Update bdr mode && add description related bdr sql

TOC.md

@@ -710,6 +710,7 @@
       - [`ADMIN PAUSE DDL`](/sql-statements/sql-statement-admin-pause-ddl.md)
       - [`ADMIN RECOVER INDEX`](/sql-statements/sql-statement-admin-recover.md)
       - [`ADMIN RESUME DDL`](/sql-statements/sql-statement-admin-resume-ddl.md)
+      - [`ADMIN [SET|SHOW|UNSET] BDR ROLE`](/sql-statements/sql-statement-admin-bdr-role.md)
       - [`ADMIN SHOW DDL [JOBS|JOB QUERIES]`](/sql-statements/sql-statement-admin-show-ddl.md)
       - [`ADMIN SHOW TELEMETRY`](/sql-statements/sql-statement-admin-show-telemetry.md)
       - [`ALTER DATABASE`](/sql-statements/sql-statement-alter-database.md)

error-codes.md

@@ -526,6 +526,10 @@ TiDB is compatible with the error codes in MySQL, and in most cases returns the
 
     DDL is paused by `ADMIN PAUSE` and cannot be paused again.
 
+* Error Number: 8263
+
+    This DDL cannot be executed under a specific BDR role. Make sure that the cluster is in [bidirectional replication](/ticdc/ticdc-bidirectional-replication.md). If the cluster is not in bidirectional replication, you can use `ADMIN UNSET BDR ROLE;` to make DDL normal.
+
 * Error Number: 9001
 
     The complete error message: `ERROR 9001 (HY000): PD Server Timeout`

keywords.md

@@ -393,7 +393,6 @@ The following list shows the keywords in TiDB. Reserved keywords are marked with
 - LIST
 - LOAD (R)
 - LOCAL
-- LOCAL_ONLY
 - LOCALTIME (R)
 - LOCALTIMESTAMP (R)
 - LOCATION
@@ -732,6 +731,7 @@ The following list shows the keywords in TiDB. Reserved keywords are marked with
 - UNIQUE (R)
 - UNKNOWN
 - UNLOCK (R)
+- UNSET
 - UNSIGNED (R)
 - UNTIL (R)
 - UPDATE (R)

sql-statements/sql-statement-admin-bdr-role.md

@@ -0,0 +1,88 @@
+---
+title: ADMIN [SET|SHOW|UNSET] BDR ROLE
+summary: An overview of the usage of ADMIN [SET|SHOW|UNSET] BDR ROLE for the TiDB database.
+---
+
+# ADMIN [SET|SHOW|UNSET] BDR ROLE
+
+- Use `ADMIN SET BDR ROLE` to set the BDR role of the cluster. Currently, you can set the following BDR roles for a TiDB cluster: `PRIMARY` and `SECONDARY`. For more information about BDR roles, see [DDL Synchronization in TiCDC Bidirectional Replication](/ticdc/ticdc-bidirectional-replication.md#ddl-replication).
+- Use `ADMIN SHOW BDR ROLE` to show the BDR role of the cluster.
+- Use `ADMIN UNSET BDR ROLE` to unset the BDR role of the cluster.
+
+> **Warning:**
+>
+> This feature is experimental. It is not recommended that you use it in the production environment. This feature might be changed or removed without prior notice. If you find a bug, you can report an [issue](https://github.com/pingcap/tidb/issues) on GitHub.
+
+## Synopsis
+
+```ebnf+diagram
+AdminShowBDRRoleStmt ::=
+    'ADMIN' 'SHOW' 'BDR' 'ROLE'
+
+AdminSetBDRRoleStmt ::=
+    'ADMIN' 'SET' 'BDR' 'ROLE' ('PRIMARY' | 'SECONDARY')
+
+AdminUnsetBDRRoleStmt ::=
+    'ADMIN' 'UNSET' 'BDR' 'ROLE'
+```
+
+## Examples
+
+By default, a TiDB cluster has no BDR role. Run the folloiwng command to show the BDR role of the cluster.
+
+```sql
+ADMIN SHOW BDR ROLE;
+```
+
+```sql
++------------+
+| BDR_ROLE   |
++------------+
+|            |
++------------+
+1 row in set (0.01 sec)
+```
+
+Run the following command to set the BDR role to `PRIMARY`.
+
+```sql
+ADMIN SET BDR ROLE PRIMARY;
+```
+
+```sql
+Query OK, 0 rows affected (0.01 sec)
+```
+
+```sql
+ADMIN SHOW BDR ROLE;
++----------+
+| BDR_ROLE |
++----------+
+| primary  |
++----------+
+1 row in set (0.00 sec)
+```
+
+Run the following command to unset the BDR role of the cluster.
+
+```sql
+ADMIN UNSET BDR ROLE;
+```
+
+```sql
+Query OK, 0 rows affected (0.01 sec)
+```
+
+```sql
+ADMIN SHOW BDR ROLE;
++----------+
+| BDR_ROLE |
++----------+
+|          |
++----------+
+1 row in set (0.01 sec)
+```
+
+## MySQL compatibility
+
+This statement is a TiDB extension to MySQL syntax.

ticdc/ticdc-bidirectional-replication.md

@@ -5,7 +5,7 @@ summary: Learn how to use bidirectional replication of TiCDC.
 
 # Bidirectional Replication
 
-Starting from v6.5.0, TiCDC supports bi-directional replication among two TiDB clusters. Based on this feature, you can create a multi-active TiDB solution using TiCDC.
+TiCDC supports bi-directional replication (BDR) among two TiDB clusters. Based on this feature, you can create a multi-active TiDB solution using TiCDC.
 
 This section describes how to use bi-directional replication taking two TiDB clusters as an example.
 
@@ -34,47 +34,117 @@ TiCDC only replicates incremental data changes that occur after a specified time
 
 After the configuration takes effect, the clusters can perform bi-directional replication.
 
-## Execute DDL
-
-After the bidirectional replication is enabled, TiCDC does not replicate any DDL statements. You need to execute DDL statements in the upstream and downstream clusters respectively.
-
-Note that some DDL statements might cause table structure changes or data change time sequence problems, which might lead to data inconsistency after the replication. Therefore, after enabling bidirectional replication, only the DDL statements in the following table can be executed without stopping the write operations of the application.
-
-| Event | Does it cause changefeed errors | Note |
-|---|---|---|
-| create database | Yes | After you manually execute the DDL statements in the upstream and downstream clusters, the errors can be automatically recovered. |
-| drop database | Yes | You need to manually restart the changefeed and specify `--overwrite-checkpoint-ts` as the `commitTs` of the DDL statement to recover the errors. |
-| create table | Yes | After you manually execute the DDL statements in the upstream and downstream clusters, the errors can be automatically recovered. |
-| drop table | Yes | You need to manually restart the changefeed and specify `--overwrite-checkpoint-ts` as the `commitTs` of the DDL statement to recover the errors. |
-| alter table comment | No |  |
-| rename index | No |  |
-| alter table index visibility | No |  |
-| add partition | Yes | After you manually execute the DDL statements in the upstream and downstream clusters, the errors can be automatically recovered. |
-| drop partition | No |  |
-| create view | No |  |
-| drop view | No |  |
-| alter column default value | No |  |
-| reorganize partition | Yes | After you manually execute the DDL statements in the upstream and downstream clusters, the errors can be automatically recovered. |
-| alter table ttl | No |  |
-| alter table remove ttl | No |  |
-| add **not unique** index | No |  |
-| drop **not unique** index | No |  |
-
-If you need to execute DDL statements that are not in the preceding table, take the following steps:
-
-1. Pause the write operations in the tables that need to execute DDL in all clusters.
-2. After the write operations of the corresponding tables in all clusters have been replicated to other clusters, manually execute all DDL statements in each TiDB cluster.
-3. After the DDL statements are executed, resume the write operations.
+## DDL types
+
+Starting from v7.6.0, to support DDL replication as much as possible in bi-directional replication, TiDB divides the [DDLs that TiCDC originally supports](/ticdc/ticdc-ddl.md) into two types: replicable DDLs and non-replicable DDLs, according to the impact of DDLs on the business.
+
+### Replicable DDLs
+
+Replicable DDLs are the DDLs that can be directly executed and replicated to other TiDB clusters in bi-directional replication.
+
+Replicable DDLs include:
+
+- `CREATE DATABASE`
+- `CREATE TABLE`
+- `ADD COLUMN`: the column can be `null`, or has `not null` and `default value` at the same time
+- `ADD NON-UNIQUE INDEX`
+- `DROP INDEX`
+- `MODIFY COLUMN`: you can only modify the `default value` and `comment` of the column
+- `ALTER COLUMN DEFAULT VALUE`
+- `MODIFY TABLE COMMENT`
+- `RENAME INDEX`
+- `ADD TABLE PARTITION`
+- `DROP PRIMARY KEY`
+- `ALTER TABLE INDEX VISIBILITY`
+- `ALTER TABLE TTL`
+- `ALTER TABLE REMOVE TTL`
+- `CREATE VIEW`
+- `DROP VIEW`
+
+### Non-replicable DDLs
+
+Non-replicable DDLs are the DDLs that have a great impact on the business, and might cause data inconsistency between clusters. Non-replicable DDLs cannot be directly replicated to other TiDB clusters in bi-directional replication through TiCDC. Non-replicable DDLs must be executed through specific operations.
+
+Non-replicable DDLs include:
+
+- `DROP DATABASE`
+- `DROP TABLE`
+- `ADD COLUMN`: the column is `not null` and does not have a `default value`
+- `DROP COLUMN`
+- `ADD UNIQUE INDEX`
+- `TRUNCATE TABLE`
+- `MODIFY COLUMN`: you can modify the attributes of the column except `default value` and `comment`
+- `RENAME TABLE`
+- `DROP PARTITION`
+- `TRUNCATE PARTITION`
+- `ALTER TABLE CHARACTER SET`
+- `ALTER DATABASE CHARACTER SET`
+- `RECOVER TABLE`
+- `ADD PRIMARY KEY`
+- `REBASE AUTO ID`
+- `EXCHANGE PARTITION`
+- `REORGANIZE PARTITION`
+
+## DDL replication
+
+To solve the problem of replicable DDLs and non-replicable DDLs, TiDB introduces the following BDR roles:
+
+- `PRIMARY`: you can execute replicable DDLs, but cannot execute non-replicable DDLs. Replicable DDLs will be replicated to the downstream by TiCDC.
+- `SECONDARY`: you cannot execute replicable DDLs or non-replicable DDLs, but can execute the DDLs replicated by TiCDC.
+
+When no BDR role is set, you can execute any DDL. But after you set `bdr_mode=true` on TiCDC, the executed DDL will not be replicated by TiCDC.
+
+> **Warning:**
+>
+> This feature is experimental. It is not recommended that you use it in the production environment. This feature might be changed or removed without prior notice. If you find a bug, you can report an [issue](https://github.com/pingcap/tidb/issues) on GitHub.
+
+### Replication scenarios of replicable DDLs
+
+1. Choose a TiDB cluster and execute `ADMIN SET BDR ROLE PRIMARY` to set it as the primary cluster.
+2. On other TiDB clusters, execute `ADMIN SET BDR ROLE SECONDARY` to set them as the secondary clusters.
+3. Execute **replicable DDLs** on the primary cluster. The successfully executed DDLs will be replicated to the secondary clusters by TiCDC.
+
+> **Note:**
+>
+> To prevent misoperations:
+>
+> - If you try to execute **non-replicable DDLs** on the primary cluster, you will get the [Error 8263](/error-codes.md).
+> - If you try to execute **replicable DDLs** or **non-replicable DDLs** on the secondary clusters, you will get the [Error 8263](/error-codes.md).
+
+### Replication scenarios of non-replicable DDLs
+
+1. Execute `ADMIN UNSET BDR ROLE` on all TiDB clusters to unset the BDR role.
+2. Stop writing data to the tables that need to execute DDLs in all clusters.
+3. Wait until all writes to the corresponding tables in all clusters are replicated to other clusters, and then manually execute all DDLs on each TiDB cluster.
+4. Wait until the DDLs are completed, and then resume writing data.
+5. Follow the steps in [Replication scenarios of replicable DDLs](#replication-scenarios-of-replicable-ddls) to switch back to the replication scenario of replicable DDLs.
+
+> **Warning:**
+>
+> After you execute `ADMIN UNSET BDR ROLE` on all TiDB clusters, none of the DDLs are replicated by TiCDC. You need to manually execute the DDLs on each cluster separately.
 
 ## Stop bi-directional replication
 
 After the application has stopped writing data, you can insert a special record into each cluster. By checking the two special records, you can make sure that data in two clusters are consistent.
 
-After the check is completed, you can stop the changefeed to stop bi-directional replication.
+After the check is completed, you can stop the changefeed to stop bi-directional replication, and execute `ADMIN UNSET BDR ROLE` on all TiDB clusters.
 
 ## Limitations
 
-- For the limitations of DDL, see [Execute DDL](#execute-ddl).
+- Use BDR role only in the following scenarios:
+
+    - 1 `PRIMARY` cluster and n `SECONDARY` clusters (replication scenarios of replicable DDLs)
+    - n clusters that have no BDR roles (replication scenarios in which you can manually execute non-replicable DDLs on each cluster)
+
+    **Do not set the BDR role in other scenarios, for example, setting `PRIMARY`, `SECONDARY`, and no BDR roles at the same time. If you set the BDR role incorrectly, TiDB cannot guarantee data correctness and consistency during data replication.**
+
+- Usually do not use `AUTO_INCREMENT` or `AUTO_RANDOM` to avoid data conflicts in the replicated tables. If you need to use `AUTO_INCREMENT` or `AUTO_RANDOM`, you can set different `auto_increment_increment` and `auto_increment_offset` for different clusters to ensure that different clusters can be assigned different primary keys. For example, if there are three TiDB clusters (A, B, and C) in bi-directional replication, you can set them as follows:
+
+    - In Cluster A, set `auto_increment_increment=3` and `auto_increment_offset=2000`
+    - In Cluster B, set `auto_increment_increment=3` and `auto_increment_offset=2001`
+    - In Cluster C, set `auto_increment_increment=3` and `auto_increment_offset=2002`
+
+    This way, A, B, and C will not conflict with each other in the implicitly assigned `AUTO_INCREMENT` ID and `AUTO_RANDOM` ID. If you need to add a cluster in BDR mode, you need to temporarily stop writing data of the related application, set the appropriate values for `auto_increment_increment` and `auto_increment_offset` on all clusters, and then resume writing data of the related application.
 
 - Bi-directional replication clusters cannot detect write conflicts, which might cause undefined behaviors. Therefore, you must ensure that there are no write conflicts from the application side.