter to match schema or table names.

Filtering process

The filtering rules corresponding to do-dbs and ignore-dbs are similar to the Evaluation of Database-Level Replication and Binary Logging Options in MySQL. The filtering rules corresponding to do-tables and ignore-tables are similar to the Evaluation of Table-Level Replication Options in MySQL.

Note

In DM and in MySQL, the allow and block lists filtering rules are different in the following ways:

In MySQL, replicate-wild-do-table and replicate-wild-ignore-table support wildcard characters. In DM, some parameter values directly supports regular expressions that start with the ~ character.
DM currently only supports binlogs in the ROW format, and does not support those in the STATEMENT or MIXED format. Therefore, the filtering rules in DM correspond to those in the ROW format in MySQL.
MySQL determines a DDL statement only by the database name explicitly specified in the USE section of the statement. DM determines a statement first based on the database name section in the DDL statement. If the DDL statement does not contain such a section, DM determines the statement by the USE section. Suppose that the SQL statement to be determined is USE test_db_2; CREATE TABLE test_db_1.test_table (c1 INT PRIMARY KEY); that replicate-do-db=test_db_1 is configured in MySQL and do-dbs: ["test_db_1"] is configured in DM. Then this rule only applies to DM and not to MySQL.

The filtering process is as follows:

Filter at the schema level:
- If do-dbs is not empty, judge whether a matched schema exists in do-dbs.
  - If yes, continue to filter at the table level.
  - If not, filter test.t.
- If do-dbs is empty and ignore-dbs is not empty, judge whether a matched schema exits in ignore-dbs.
  - If yes, filter test.t.
  - If not, continue to filter at the table level.
- If both do-dbs and ignore-dbs are empty, continue to filter at the table level.
Filter at the table level:
1. If do-tables is not empty, judge whether a matched table exists in do-tables.
  - If yes, migrate test.t.
  - If not, filter test.t.
2. If ignore-tables is not empty, judge whether a matched table exists in ignore-tables.
  - If yes, filter test.t.
  - If not, migrate test.t.
3. If both do-tables and ignore-tables are empty, migrate test.t.

Note

To judge whether the schema test should be filtered, you only need to filter at the schema level.

Usage example

Assume that the upstream MySQL instances include the following tables:

`logs`.`messages_2016`
`logs`.`messages_2017`
`logs`.`messages_2018`
`forum`.`users`
`forum`.`messages`
`forum_backup_2016`.`messages`
`forum_backup_2017`.`messages`
`forum_backup_2018`.`messages`

The configuration is as follows:

block-allow-list:  # Use black-white-list if the DM version is earlier than or equal to v2.0.0-beta.2.
  bw-rule:
    do-dbs: ["forum_backup_2018", "forum"]
    ignore-dbs: ["~^forum_backup_"]
    do-tables:
    - db-name: "logs"
      tbl-name: "~_2018$"
    - db-name: "~^forum.*"
      tbl-name: "messages"
    ignore-tables:
    - db-name: "~.*"
      tbl-name: "^messages.*"

After using the bw-rule rule:

Table	Whether to filter	Why filter
`logs`.`messages_2016`	Yes	The schema `logs` fails to match any `do-dbs`.
`logs`.`messages_2017`	Yes	The schema `logs` fails to match any `do-dbs`.
`logs`.`messages_2018`	Yes	The schema `logs` fails to match any `do-dbs`.
`forum_backup_2016`.`messages`	Yes	The schema `forum_backup_2016` fails to match any `do-dbs`.
`forum_backup_2017`.`messages`	Yes	The schema `forum_backup_2017` fails to match any `do-dbs`.
`forum`.`users`	Yes	1. The schema `forum` matches `do-dbs` and continues to filter at the table level. 2. The schema and table fail to match any of `do-tables` and `ignore-tables` and `do-tables` is not empty.
`forum`.`messages`	No	1. The schema `forum` matches `do-dbs` and continues to filter at the table level. 2. The table `messages` is in the `db-name: "~^forum.*",tbl-name: "messages"` of `do-tables`.
`forum_backup_2018`.`messages`	No	1. The schema `forum_backup_2018` matches `do-dbs` and continues to filter at the table level. 2. The schema and table match the `db-name: "~^forum.*",tbl-name: "messages"` of `do-tables`.

Binlog event filter

Binlog event filter is a more fine-grained filtering rule than the block and allow lists filtering rule. You can use statements like INSERT or TRUNCATE TABLE to specify the binlog events of schema/table that you need to migrate or filter out.

Note

If the same table matches multiple rules, these rules are applied in order and the block list has priority over the allow list. This means if both the Ignore and Do rules are applied to a table, the Ignore rule takes effect.
Starting from DM v2.0.2, you can configure binlog event filters in the source configuration file. For details, see Upstream Database Configuration File.

Parameter configuration

filters:
  rule-1:
    schema-pattern: "test_*"
    table-pattern: "t_*"
    events: ["truncate table", "drop table"]
    sql-pattern: ["^DROP\\s+PROCEDURE", "^CREATE\\s+PROCEDURE"]
    action: Ignore

Parameter explanation

schema-pattern/table-pattern: the binlog events or DDL SQL statements of upstream MySQL or MariaDB instance tables that match schema-pattern/table-pattern are filtered by the rules below.

events: the binlog event array. You can only select one or more Events from the following table:

Events	Type	Description
`all`		Includes all the events below
`all dml`		Includes all DML events below
`all ddl`		Includes all DDL events below
`none`		Includes none of the events below
`none ddl`		Includes none of the DDL events below
`none dml`		Includes none of the DML events below
`insert`	DML	The `INSERT` DML event
`update`	DML	The `UPDATE` DML event
`delete`	DML	The `DELETE` DML event
`create database`	DDL	The `CREATE DATABASE` DDL event
`drop database`	DDL	The `DROP DATABASE` DDL event
`create table`	DDL	The `CREATE TABLE` DDL event
`create index`	DDL	The `CREATE INDEX` DDL event
`drop table`	DDL	The `DROP TABLE` DDL event
`truncate table`	DDL	The `TRUNCATE TABLE` DDL event
`rename table`	DDL	The `RENAME TABLE` DDL event
`drop index`	DDL	The `DROP INDEX` DDL event
`alter table`	DDL	The `ALTER TABLE` DDL event

sql-pattern: it is used to filter specified DDL SQL statements. The matching rule supports using a regular expression. For example, "^DROP\\s+PROCEDURE".
action: the string (Do/Ignore). Based on the following rules, it judges whether to filter. If either of the two rules is satisfied, the binlog is filtered; otherwise, the binlog is not filtered.
- Do: the allow list. The binlog is filtered in either of the following two conditions:
  - The type of the event is not in the event list of the rule.
  - The SQL statement of the event cannot be matched by sql-pattern of the rule.
- Ignore: the block list. The binlog is filtered in either of the following two conditions:
  - The type of the event is in the event list of the rule.
  - The SQL statement of the event can be matched by sql-pattern of the rule.

Usage examples

This section shows the usage examples in the scenario of sharding (sharded schemas and tables).

Filter all sharding deletion operations

To filter out all deletion operations, configure the following two filtering rules:

filter-table-rule filters out the truncate table, drop table and delete statement operations of all tables that match the test_*.t_* pattern.
filter-schema-rule filters out the drop database operation of all schemas that match the test_* pattern.

filters:
  filter-table-rule:
    schema-pattern: "test_*"
    table-pattern: "t_*"
    events: ["truncate table", "drop table", "delete"]
    action: Ignore
  filter-schema-rule:
    schema-pattern: "test_*"
    events: ["drop database"]
    action: Ignore

Only migrate sharding DML statements

To only migrate sharding DML statements, configure the following two filtering rules:

do-table-rule only migrates the create table, insert, update and delete statements of all tables that match the test_*.t_* pattern.
do-schema-rule only migrates the create database statement of all schemas that match the test_* pattern.

Note

The reason why the create database/table statement is migrated is that you can migrate DML statements only after the schema and table are created.

filters:
  do-table-rule:
    schema-pattern: "test_*"
    table-pattern: "t_*"
    events: ["create table", "all dml"]
    action: Do
  do-schema-rule:
    schema-pattern: "test_*"
    events: ["create database"]
    action: Do

Filter out the SQL statements that TiDB does not support

To filter out the PROCEDURE statements that TiDB does not support, configure the following filter-procedure-rule:

filters:
  filter-procedure-rule:
    schema-pattern: "test_*"
    table-pattern: "t_*"
    sql-pattern: ["^DROP\\s+PROCEDURE", "^CREATE\\s+PROCEDURE"]
    action: Ignore

filter-procedure-rule filters out the ^CREATE\\s+PROCEDURE and ^DROP\\s+PROCEDURE statements of all tables that match the test_*.t_* pattern.

Filter out the SQL statements that the TiDB parser does not support

For the SQL statements that the TiDB parser does not support, DM cannot parse them and get the schema/table information. So you must use the global filtering rule: schema-pattern: "*".

Note

To avoid filtering out data that need to be migrated, you must configure the global filtering rule as strictly as possible.

To filter out the PARTITION statements that the TiDB parser (of some version) does not support, configure the following filtering rule:

filters:
  filter-partition-rule:
    schema-pattern: "*"
    sql-pattern: ["ALTER\\s+TABLE[\\s\\S]*ADD\\s+PARTITION", "ALTER\\s+TABLE[\\s\\S]*DROP\\s+PARTITION"]
    action: Ignore

Online DDL tools

In the MySQL ecosystem, tools such as gh-ost and pt-osc are widely used. DM provides supports for these tools to avoid migrating unnecessary intermediate data.

Restrictions

DM only supports gh-ost and pt-osc.
When online-ddl is enabled, the checkpoint corresponding to incremental replication should not be in the process of online DDL execution. For example, if an upstream online DDL operation starts at position-A and ends at position-B of the binlog, the starting point of incremental replication should be earlier than position-A or later than position-B; otherwise, an error occurs. For details, refer to FAQ.

Parameter configuration

v2.0.5 and later
earlier than v2.0.5

In v2.0.5 and later versions, you need to use the online-ddl configuration item in the task configuration file.

If the upstream MySQL/MariaDB (at the same time) uses the gh-ost or pt-osc tool, set online-ddl to true in the task configuration file:

online-ddl: true

Note

Since v2.0.5, online-ddl-scheme has been deprecated, so you need to use online-ddl instead of online-ddl-scheme. That means that setting online-ddl: true overwrites online-ddl-scheme, and setting online-ddl-scheme: "pt" or online-ddl-scheme: "gh-ost" is converted to online-ddl: true.

Before v2.0.5 (not including v2.0.5), you need to use the online-ddl-scheme configuration item in the task configuration file.

If the upstream MySQL/MariaDB uses the gh-ost tool, set it in the task configuration file:

online-ddl-scheme: "gh-ost"

If the upstream MySQL/MariaDB uses the pt tool, set it in the task configuration file:

online-ddl-scheme: "pt"

Shard merge

DM supports merging the DML and DDL data in the upstream MySQL/MariaDB sharded tables and migrating the merged data to the downstream TiDB tables.

Restrictions

Currently, the shard merge feature is supported only in limited scenarios. For details, refer to Sharding DDL usage Restrictions in the pessimistic mode and Sharding DDL usage Restrictions in the optimistic mode.

Parameter configuration

Set shard-mode to pessimistic in the task configuration file:

shard-mode: "pessimistic" # The shard merge mode. Optional modes are ""/"pessimistic"/"optimistic". The "" mode is used by default which means sharding DDL merge is disabled. If the task is a shard merge task, set it to the "pessimistic" mode. After getting a deep understanding of the principles and restrictions of the "optimistic" mode, you can set it to the "optimistic" mode.

Handle sharding DDL locks manually

In some abnormal scenarios, you need to handle sharding DDL Locks manually.