Working with MySQL read replicas

Following, you can find specific information about working with read replicas on RDS for MySQL. For general information about read replicas and instructions for using them, see Working with DB instance read replicas.

Configuring read replicas with MySQL

Before a MySQL DB instance can serve as a replication source, make sure to enable automatic backups on the source DB instance. To do this, set the backup retention period to a value other than 0. This requirement also applies to a read replica that is the source DB instance for another read replica. Automatic backups are supported for read replicas running any version of MySQL. You can configure replication based on binary log coordinates for a MySQL DB instance.

On RDS for MySQL version 5.7.44 and higher MySQL 5.7 versions and RDS for MySQL 8.0.28 and higher 8.0 versions, you can configure replication using global transaction identifiers (GTIDs). For more information, see Using GTID-based replication.

You can create up to 15 read replicas from one DB instance within the same Region. For replication to operate effectively, each read replica should have the same amount of compute and storage resources as the source DB instance. If you scale the source DB instance, also scale the read replicas.

RDS for MySQL supports cascading read replicas. To learn how to configure cascading read replicas, see Using cascading read replicas with RDS for MySQL.

You can run multiple read replica create and delete actions at the same time that reference the same source DB instance. When you perform these actions, stay within the limit of 15 read replicas for each source instance.

A read replica of a MySQL DB instance can't use a lower DB engine version than its source DB instance.

Preparing MySQL DB instances that use MyISAM

If your MySQL DB instance uses a nontransactional engine such as MyISAM, you need to perform the following steps to successfully set up your read replica. These steps are required to make sure that the read replica has a consistent copy of your data. These steps are not required if all of your tables use a transactional engine such as InnoDB.

Configuring replication filters with MySQL

You can use replication filters to specify which databases and tables are replicated with a read replica. Replication filters can include databases and tables in replication or exclude them from replication.

The following are some use cases for replication filters:

Setting replication filtering parameters for RDS for MySQL

To configure replication filters, set the following replication filtering parameters on the read replica:

The parameters are evaluated in the order that they are listed. For more information about how these parameters work, see the MySQL documentation:

By default, each of these parameters has an empty value. On each read replica, you can use these parameters to set, change, and delete replication filters. When you set one of these parameters, separate each filter from others with a comma.

You can use the % and _ wildcard characters in the replicate-wild-do-table and replicate-wild-ignore-table parameters. The % wildcard matches any number of characters, and the _ wildcard matches only one character.

The binary logging format of the source DB instance is important for replication because it determines the record of data changes. The setting of the binlog_format parameter determines whether the replication is row-based or statement-based. For more information, see Configuring MySQL binary logging.

Replication filtering limitations for RDS for MySQL

The following limitations apply to replication filtering for RDS for MySQL:

Replication filtering examples for RDS for MySQL

To configure replication filtering for a read replica, modify the replication filtering parameters in the parameter group associated with the read replica.

You can set parameters in a parameter group using the Amazon Web Services Management Console, Amazon CLI, or RDS API. For information about setting parameters, see Modifying parameters in a DB parameter group. When you set parameters in a parameter group, all of the DB instances associated with the parameter group use the parameter settings. If you set the replication filtering parameters in a parameter group, make sure that the parameter group is associated only with read replicas. Leave the replication filtering parameters empty for source DB instances.

The following examples set the parameters using the Amazon CLI. These examples set ApplyMethod to immediate so that the parameter changes occur immediately after the CLI command completes. If you want a pending change to be applied after the read replica is rebooted, set ApplyMethod to pending-reboot.

The following examples set replication filters:

aws rds modify-db-parameter-group \
  --db-parameter-group-name myparametergroup \
  --parameters "ParameterName=replicate-do-db,ParameterValue='mydb1,mydb2',ApplyMethod=immediate"

aws rds modify-db-parameter-group ^
  --db-parameter-group-name myparametergroup ^
  --parameters "ParameterName=replicate-do-db,ParameterValue='mydb1,mydb2',ApplyMethod=immediate"

aws rds modify-db-parameter-group \
  --db-parameter-group-name myparametergroup \
  --parameters "ParameterName=replicate-do-table,ParameterValue='mydb1.table1,mydb1.table2',ApplyMethod=immediate"

aws rds modify-db-parameter-group ^
  --db-parameter-group-name myparametergroup ^
  --parameters "ParameterName=replicate-do-table,ParameterValue='mydb1.table1,mydb1.table2',ApplyMethod=immediate"

aws rds modify-db-parameter-group \
  --db-parameter-group-name myparametergroup \
  --parameters "ParameterName=replicate-wild-do-table,ParameterValue='mydb.order%,mydb.return%',ApplyMethod=immediate"

aws rds modify-db-parameter-group ^
  --db-parameter-group-name myparametergroup ^
  --parameters "ParameterName=replicate-wild-do-table,ParameterValue='mydb.order%,mydb.return%',ApplyMethod=immediate"

aws rds modify-db-parameter-group \
  --db-parameter-group-name myparametergroup \
  --parameters "ParameterName=replicate-ignore-db,ParameterValue='mydb5,mydb6',ApplyMethod=immediate"

aws rds modify-db-parameter-group ^
  --db-parameter-group-name myparametergroup ^
  --parameters "ParameterName=replicate-ignore-db,ParameterValue='mydb5,mydb6',ApplyMethod=immediate"

aws rds modify-db-parameter-group \
  --db-parameter-group-name myparametergroup \
  --parameters "ParameterName=replicate-ignore-table,ParameterValue='mydb5.table1,mydb6.table2',ApplyMethod=immediate"

aws rds modify-db-parameter-group ^
  --db-parameter-group-name myparametergroup ^
  --parameters "ParameterName=replicate-ignore-table,ParameterValue='mydb5.table1,mydb6.table2',ApplyMethod=immediate"

aws rds modify-db-parameter-group \
  --db-parameter-group-name myparametergroup \
  --parameters "ParameterName=replicate-wild-ignore-table,ParameterValue='mydb7.order%,mydb7.return%',ApplyMethod=immediate"

aws rds modify-db-parameter-group ^
  --db-parameter-group-name myparametergroup ^
  --parameters "ParameterName=replicate-wild-ignore-table,ParameterValue='mydb7.order%,mydb7.return%',ApplyMethod=immediate"

Viewing the replication filters for a read replica

You can view the replication filters for a read replica in the following ways:

Configuring delayed replication with MySQL

You can use delayed replication as a strategy for disaster recovery. With delayed replication, you specify the minimum amount of time, in seconds, to delay replication from the source to the read replica. In the event of a disaster, such as a table deleted unintentionally, you complete the following steps to recover from the disaster quickly:

Configuring delayed replication during read replica creation

To configure delayed replication for any future read replica created from a DB instance, run the mysql.rds_set_configuration stored procedure with the target delay parameter.

Modifying delayed replication for an existing read replica

To modify delayed replication for an existing read replica, run the mysql.rds_set_source_delay stored procedure.

Setting a location to stop replication to a read replica

After stopping replication to the read replica, you can start replication and then stop it at a specified binary log file location using the mysql.rds_start_replication_until stored procedure.

Replication stops automatically when the stop point is reached. The following RDS event is generated: Replication has been stopped since the replica reached the stop point specified by the rds_start_replication_until stored procedure.

Promoting a read replica

After replication is stopped, in a disaster recovery scenario, you can promote a read replica to be the new source DB instance. For information about promoting a read replica, see Promoting a read replica to be a standalone DB instance.

Updating read replicas with MySQL

Read replicas are designed to support read queries, but you might need occasional updates. For example, you might need to add an index to optimize the specific types of queries accessing the replica.

Although you can enable updates by setting the read_only parameter to 0 in the DB parameter group for the read replica, we recommend that you don't do so because it can cause problems if the read replica becomes incompatible with the source DB instance. For maintenance operations, we recommend that you use blue/green deployments. For more information, see Using Blue/Green Deployments for database updates.

If you disable read-only on a read replica, change the value of the read_only parameter back to 1 as soon as possible.

Working with Multi-AZ read replica deployments with MySQL

You can create a read replica from either single-AZ or Multi-AZ DB instance deployments. You use Multi-AZ deployments to improve the durability and availability of critical data, but you can't use the Multi-AZ secondary to serve read-only queries. Instead, you can create read replicas from high-traffic Multi-AZ DB instances to offload read-only queries. If the source instance of a Multi-AZ deployment fails over to the secondary, any associated read replicas automatically switch to use the secondary (now primary) as their replication source. For more information, see Configuring and managing a Multi-AZ deployment.

You can create a read replica as a Multi-AZ DB instance. Amazon RDS creates a standby of your replica in another Availability Zone for failover support for the replica. Creating your read replica as a Multi-AZ DB instance is independent of whether the source database is a Multi-AZ DB instance.

Using cascading read replicas with RDS for MySQL

RDS for MySQL supports cascading read replicas. With cascading read replicas, you can scale reads without adding overhead to your source RDS for MySQL DB instance.

With cascading read replicas, your RDS for MySQL DB instance sends data to the first read replica in the chain. That read replica then sends data to the second replica in the chain, and so on. The end result is that all read replicas in the chain have the changes from the RDS for MySQL DB instance, but without the overhead solely on the source DB instance.

You can create a series of up to three read replicas in a chain from a source RDS for MySQL DB instance. For example, suppose that you have an RDS for MySQL DB instance, mysql-main. You can do the following:

You can't create another read replica beyond this third cascading read replica in the series for mysql-main. A complete series of instances from an RDS for MySQL source DB instance through to the end of a series of cascading read replicas can consist of at most four DB instances.

For cascading read replicas to work, each source RDS for MySQL DB instance must have automated backups turned on. To turn on automatic backups on a read replica, first create the read replica, and then modify the read replica to turn on automatic backups. For more information, see Creating a read replica.

As with any read replica, you can promote a read replica that's part of a cascade. Promoting a read replica from within a chain of read replicas removes that replica from the chain. For example, suppose that you want to move some of the workload from your mysql-main DB instance to a new instance for use by the accounting department only. Assuming the chain of three read replicas from the example, you decide to promote read-replica-2. The chain is affected as follows:

For more information about promoting read replicas, see Promoting a read replica to be a standalone DB instance.

Monitoring MySQL read replicas

For MySQL read replicas, you can monitor replication lag in Amazon CloudWatch by viewing the Amazon RDS ReplicaLag metric. The ReplicaLag metric reports the value of the Seconds_Behind_Master field of the SHOW REPLICA STATUS command.

Common causes for replication lag for MySQL are the following:

When the ReplicaLag metric reaches 0, the replica has caught up to the source DB instance. If the ReplicaLag metric returns -1, then replication is currently not active. ReplicaLag = -1 is equivalent to Seconds_Behind_Master = NULL.

Starting and stopping replication with MySQL read replicas

You can stop and restart the replication process on an Amazon RDS DB instance by calling the system stored procedures mysql.rds_stop_replication and mysql.rds_start_replication. You can do this when replicating between two Amazon RDS instances for long-running operations such as creating large indexes. You also need to stop and start replication when importing or exporting databases. For more information, see Importing data to an Amazon RDS MariaDB or MySQL database with reduced downtime and Exporting data from a MySQL DB instance by using replication.

If replication is stopped for more than 30 consecutive days, either manually or due to a replication error, Amazon RDS terminates replication between the source DB instance and all read replicas. It does so to prevent increased storage requirements on the source DB instance and long failover times. The read replica DB instance is still available. However, replication can't be resumed because the binary logs required by the read replica are deleted from the source DB instance after replication is terminated. You can create a new read replica for the source DB instance to reestablish replication.

Troubleshooting a MySQL read replica problem

For MySQL DB instances, in some cases read replicas present replication errors or data inconsistencies (or both) between the read replica and its source DB instance. This problem occurs when some binary log (binlog) events or InnoDB redo logs aren't flushed during a failure of the read replica or the source DB instance. In these cases, manually delete and recreate the read replicas. You can reduce the chance of this happening by setting the following parameter values: sync_binlog=1 and innodb_flush_log_at_trx_commit=1. These settings might reduce performance, so test their impact before implementing the changes in a production environment.

The replication technologies for MySQL are asynchronous. Because they are asynchronous, occasional BinLogDiskUsage increases on the source DB instance and ReplicaLag on the read replica are to be expected. For example, a high volume of write operations to the source DB instance can occur in parallel. In contrast, write operations to the read replica are serialized using a single I/O thread, which can lead to a lag between the source instance and read replica. For more information about read-only replicas in the MySQL documentation, see Replication implementation details.

You can do several things to reduce the lag between updates to a source DB instance and the subsequent updates to the read replica, such as the following:

Amazon RDS monitors the replication status of your read replicas and updates the Replication State field of the read replica instance to Error if replication stops for any reason. An example might be if DML queries run on your read replica conflict with the updates made on the source DB instance.

You can review the details of the associated error thrown by the MySQL engine by viewing the Replication Error field. Events that indicate the status of the read replica are also generated, including RDS-EVENT-0045, RDS-EVENT-0046, and RDS-EVENT-0047. For more information about events and subscribing to events, see Working with Amazon RDS event notification. If a MySQL error message is returned, review the error number in the MySQL error message documentation.

One common issue that can cause replication errors is when the value for the max_allowed_packet parameter for a read replica is less than the max_allowed_packet parameter for the source DB instance. The max_allowed_packet parameter is a custom parameter that you can set in a DB parameter group. You use max_allowed_packet to specify the maximum size of DML code that can be run on the database. In some cases, the max_allowed_packet value in the DB parameter group associated with a read replica is smaller than the max_allowed_packet value in the DB parameter group associated with the source DB instance. In these cases, the replication process can throw the error Packet bigger than 'max_allowed_packet' bytes and stop replication. To fix the error, have the source DB instance and read replica use DB parameter groups with the same max_allowed_packet parameter values.

Other common situations that can cause replication errors include the following:

If you decide that you can safely skip an error, you can follow the steps described in the section Skipping the current replication error. Otherwise, you can first delete the read replica. Then you create an instance using the same DB instance identifier so that the endpoint remains the same as that of your old read replica. If a replication error is fixed, the Replication State changes to replicating.