CDL Usage Instructions

Data synchronization task

• The CDL supports the following types of data synchronization tasks:

  Table 1 Data synchronization task types supported by the CDL

  Data source	Destination end	Description
  MySQL	Hudi	This task synchronizes data from the MySQL database to Hudi.
  	Kafka	This task synchronizes data from the MySQL database to Kafka.
  PgSQL	Hudi	This task synchronizes data from the PgSQL database to Hudi.
  	Kafka	This task synchronizes data from the PgSQL database to Kafka.
  Hudi	DWS	This task synchronizes data from the Hudi database to DWS.
  	ClickHouse	This task synchronizes data from the Hudi database to ClickHouse.
  ThirdKafka	Hudi	This task synchronizes data from the ThirdKafka database to Hudi.
  	Kafka	This task synchronizes data from the ThirdKafka database to Kafka.
  openGuass (supported by MRS 3.3.0 and later versions)	ThirdKafka (DMS/DRS) -> Hudi	Synchronizes data from GaussDB(for openGauss) to Hudi through ThirdKafka (DMS/DRS).
  	Hudi	Synchronizes data from openGauss to Hudi.
  	Kafka	Synchronizes data from openGauss to Kafka.
  ogg-oracle-avro (supported by MRS 3.3.0 and later versions)	ThirdKafka (DMS/DRS) -> Hudi	Synchronizes data from avro-oracle to Hudi through ThirdKafka (DMS/DRS).

• Database versions supported by CDL in a data synchronization task:
  Kafka (including ThirdKafka that uses MRS Kafka as the source), Hudi, and ClickHouse use related components in the MRS cluster as data sources. For details about the version numbers, see List of MRS Component Versions. Table 2 lists the database versions.

  Table 2 Database versions

  Database Name	Data Source	Destination End	Version
  MySQL	Support	Not supported	5.7 and 8.0.x
  PostgreSQL	Support	Not supported	9.6, 10, 11, 12, and 13
  Opengauss (Open source version)	Support	Not supported	2.1.0 or later
  DWS	Not supported	Support	8.1.1 or later

• Usage Constraints:
  • If CDL is required, the value of log.cleanup.policy of Kafka must be delete.
  • The CDL service has been installed in the MRS cluster.
  • CDL can capture incremental data only from non-system tables, but not from built-in databases of databases such as MySQL, and PostgreSQL.
  • When data is synchronized from Hudi to DWS or ClickHouse, the data that is deleted from Hudi will not be deleted from the destination. For example, if you run the delete from tableName command in Hudi to physically delete table data, the table data still exists on the destination DWS or ClickHouse.
  • Binary logging (enabled by default) and GTID have been enabled for the MySQL database. CDL cannot fetch tables whose names contain special characters such as the dollar sign ($) or Chinese characters.

    To check whether binary logging is enabled for the MySQL database:

    Use a tool (Navicat is used in this example) or CLI to connect to the MySQL database and run the show variables like 'log_%' command to view the configuration.

    For example, in Navicat, choose File > New Query to create a query, enter the following SQL statement, and click Run. If log_bin is displayed as ON in the result, the function is enabled successfully.

    show variables like 'log_%'

    

    If the bin log function of the MySQL database is not enabled, perform the following operations:

    Modify the MySQL configuration file my.cnf (my.ini for Windows) as follows:

    server-id         = 223344
    log_bin           = mysql-bin
    binlog_format     = ROW
    binlog_row_image  = FULL
    expire_logs_days  = 10

    After the modification, restart MySQL for the configurations to take effect.

    To check whether GTID is enabled for the MySQL database:

    Run the show global variables like '%gtid%' command to check whether GTID is enabled. For details, see the official documentation of the corresponding MySQL version. (For details about how to enable the function in MySQL 8.x, see https://dev.mysql.com/doc/refman/8.0/en/replication-mode-change-online-enable-gtids.html.)

    

    Set user permissions:

    To execute MySQL tasks, users must have the SELECT, RELOAD, SHOW DATABASES, REPLICATION SLAVE and REPLICATION CLIENT permissions.

    Run the following command to grant the permissions, There can be security risks if a command contains the authentication password. You are advised to disable the command recording function (history) before running the command.

    GRANT SELECT, RELOAD, SHOW DATABASES, REPLICATION SLAVE, REPLICATION CLIENT ON *.* TO 'Username' IDENTIFIED BY 'Password';

    Run the following command to update the permissions:

    FLUSH PRIVILEGES;

  • The write-ahead log policy is modified for the PostgreSQL database.
    

    • The user for connecting to the PostgreSQL database must have the replication permission, the CREATE permission on the database, and is the owner of tables.
    • CDL cannot fetch tables whose names contain special characters such as the dollar sign ($) or Chinese characters.
    • For PostgreSQL databases, you must have the permission to set the statement_timeout and lock_timeout parameters and the permission to query and delete slots and publications.
    • You are advised to set max_wal_senders to 1.5 or 2 times the value of Slot.
    • If the replication identifier of a PostgreSQL table is default, enable the full field completion function in the following scenarios:
      • Scenario 1:

        When the delete operation is performed on the source database, a delete event contains only the primary key information. In this case, for the delete data written to Hudi, only the primary key has values, and the values of other service fields are null.

      • Scenario 2:

        When the size of a single piece of data in the database exceeds 8 KB (including 8 KB), an update event contains only changed fields. In this case, the values of some fields in the Hudi data are __debezium_unavailable_value.

      The related commands are as follows:

      • Command for querying the replication identifier of a PostgreSQL table:

        SELECT CASE relreplident WHEN 'd' THEN 'default' WHEN 'n' THEN 'nothing' WHEN 'f' THEN 'full' WHEN 'i' THEN 'index' END AS replica_identity FROM pg_class WHERE oid = 'tablename'::regclass;

      • Command for enabling the full field completion function for a table:

        ALTER TABLE tablename REPLICA IDENTITY FULL;

    1. Modify wal_level = logical in the database configuration file postgresql.conf (which is stored in the data folder in the PostgreSQL installation directory by default).

       #------------------------------------------------
       #WRITE-AHEAD LOG
       #------------------------------------------------
       
       # - Settings -
       wal_level = logical         # minimal, replica, or logical
                               # (change requires restart)
       #fsync = on             #flush data to disk for crash safety
       ...

    2. Restart the database service.

       # Stop
       pg_ctl stop
       # Start
       pg_ctl start

  • Prerequisites for the DWS database

    Before a synchronization task is started, both the source and target tables exist and have the same table structure. The value of ads_last_update_date in the DWS table is the current system time.

  • Prerequisites for ThirdPartyKafka

    The upper-layer source supports openGauss and OGG. Kafka topics at the source end can be consumed by Kafka in the MRS cluster.

    

    ThirdKafka does not support data synchronization from the distributed openGauss database to CDL.

  • Prerequisites for ClickHouse

    You have the permissions to operate ClickHouse. For details, see ClickHouse User and Permission Management.

  • Write-ahead logging required for the openGauss database (supported in MRS 3.3.0 and later versions)
    

    • The user connecting to openGauss databases must have logical replication permission.
    • openGauss interval partitioned tables and Chinese table names cannot be synchronized.
    • If a table in the source openGauss database does not have a primary key, data in the table cannot be deleted.
    • If the data in the source openGauss database needs to be deleted, run the following commands to enable the field completion function:
      • Command for querying the replication identifier of a openGauss table:

        SELECT CASE relreplident WHEN 'd' THEN 'default' WHEN 'n' THEN 'nothing' WHEN 'f' THEN 'full' WHEN 'i' THEN 'index' END AS replica_identity FROM pg_class WHERE oid = 'tablename'::regclass;

      • Command for enabling the full field completion function for a table:

        ALTER TABLE tablename REPLICA IDENTITY FULL;

    1. Modify wal_level = logical in the database configuration file postgresql.conf (which is stored in the data folder in the openGauss installation directory by default).

       #------------------------------------------------
       #WRITE-AHEAD LOG
       #------------------------------------------------
       
       # - Settings -
       wal_level = logical         # minimal, replica, or logical
                               # (change requires restart)
       #fsync = on             #flush data to disk for crash safety
       ...

    2. Restart the database service.

       # Stop
       pg_ctl stop
       # Start
       pg_ctl start

Data Types and Mapping Supported by CDL Synchronization Tasks

This section describes the data types supported by CDL synchronization tasks and the mapping between data types of the source database and Spark data types.

Data comparison task

Data comparison checks the consistency between data in the source database and that in the target Hive. If the data is inconsistent, CDL can attempt to repair the inconsistent data. For detail, see Creating a CDL Data Comparison Job.