Doc Menu

Key Features

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:             # This configuration applies to DM versions higher than v2.0.0-beta.2. Use black-white-list otherwise.
  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
  • ignore-tables: block lists of the tables to be migrated, similar to replicate-ignore-table in MySQL

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:

  1. 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.

  2. 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:  # This configuration applies to DM versions higher than v2.0.0-beta.2. Use black-white-list otherwise.
  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:

TableWhether to filterWhy filter
logs.messages_2016YesThe schema logs fails to match any do-dbs.
logs.messages_2017YesThe schema logs fails to match any do-dbs.
logs.messages_2018YesThe schema logs fails to match any do-dbs.
forum_backup_2016.messagesYesThe schema forum_backup_2016 fails to match any do-dbs.
forum_backup_2017.messagesYesThe schema forum_backup_2017 fails to match any do-dbs.
forum.usersYes1. 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.messagesNo1. 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.messagesNo1. 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 single table, the Ignore rule takes effect.

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:

    EventsTypeDescription
    allIncludes all the events below
    all dmlIncludes all DML events below
    all ddlIncludes all DDL events below
    noneIncludes none of the events below
    none ddlIncludes none of the DDL events below
    none dmlIncludes none of the DML events below
    insertDMLThe INSERT DML event
    updateDMLThe UPDATE DML event
    deleteDMLThe DELETE DML event
    create databaseDDLThe CREATE DATABASE DDL event
    drop databaseDDLThe DROP DATABASE DDL event
    create tableDDLThe CREATE TABLE DDL event
    create indexDDLThe CREATE INDEX DDL event
    drop tableDDLThe DROP TABLE DDL event
    truncate tableDDLThe TRUNCATE TABLE DDL event
    rename tableDDLThe RENAME TABLE DDL event
    drop indexDDLThe DROP INDEX DDL event
    alter tableDDLThe 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

Column mapping

Note:

The column mapping is not recommended as the primary solution due to its usage restrictions. The preferable solution is handling conflicts of auto-increment primary key.

The column mapping feature supports modifying the value of table columns. You can execute different modification operations on the specified column according to different expressions. Currently, only the built-in expressions provided by DM are supported.

Note:

  • It does not support modifying the column type and the table schema.
  • It does not support configuring multiple different column mapping rules for the same table.

Parameter configuration

column-mappings:
  rule-1:
​    schema-pattern: "test_*"
​    table-pattern: "t_*"
​    expression: "partition id"
​    source-column: "id"
​    target-column: "id"
​    arguments: ["1", "test", "t", "_"]
  rule-2:
​    schema-pattern: "test_*"
​    table-pattern: "t_*"
​    expression: "partition id"
​    source-column: "id"
​    target-column: "id"
​    arguments: ["2", "test", "t", "_"]

Parameter explanation

  • schema-pattern/table-pattern: to execute column value modifying operations on the upstream MySQL or MariaDB instance tables that match the schema-pattern/table-pattern filtering rule.
  • source-column, target-column: to modify the value of the source-column column according to specified expression and assign the new value to target-column.
  • expression: the expression used to modify data. Currently, only the partition id built-in expression is supported.

The partition id expression

partition id is used to resolve the conflicts of auto-increment primary keys of sharded tables.

partition id restrictions

Note the following restrictions:

  • The partition id expression only supports the bigint type of auto-increment primary key.
  • If the schema prefix is not empty, the schema name format must be schema prefix or schema prefix + separator + number (the schema ID). For example, it supports s and s_1, but does not support s_a.
  • If the table prefix is not empty, the table name format must be table prefix or table prefix + separator + number (the table ID).
  • If the schema/table name does not contain the … + separator + number part, the corresponding ID is considered as 0.
  • Restrictions on sharding size:
    • It supports 16 MySQL or MariaDB instances at most (Requirement: 0 <= instance ID <= 15).
    • Each instance supports 128 schemas at most (Requirement: 0 <= schema ID <= 127).
    • Each schema of each instance supports 256 tables at most (Requirement: 0 <= table ID <= 255).
    • The range of the mapped column should meet the requirement: 0 <= ID <= 17592186044415.
    • The {instance ID, schema ID, table ID} group must be unique.
  • Currently, the partition id expression is a customized feature. If you want to modify this feature, contact the corresponding developers.

partition id arguments configuration

Configure the following three or four arguments in order:

  • instance_id: the ID of the upstream sharded MySQL or MariaDB instance (0 <= instance ID <= 15)
  • schema prefix: used to parse the schema name and get the schema ID
  • table prefix: used to parse the table name and get the table ID
  • The separator: used to separate between the prefix and the IDs, and can be omitted to use an empty string as a separator

Any of instance_id, schema prefix and table prefix can be set to an empty string ("") to indicate that the corresponding parts are not encoded into the partition ID.

partition id expression rules

partition id fills the beginning bit of the auto-increment primary key ID with the argument number, and computes an int64 (MySQL bigint) type of value. The specific rules are as follows:

instance_idschema prefixtable prefixEncoding
☑ defined☑ defined☑ defined[S: 1 bit][`I`: 4 bits] [D: 7 bits][`T`: 8 bits] [P: 44 bits]
☐ empty☑ defined☑ defined[S: 1 bit][`D`: 7 bits] [T: 8 bits][`P`: 48 bits]
☑ defined☐ empty☑ defined[S: 1 bit][`I`: 4 bits] [T: 8 bits][`P`: 51 bits]
☑ defined☑ defined☐ empty[S: 1 bit][`I`: 4 bits] [D: 7 bits][`P`: 52 bits]
☐ empty☐ empty☑ defined[S: 1 bit][`T`: 8 bits] [P: 55 bits]
☐ empty☑ defined☐ empty[S: 1 bit][`D`: 7 bits] [P: 56 bits]
☑ defined☐ empty☐ empty[S: 1 bit][`I`: 4 bits] [P: 59 bits]
  • S: the sign bit, reserved
  • I: the instance ID, 4 bits by default if set
  • D: the schema ID, 7 bits by default if set
  • T: the table ID, 8 bits by default if set
  • P: the auto-increment primary key ID, occupying the rest of bits (≥44 bits)

Usage example

Assuming in the sharding scenario where all tables have the auto-increment primary key, you want to migrate two upstream MySQL instances test_{1,2,3...}.t_{1,2,3...} to the downstream TiDB instances test.t.

Configure the following two rules:

column-mappings:
  rule-1:
​    schema-pattern: "test_*"
​    table-pattern: "t_*"
​    expression: "partition id"
​    source-column: "id"
​    target-column: "id"
​    arguments: ["1", "test", "t", "_"]
  rule-2:
​    schema-pattern: "test_*"
​    table-pattern: "t_*"
​    expression: "partition id"
​    source-column: "id"
​    target-column: "id"
​    arguments: ["2", "test", "t", "_"]
  • The column ID of the MySQL instance 1 table test_1.t_1 is converted from 1 to 1 << (64-1-4) | 1 << (64-1-4 -7) | 1 << 44 | 1 = 580981944116838401.
  • The row ID of the MySQL instance 2 table test_1.t_2 is converted from 2 to 2 << (64-1-4) | 1 << (64-1-4 -7) | 2 << 44 | 2 = 1157460288606306306.

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-scheme 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

  • If the upstream MySQL/MariaDB uses gh-ost, set online-ddl-scheme to "gh-ost" in the task configuration file:
online-ddl-scheme: "gh-ost"
  • If the upstream MySQL/MariaDB uses pt-osc, set online-ddl-scheme to "pt" in the task configuration file:
online-ddl-scheme: "pt"

For more information about online DDL tools, refer to Online DDL Scheme.

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 is-sharding to true in the task configuration file:

shard-mode: "pessimistic" # You need to configure this if it is a shard merge task. The "pessimistic" mode is used by default. After understanding 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.