Manual
Manual
Environment
JAVA,JDK 1.8+.
The migration scene we support:
Source | Target |
---|---|
MySQL(5.1.15 ~ 5.7.x) | MySQL(5.1.15 ~ 5.7.x) |
PostgreSQL(9.4 ~ ) | PostgreSQL(9.4 ~ ) |
openGauss(2.1.0) | openGauss(2.1.0) |
Attention:
If the backend database is in following table, please download JDBC driver jar and put it into ${shardingsphere-proxy}/lib
directory.
RDBMS | JDBC driver | Reference |
---|---|---|
MySQL | mysql-connector-java-5.1.47.jar | Connector/J Versions |
openGauss | opengauss-jdbc-2.0.1-compatibility.jar |
Supported features:
Feature | MySQL | PostgreSQL | openGauss |
---|---|---|---|
Inventory migration | Supported | Supported | Supported |
Incremental migration | Supported | Supported | Supported |
Create table automatically | Supported | Unsupported | Supported |
DATA_MATCH data consistency check | Supported | Supported | Supported |
CRC32_MATCH data consistency check | Supported | Unsupported | Unsupported |
Attention:
For RDBMS which Create table automatically
feature is not supported, we need to create sharding tables manually.
Privileges
MySQL
- Enable
binlog
Configuration Example of MySQL 5.7 my.cnf
:
[mysqld]
server-id=1
log-bin=mysql-bin
binlog-format=row
binlog-row-image=full
max_connections=600
Execute the following SQL to confirm whether binlog is turned on or not:
show variables like '%log_bin%';
show variables like '%binlog%';
As shown below, it means binlog has been turned on:
+-----------------------------------------+---------------------------------------+
| Variable_name | Value |
+-----------------------------------------+---------------------------------------+
| log_bin | ON |
| binlog_format | ROW |
| binlog_row_image | FULL |
+-----------------------------------------+---------------------------------------+
- Privileges of account that scaling use should include Replication privileges.
Execute the following SQL to confirm whether the user has migration permission or not:
SHOW GRANTS 'user';
Result Example:
+------------------------------------------------------------------------------+
|Grants for ${username}@${host} |
+------------------------------------------------------------------------------+
|GRANT REPLICATION SLAVE, REPLICATION CLIENT ON *.* TO ${username}@${host} |
|....... |
+------------------------------------------------------------------------------+
PostgreSQL
Enable test_decoding feature.
Adjust WAL configuration
Configuration Example of postgresql.conf
:
wal_level = logical
max_replication_slots = 10
Please refer to Write Ahead Log and Replication for more details.
DistSQL API for auto mode
Preview current sharding rule
Example:
preview select count(1) from t_order;
Response:
mysql> preview select count(1) from t_order;
+------------------+--------------------------------+
| data_source_name | sql |
+------------------+--------------------------------+
| ds_0 | select count(1) from t_order_0 |
| ds_0 | select count(1) from t_order_1 |
| ds_1 | select count(1) from t_order_0 |
| ds_1 | select count(1) from t_order_1 |
+------------------+--------------------------------+
4 rows in set (0.00 sec)
Start scaling job
- Add new data source resources
Please refer to RDL#Data Source for more details.
Create database on underlying RDBMS first, it will be used in following DistSQL
.
Example:
ADD RESOURCE ds_2 (
URL="jdbc:mysql://127.0.0.1:3306/scaling_ds_2?serverTimezone=UTC&useSSL=false",
USER=root,
PASSWORD=root,
PROPERTIES("maximumPoolSize"=10,"idleTimeout"="30000")
), ds_3 (
URL="jdbc:mysql://127.0.0.1:3306/scaling_ds_3?serverTimezone=UTC&useSSL=false",
USER=root,
PASSWORD=root,
PROPERTIES("maximumPoolSize"=10,"idleTimeout"="30000")
), ds_4 (
URL="jdbc:mysql://127.0.0.1:3306/scaling_ds_4?serverTimezone=UTC&useSSL=false",
USER=root,
PASSWORD=root,
PROPERTIES("maximumPoolSize"=10,"idleTimeout"="30000")
);
- Alter all sharding table rule
Currently, scaling job could only be emitted by executing ALTER SHARDING TABLE RULE
DistSQL.
Please refer to RDL#Sharding for more details.
SHARDING TABLE RULE
support two types: TableRule
and AutoTableRule
. Following is a comparison of the two sharding rule types:
Type | AutoTableRule | TableRule |
---|---|---|
Definition | Auto Sharding Algorithm | User-Defined Sharding Algorithm |
Meaning of fields in DistSQL is the same as YAML configuration, please refer to YAML Configuration#Sharding for more details.
Example of alter AutoTableRule
:
ALTER SHARDING TABLE RULE t_order (
RESOURCES(ds_2, ds_3, ds_4),
SHARDING_COLUMN=order_id,
TYPE(NAME=hash_mod,PROPERTIES("sharding-count"=6)),
GENERATED_KEY(COLUMN=order_id,TYPE(NAME=snowflake,PROPERTIES("worker-id"=123)))
);
RESOURCES
is altered from (ds_0, ds_1)
to (ds_2, ds_3, ds_4)
, and sharding-count
is altered from 4
to 6
, it will emit scaling job.
Uncompleted example of alter TableRule
:
ALTER SHARDING ALGORITHM database_inline (
TYPE(NAME=INLINE,PROPERTIES("algorithm-expression"="ds_${user_id % 3 + 2}"))
);
ALTER SHARDING TABLE RULE t_order (
DATANODES("ds_${2..4}.t_order_${0..1}"),
DATABASE_STRATEGY(TYPE=standard,SHARDING_COLUMN=user_id,SHARDING_ALGORITHM=database_inline),
TABLE_STRATEGY(TYPE=standard,SHARDING_COLUMN=order_id,SHARDING_ALGORITHM=t_order_inline),
GENERATED_KEY(COLUMN=order_id,TYPE(NAME=snowflake,PROPERTIES("worker-id"=123)))
), t_order_item (
DATANODES("ds_${2..4}.t_order_item_${0..1}"),
DATABASE_STRATEGY(TYPE=standard,SHARDING_COLUMN=user_id,SHARDING_ALGORITHM=database_inline),
TABLE_STRATEGY(TYPE=standard,SHARDING_COLUMN=order_id,SHARDING_ALGORITHM=t_order_item_inline),
GENERATED_KEY(COLUMN=order_item_id,TYPE(NAME=snowflake,PROPERTIES("worker-id"=123)))
);
algorithm-expression
of database_inline
is alerted from ds_${user_id % 2}
to ds_${user_id % 3 + 2}
, and DATANODES
of t_order
is alerted from ds_${0..1}.t_order_${0..1}
to ds_${2..4}.t_order_${0..1}
, it will emit scaling job.
Currently, ALTER SHARDING ALGORITHM
will take effect immediately, but table rule will not, it might cause inserting data into source side failure, so alter sharding table rule to AutoTableRule
is recommended for now.
List scaling jobs
Please refer to RAL#Scaling for more details.
Example:
show scaling list;
Response:
mysql> show scaling list;
+--------------------+-----------------------+----------------------+--------+---------------------+---------------------+
| id | tables | sharding_total_count | active | create_time | stop_time |
+--------------------+-----------------------+----------------------+--------+---------------------+---------------------+
| 659853312085983232 | t_order_item, t_order | 2 | false | 2021-10-26 20:21:31 | 2021-10-26 20:24:01 |
| 660152090995195904 | t_order_item, t_order | 2 | false | 2021-10-27 16:08:43 | 2021-10-27 16:11:00 |
+--------------------+-----------------------+----------------------+--------+---------------------+---------------------+
2 rows in set (0.04 sec)
Get scaling progress
Example:
show scaling status {jobId};
Response:
mysql> show scaling status 660152090995195904;
+------+-------------+----------+-------------------------------+--------------------------+
| item | data_source | status | inventory_finished_percentage | incremental_idle_minutes |
+------+-------------+----------+-------------------------------+--------------------------+
| 0 | ds_1 | FINISHED | 100 | 2834 |
| 1 | ds_0 | FINISHED | 100 | 2834 |
+------+-------------+----------+-------------------------------+--------------------------+
2 rows in set (0.00 sec)
Current scaling job is finished, new sharding rule should take effect, and not if scaling job is failed.
status
values:
Value | Description |
---|---|
PREPARING | preparing |
RUNNING | running |
EXECUTE_INVENTORY_TASK | inventory task running |
EXECUTE_INCREMENTAL_TASK | incremental task running |
ALMOST_FINISHED | almost finished |
FINISHED | finished (The whole process is completed, and the new rules have been taken effect) |
PREPARING_FAILURE | preparation failed |
EXECUTE_INVENTORY_TASK_FAILURE | inventory task failed |
EXECUTE_INCREMENTAL_TASK_FAILURE | incremental task failed |
If status
fails, you can check the log of proxy
to view the error stack and analyze the problem.
Preview new sharding rule
Example:
preview select count(1) from t_order;
Response:
mysql> preview select count(1) from t_order;
+------------------+--------------------------------+
| data_source_name | sql |
+------------------+--------------------------------+
| ds_2 | select count(1) from t_order_0 |
| ds_2 | select count(1) from t_order_1 |
| ds_3 | select count(1) from t_order_0 |
| ds_3 | select count(1) from t_order_1 |
| ds_4 | select count(1) from t_order_0 |
| ds_4 | select count(1) from t_order_1 |
+------------------+--------------------------------+
6 rows in set (0.01 sec)
Other DistSQL
Please refer to RAL#Scaling for more details.
DistSQL API for manual mode
Data consistency check and switch configuration could be emitted manually. Please refer to RAL#Scaling for more details.
Attention: It’s still under development.