Workload Manager best practices for MySQL

Instances in the evaluation scope must have the Google Cloud Agent for Compute Workloads configured to run Workload Manager evaluations. If you have not configured the agent correctly, evaluation results can be incomplete or inaccurate. The minimum recommended version is 1.2.

To ensure that the VM restarts automatically in the event of a failure, enable the Compute Engine automatic restart policy for any VM that is running a MySQL workload.

The default provisioned IOPS and throughput values for small Hyperdisks can be too low to guarantee proper performance. We recommend that you raise these values to a minimum of 10,000 IOPS and 1 GiB/s throughput.

For more information, see Default performance.

To prevent any platform maintenance events from stopping or restarting a VM that is running MySQL workloads, the onHostMaintenance parameter for the VM must be set to the recommended option MIGRATE.

For more information, see Set VM host maintenance policy.

Google Cloud recommends that you do not run MySQL workloads on PD Standard and Hyperdisk Throughput disk types, because these are hard disk-based disks and might lead to degraded performance.

For more information, see Persistent disk types, Hyperdisks, and Hyperdisk Throughput.

To maintain proper high availability, we strongly recommend that you send your replication to a machine in a different zone or region from the primary location.

For more information, see Well-Architected Framework: Reliability pillar.

For optimized performance, Google Cloud recommends allocating the innodb_buffer_pool_size parameter a 50% or greater share of the machine's memory. This ensures that the buffer pool has enough memory to store data and improve performance. This guideline doesn't apply to smaller VMs with less than 4 GB of total memory where such an allocation might not be possible.