Manage read replicas

This page describes how to manage read replicas. These operations include disabling and enabling replication, promoting a replica, configuring parallel replication, and checking the replication status.

For more information about how replication works, see Replication in Cloud SQL.

Disable replication

By default, a replica starts with replication enabled. However, you can disable replication, for example, to debug or analyze the state of an instance. When you are ready, you explicitly re-enable replication. Disabling or re-enabling replication doesn't restart the replica instance.

Disabling replication does not stop the replica instance; it becomes a read-only instance that is no longer replicating from its primary instance. You continue to be charged for the instance. On the disabled replica, you can re-enable replication, delete the replica, or promote the replica to a stand-alone instance.

When you disable the replication for an extended period of time, your disk storage requirements might increase. For example, your instance might accumulate transactional logs to let you resume replication when you re-enable replication. To avoid increasing disk storage requirements, instead of disabling replication for an extended period of time, consider promoting the replica or creating a clone of the primary instance.

To disable replication:

Console

  1. In the Google Cloud console, go to the Cloud SQL Instances page.

    Go to Cloud SQL Instances

  2. Select a replica instance by clicking its name.
  3. Click Disable replication in the button bar.
  4. Click OK.

gcloud

gcloud sql instances patch REPLICA_NAME \
--no-enable-database-replication

REST v1

To execute this cURL command at a command line prompt, you acquire an access token by using the gcloud auth print-access-token command. You can also use the APIs Explorer on the Instances:patch page to send the REST API request.

Before using any of the request data, make the following replacements:

  • project-id: The project ID
  • replica-name: The name of the replica instance

HTTP method and URL: 

PATCH https://sqladmin.googleapis.com/v1/projects/project-id/instances/replica-name

Request JSON body: 

{
  "settings":
  {
    "databaseReplicationEnabled": "False"
  }
}

To send your request, expand one of these options:

You should receive a JSON response similar to the following:

REST v1beta4

To execute this cURL command at a command line prompt, you acquire an access token by using the gcloud auth print-access-token command. You can also use the APIs Explorer on the Instances:patch page to send the REST API request.

Before using any of the request data, make the following replacements:

  • project-id: The project ID
  • replica-name: The name of the replica instance

HTTP method and URL: 

PATCH https://sqladmin.googleapis.com/sql/v1beta4/projects/project-id/instances/replica-name

Request JSON body: 

{
  "settings":
  {
    "databaseReplicationEnabled": "False"
  }
}

To send your request, expand one of these options:

You should receive a JSON response similar to the following:

Enable replication

If a replica has not been replicating for a long time, it will take longer for it to catch up to the primary instance. In this case, delete the replica and create a new one.

To enable replication:

Console

  1. In the Google Cloud console, go to the Cloud SQL Instances page.

    Go to Cloud SQL Instances

  2. Select a replica instance by clicking its name.
  3. Click Enable replication.
  4. Click Ok.

gcloud

gcloud sql instances patch REPLICA_NAME \
--enable-database-replication

REST v1

To execute this cURL command at a command line prompt, you acquire an access token by using the gcloud auth print-access-token command. You can also use the APIs Explorer on the Instances:patch page to send the REST API request.

Before using any of the request data, make the following replacements:

  • project-id: The project ID
  • replica-name: The name of the replica instance

HTTP method and URL: 

PATCH https://sqladmin.googleapis.com/v1/projects/project-id/instances/replica-name

Request JSON body: 

{
  "settings":
  {
    "databaseReplicationEnabled": "True"
  }
}

To send your request, expand one of these options:

You should receive a JSON response similar to the following:

REST v1beta4

To execute this cURL command at a command line prompt, you acquire an access token by using the gcloud auth print-access-token command. You can also use the APIs Explorer on the Instances:patch page to send the REST API request.

Before using any of the request data, make the following replacements:

  • project-id: The project ID
  • replica-name: The name of the replica instance

HTTP method and URL: 

PATCH https://sqladmin.googleapis.com/sql/v1beta4/projects/project-id/instances/replica-name

Request JSON body: 

{
  "settings":
  {
    "databaseReplicationEnabled": "True"
  }
}

To send your request, expand one of these options:

You should receive a JSON response similar to the following:

Promote a replica

Promoting a read replica stops replication and converts the instance to a standalone Cloud SQL primary instance with read and write capabilities.

When promoted, read replicas are automatically configured with backups, but they aren't automatically configured as high availability (HA) instances. You can enable high availability after promoting the replica just as you would for any non-replica instance. Configuring a read replica for high availability is done the same way as for a primary instance. Learn more about configuring the instance for high availability.

Before promoting a read replica, if the primary is still available and serving clients, you should do the following:

  1. Stop all writes to the primary instance.
  2. Check the replication status of the replica (follow the instructions in the mysql Client tab).
  3. Verify that the replica is replicating, and then wait until the replication lag reported by the Seconds_Behind_Master metric is 0.

Otherwise, a newly promoted instance may be missing some transactions that were committed to the primary instance.

To promote a replica to a standalone instance:

Console

  1. In the Google Cloud console, go to the Cloud SQL Instances page.

    Go to Cloud SQL Instances

  2. Select a replica instance by clicking its name.
  3. Click Promote replica.
  4. Click Ok.

gcloud

gcloud sql instances promote-replica REPLICA_NAME

REST v1

To execute this cURL command at a command line prompt, you acquire an access token by using the gcloud auth print-access-token command. You can also use the APIs Explorer on the Instances:promoteReplica page to send the REST API request.

Before using any of the request data, make the following replacements:

  • project-id: The project ID
  • replica-name: The name of the replica instance

HTTP method and URL: 

POST https://sqladmin.googleapis.com/v1/projects/project-id/instances/replica-name/promoteReplica

To send your request, expand one of these options:

You should receive a JSON response similar to the following:

REST v1beta4

To execute this cURL command at a command line prompt, you acquire an access token by using the gcloud auth print-access-token command. You can also use the APIs Explorer on the Instances:promoteReplica page to send the REST API request.

Before using any of the request data, make the following replacements:

  • project-id: The project ID
  • replica-name: The name of the replica instance

HTTP method and URL: 

POST https://sqladmin.googleapis.com/sql/v1beta4/projects/project-id/instances/replica-name/promoteReplica

To send your request, expand one of these options:

You should receive a JSON response similar to the following:

Confirm that the promoted instance is configured correctly. In particular, consider configuring the instance for high availability if needed.

Configure parallel replication

Reducing replication lag is important for managing replication performance. Replication lag occurs when the updates to a read replica fall behind the updates to the primary instance. This section describes how users can enable parallel replication, which can reduce replication lag.

In MySQL replication, a replication SQL thread is used to execute the transactions that are collected in the relay log on the read replica. Parallel replication reduces replication lag by increasing the number of SQL threads that work to execute these transactions. Read replicas with parallel replication enabled are sometimes called multithreaded replicas.

Parallel replication is available in these three scenarios in Cloud SQL for MySQL:

For simplicity, this page uses the terms "primary instance" and "read replica."

Basic steps to change parallel replication flags

The steps for enabling parallel replication are the following:

  1. On a read replica, disable replication.
  2. On the read replica, set the flags for parallel replication. Use the gcloud command to set the flags. The Google Cloud console option is disabled when replication is disabled.
  3. On the read replica, enable replication.
  4. Optionally, on the primary instance, set the flags to optimize performance for parallel replication.

Read replicas: Flags for parallel replication

Cloud SQL for MySQL supports several flags for parallel replication on read replicas. For information about the flags, click these links to the MySQL 8.0 documentation:

Changing these flags does not restart the read replica.

The following table contains the allowed ranges and default values for these flags:

Read replica flag Allowed values MySQL 5.7 default value MySQL 8.0 and later default value
replica_parallel_workers 0-1024 0 0 (MySQL 8.0.26 and earlier)
4 (MySQL 8.0.27 and later)
replica_parallel_type DATABASE, LOGICAL_CLOCK DATABASE DATABASE (MySQL 8.0.26 and earlier)
LOGICAL_CLOCK (MySQL 8.0.27 and later)
replica_preserve_commit_order ON, OFF OFF OFF (MySQL 8.0.26 and earlier)
ON (MySQL 8.0.27 and later)
replica_pending_jobs_size_max 1024-1GB 16MB 128MB

The replica_preserve_commit_order flag prevents gaps in the sequence of transactions executed from the replica's relay log.

The replica_preserve_commit_order=1 setting requires the following:

The replica_pending_jobs_size_max flag sets the maximum memory, in bytes, available to applier queues holding events not yet applied.

Primary instance: Flags for parallel replication

Cloud SQL for MySQL supports several flags for use on a primary instance. You can use these flags to tune replication performance for associated read replicas with parallel replication enabled. For information about the flags, click these links to the MySQL 8.0 documentation:

Changing these flags does not restart the primary instance.

The following table contains the allowed ranges and default values for these flags:

Primary instance flag Allowed values MySQL 5.7 default value MySQL 8.0 default value MySQL 8.4 default value
binlog_transaction_dependency_history_size 1-1000000 25000 25000 25000
binlog_transaction_dependency_tracking COMMIT_ORDER, WRITESET, WRITESET_SESSION COMMIT_ORDER WRITESET n/a (deprecated in MySQL 8.4)
transaction_write_set_extraction OFF, MURMUR32, XXHASH64 OFF XXHASH64 n/a (deprecated in MySQL 8.4)

In MySQL 5.7, if binlog_transaction_dependency_tracking is set to WRITESET or WRITESET_SESSION, then transaction_write_set_extraction should be set to a non-OFF value (XXHASH64 or MURMUR32).

Check replication status

When you view a replica instance using the Google Cloud console or log into the instance using an administration client, you get details about replication, including status and metrics. When you use the gcloud CLI, you get a brief summary of the replication configuration.

Before checking the replication status for a Cloud SQL replica instance, use the
gcloud sql instances describe command to display the status of the instance. As a result, you can see whether replication is enabled for the replica instance.

The following metrics are available for replica instances. (Learn more about additional metrics available for all instances, including non-replica instances.)

MetricDescription
Replication State
(cloudsql.googleapis.com/database/replication/state)

Indicates whether replication is actively streaming logs from the primary to the replica. Possible values are:

  • Running
  • Stopped
  • Error

This metric reports Running if both the replica's I/O and SQL threads report they are running. See the Slave I/O thread running state and Slave SQL thread running state metrics below for more information, or consult Checking Replication Status in the MySQL Reference Manual.
Replication Lag
(cloudsql.googleapis.com/database/replication/replica_lag)

The amount of time that the replica's state is lagging behind the state of the primary instance. This is the difference between (1) the current time and (2) the original timestamp at which the primary committed the transaction that is currently being applied on the replica. In particular, writes may be counted as lagging even if they have been received by the replica, if the replica has not yet applied the write to the database.

For cascading replicas, each primary-replica pair is monitored separately and there's no single metric that yields the end-to-end (primary to replica) lag.

This metric reports the value of Seconds_Behind_Master when SHOW REPLICA STATUS is run on the replica. For more information, see Checking Replication Status in the MySQL Reference Manual.
Network Lag
(cloudsql.googleapis.com/database/replication/network_lag)

The amount of time, in seconds that it takes from writing the binlog in the primary database to reach the IO thread in the replica.

If the network_lag is zero, or negligible, but the `replica_lag` is high, it indicates that the SQL thread is not able to apply replication changes fast enough.
Slave I/O thread running state
(cloudsql.googleapis.com/database/mysql/replication/slave_io_running_state)

Indicates whether the I/O thread for reading the primary instance's binary log is running on the replica. Possible values are:

  • Yes
  • No
  • Connecting

This metric reports the value of Slave_IO_Running when SHOW REPLICA STATUS is run on the replica. For more information, see Checking Replication Status in the MySQL Reference Manual.
Slave SQL thread running state
(cloudsql.googleapis.com/database/mysql/replication/slave_sql_running_state)

Indicates whether the SQL thread for executing events in the relay log is running on the replica. Possible values are:

  • Yes
  • No
  • Connecting

This metric reports the value of Slave_SQL_Running when SHOW REPLICA STATUS is run on the replica. For more information, see Checking Replication Status in the MySQL Reference Manual.

To check replication status:

Console

Cloud SQL reports the Replication State and Replication Lag metrics on the default Cloud SQL monitoring dashboard.

To view other metrics for in-region and cross-region replicas, and replicas of external servers, create a custom dashboard and add the metrics you wish to monitor to it:

  1. In the Google Cloud console, go to the Monitoring page.

    Go to Monitoring

  2. Select the Dashboards tab.
  3. Click Create dashboard.
  4. Give the dashboard a name and click OK.
  5. Click Add chart.
  6. For Resource Type select Cloud SQL Database.
  7. Do any of the following:
    1. To monitor the replication state metric: in the Select a metric field, type Replication state. Then add a filter for state = "Running". The chart shows 1 if replication is running and 0 otherwise.
    2. To monitor the replication lag metric: in the Select a metric field, type replica_lag. The chart shows the amount of time that the replica's state lags behind that of its primary.
    3. To monitor the status of the replica's I/O thread: in the Select a metric field, type Slave I/O thread running state. Then add a filter on state = "Yes". The chart shows 1 if the thread is running and 0 otherwise.
    4. To monitor the status of the replica's SQL thread: in the Select a metric field, type Slave SQL thread running state. Then add a filter on state = "Yes". The chart shows 1 if the thread is running and 0 otherwise.

gcloud

For a replica instance, check the replication status with:

gcloud sql instances describe REPLICA_NAME

In the output, look for the properties databaseReplicationEnabled and masterInstanceName.

For a primary instance, check if there are replicas with:

gcloud sql instances describe PRIMARY_INSTANCE_NAME

In the output, look for the property replicaNames.

mysql Client

  1. Connect to the replica with a MySQL client.

    For information, see Connection Options for External Applications.

  2. Check the replica's status: 
    SHOW REPLICA STATUS \G

    Look for the following metrics in the output of the command:

    • Master_Host: The name of the primary instance.
    • Slave_IO_Running, Slave_SQL_Running: Whether the I/O and SQL threads, respectively, are running. These threads are responsible for transferring events from the primary to the replica's relay log and executing those events from the relay log. The value of the metric is Yes if the thread is running. Both threads must be running for replication to be active.
    • Seconds_Behind_Master: The amount of time, in seconds, by which the replica lags in processing the primary's transactions, i.e. the difference between (1) the current time and (2) the original timestamp at which the primary committed the transaction that is currently being applied on the replica. The value is NULL if replication is broken.
    • Master_Log_file, Read_Master_Log_Pos, Relay_Master_Log_File, Exec_Master_Log_Pos: These metrics show the coordinates (filename and offset) that the I/O thread has read events up to (Master_Log_file and Read_Master_Log_Pos) and that the SQL thread has executed events up to (Relay_Master_Log_File and Exec_Master_Log_Pos). If they are the same (i.e. Master_Log_file is equal to Relay_Master_Log_File and Read_Master_Log_Pos is equal to Exec_Master_Log_Pos) then the replica has processed all of the events it has received from the primary.

For more details about the output from this command, see the MySQL documentation on Checking Replication Status.

Troubleshoot

Issue Troubleshooting
Read replica didn't start replicating on creation. There's probably a more specific error in the log files. Inspect the logs in Cloud Logging to find the actual error.
Unable to create read replica - invalidFlagValue error. One of the flags in the request is invalid. It could be a flag you provided explicitly or one that was set to a default value.

First, check that the value of the max_connections flag is greater than or equal to the value on the primary.

If the max_connections flag is set appropriately, inspect the logs in Cloud Logging to find the actual error.
Unable to create read replica - unknown error. There's probably a more specific error in the log files. Inspect the logs in Cloud Logging to find the actual error.

If the error is: set Service Networking service account as servicenetworking.serviceAgent role on consumer project, then disable and re-enable the Service Networking API. This action creates the service account necessary to continue with the process.
Disk is full. The primary instance disk size can become full during replica creation. Edit the primary instance to upgrade it to a larger disk size.
The replica instance is using too much memory. The replica uses temporary memory to cache often-requested read operations, which can lead it to use more memory than the primary instance.

Restart the replica instance to reclaim the temporary memory space.
Replication stopped. The maximum storage limit was reached and automatic storage increase isn't enabled.

Edit the instance to enable automatic storage increase.
Replication lag is consistently high. The write load is too high for the replica to handle. Replication lag takes place when the SQL thread on a replica is unable to keep up with the IO thread. Some kinds of queries or workloads can cause temporary or permanent high replication lag for a given schema. Some of the typical causes of replication lag are:
  • Slow queries on the replica. Find and fix them.
  • All tables must have a unique/primary key. Every update on such a table without a unique/primary key causes full table scans on th replica.
  • Queries like DELETE ... WHERE field < 50000000 cause replication lag with row-based replication since a huge number of updates are piled up on the replica.

Some possible solutions include:
Replication lag suddenly spikes. This is caused by long-running transaction(s). When a transaction (single statement or multi-statements) commits on the source instance, the start time of the transaction is recorded in the binary log. When the replica receives this binlog event, it compares that timestamp with the current timestamp to calculate replication lag. Hence, a long-running transaction on the source would result in an immediate large replication lag on the replica. If the amount of row changes in the transaction is large, the replica would also spend a long time to execute it. During the time, replication lag is increasing. Once the replica finishes this transaction, the catch up period would depend on the write workload on the source and the replica's processing speed.

To avoid a long transaction, some possible solutions include:

  • Break the transaction into multiple small transactions
  • Chunk a single large write query into smaller batches
  • Try to separate long SELECT queries from a transaction mixed with DMLs
Changing parallel replication flags results in an error. An incorrect value is set for one of or more of these flags.

On the primary instance that's displaying the error message, set the parallel replication flags:

  1. Modify the binlog_transaction_dependency_tracking and transaction_write_set_extractionflags:
    • binlog_transaction_dependency_tracking=COMMIT_ORDER
    • transaction_write_set_extraction=OFF
  2. Add the slave_pending_jobs_size_max flag:

    slave_pending_jobs_size_max=33554432

  3. Modify the transaction_write_set_extraction flag:

    transaction_write_set_extraction=XXHASH64

  4. Modify the binlog_transaction_dependency_tracking flag:

    binlog_transaction_dependency_tracking=WRITESET
Replica creation fails with timeout. Long-running uncommitted transactions on the primary instance can cause read replica creation to fail.

Recreate the replica after stopping all running queries.

What's next