Deploy a stateful MySQL cluster on GKE

Autopilot Standard

---

This document is intended for database administrators, cloud architects, and operations professionals interested in deploying a highly available MySQL topology on Google Kubernetes Engine.

Follow this tutorial to learn how to deploy a MySQL InnoDB Cluster and a MySQL InnoDB ClusterSet, in addition to MySQL Router middleware on your GKE cluster, and how to perform upgrades.

Objectives

In this tutorial, you will learn how to:

• Create and deploy a stateful Kubernetes service.
• Deploy a MySQL InnoDB Cluster for high availability.
• Deploy Router middleware for database operation routing.
• Deploy a MySQL InnoDB ClusterSet for disaster tolerance.
• Simulate a MySQL cluster failover.
• Perform a MySQL version upgrade.

The following sections describe the architecture of the solution you will build in this tutorial.

MySQL InnoDB Cluster

In your regional GKE cluster, using a StatefulSet, you deploy a MySQL database instance with the necessary naming and configuration to create a MySQL InnoDB Cluster. To provide fault tolerance and high availability, you deploy three database instance Pods. This ensures that the majority of Pods on different zones are available at any given time for a successful primary election using a consensus protocol, and makes your MySQL InnoDB Cluster tolerant of single zonal failures.

Figure 1: Example architecture of a single MySQL InnoDB Cluster

Once deployed, you designate one Pod as the primary instance to serve both read and write operations. The other two Pods are secondary read-only replicas. If the primary instance experiences an infrastructure failure, you can promote one of these two replica Pods to become the primary.

In a separate namespace, you deploy three MySQL Router Pods to provide connection routing for improved resilience. Instead of directly connecting to the database service, your applications connect to MySQL Router Pods. Each Router Pod is aware of the status and purpose of each MySQL InnoDB Cluster Pod, and routes application operations to the respective healthy Pod. The routing state is cached in the Router Pods and updated from the cluster metadata stored on each node of the MySQL InnoDB Cluster. In the case of an instance failure, the Router adjusts the connection routing to a live instance.

MySQL InnoDB ClusterSet

You can create a MySQL InnoDB ClusterSet from an initial MySQL InnoDB Cluster. This lets you increase disaster tolerance if the primary cluster is no longer available.

Figure 2: Example multi-region ClusterSet architecture which contains one primary cluster and one replica cluster

If the MySQL InnoDB Cluster primary instance is no longer available, you can promote a replica cluster in the ClusterSet to primary. When using MySQL Router middleware, your application does not need to track the health of the primary database instance. Routing is adjusted to send connections to the new primary after the election has occurred. However, it is your responsibility to ensure that applications connecting to your MySQL Router middleware follow best practices for resilience, so that connections are retried if an error occurs during cluster failover.

Costs

In this document, you use the following billable components of Google Cloud:

• GKE
• Compute Engine

To generate a cost estimate based on your projected usage, use the pricing calculator. New Google Cloud users might be eligible for a free trial.

When you finish the tasks that are described in this document, you can avoid continued billing by deleting the resources that you created. For more information, see Clean up.

Before you begin

Set up your project

1. Sign in to your Google Cloud account. If you're new to Google Cloud, create an account to evaluate how our products perform in real-world scenarios. New customers also get $300 in free credits to run, test, and deploy workloads.

2. In the Google Cloud console, on the project selector page, click Create project to begin creating a new Google Cloud project.

   Go to project selector

3. Make sure that billing is enabled for your Google Cloud project.

4. Enable the GKE API.

   Enable the API

5. In the Google Cloud console, on the project selector page, click Create project to begin creating a new Google Cloud project.

   Go to project selector

6. Make sure that billing is enabled for your Google Cloud project.

7. Enable the GKE API.

   Enable the API

Set up roles

1. Grant roles to your user account. Run the following command once for each of the following IAM roles: role/storage.objectViewer, role/logging.logWriter, role/artifactregistry.Admin, roles/container.clusterAdmin, role/container.serviceAgent, roles/serviceusage.serviceUsageAdmin, roles/iam.serviceAccountAdmin

   gcloud projects add-iam-policy-binding PROJECT_ID --member="user:USER_IDENTIFIER" --role=ROLE

   • Replace PROJECT_ID with your project ID.

   • Replace USER_IDENTIFIER with the identifier for your user account. For example, user:[e-mail].

   • Replace ROLE with each individual role.

Set up your environment

In this tutorial, you use Cloud Shell to manage resources hosted on Google Cloud. Cloud Shell comes preinstalled with Docker and the kubectl and gcloud CLI.

To use Cloud Shell to set up your environment:

1. Set environment variables.

   export PROJECT_ID=PROJECT_ID
   export CLUSTER_NAME=gkemulti-west
   export REGION=COMPUTE_REGION
   

   Replace the following values:

   • PROJECT_ID: your Google Cloud project ID.
   • COMPUTE_REGION: your Compute Engine region. For this tutorial, the region is us-west1. Typically, you want a region that is close to you.

2. Set the default environment variables.

    gcloud config set project PROJECT_ID
    gcloud config set compute/region COMPUTE_REGION
   

3. Clone the code repository.

   git clone https://github.com/GoogleCloudPlatform/kubernetes-engine-samples
   

4. Change to the working directory.

   cd kubernetes-engine-samples/databases/gke-stateful-mysql/kubernetes
   

Create a GKE cluster

In this section, you create a regional GKE cluster. Unlike a zonal cluster, a regional cluster's control plane is replicated into several zones, so an outage in a single zone doesn't make the control plane unavailable.

To create a GKE cluster, follow these steps:

Autopilot

1. In Cloud Shell, create a GKE Autopilot cluster in the us-west1 region.

   gcloud container clusters create-auto $CLUSTER_NAME \
       --region=$REGION
   

2. Get the GKE cluster credentials.

   gcloud container clusters get-credentials $CLUSTER_NAME \
     --region=$REGION
   

3. Deploy a Service across three zones.

   apiVersion: apps/v1
   kind: Deployment
   metadata:
     name: prepare-three-zone-ha
     labels:
       app: prepare-three-zone-ha
   spec:
     replicas: 3
     selector:
       matchLabels:
         app: prepare-three-zone-ha
     template:
       metadata:
         labels:
           app: prepare-three-zone-ha
       spec:
         affinity:
           # Tell Kubernetes to avoid scheduling a replica in a zone where there
           # is already a replica with the label "app: prepare-three-zone-ha"
           podAntiAffinity:
             requiredDuringSchedulingIgnoredDuringExecution:
             - labelSelector:
                 matchExpressions:
                 - key: app
                   operator: In
                   values:
                   - prepare-three-zone-ha
               topologyKey: "topology.kubernetes.io/zone"
         containers:
         - name: prepare-three-zone-ha
           image: busybox:latest
           command:
               - "/bin/sh"
               - "-c"
               - "while true; do sleep 3600; done"
           resources:
             limits:
               cpu: "500m"
               ephemeral-storage: "10Mi"
               memory: "0.5Gi"
             requests:
               cpu: "500m"
               ephemeral-storage: "10Mi"
               memory: "0.5Gi"

   kubectl apply -f prepare-for-ha.yaml
   

   By default, Autopilot provisions resources in two zones. The Deployment defined in prepare-for-ha.yaml ensures that Autopilot provisions nodes across three zones in your cluster, by setting replicas:3, podAntiAffinity with requiredDuringSchedulingIgnoredDuringExecution, and topologyKey: "topology.kubernetes.io/zone".

4. Check the status of the Deployment.

   kubectl get deployment prepare-three-zone-ha --watch
   

   When you see three Pods in the ready state, cancel this command with CTRL+C. The output is similar to the following:

   NAME                    READY   UP-TO-DATE   AVAILABLE   AGE
   prepare-three-zone-ha   0/3     3            0           9s
   prepare-three-zone-ha   1/3     3            1           116s
   prepare-three-zone-ha   2/3     3            2           119s
   prepare-three-zone-ha   3/3     3            3           2m16s
   

5. Run this script to validate that your Pods have been deployed across three zones.

   bash ../scripts/inspect_pod_node.sh default
   

   Each line of the output corresponds to a Pod, and the second column indicates the cloud zone. The output is similar to the following:

   gk3-gkemulti-west1-default-pool-eb354e2d-z6mv us-west1-b prepare-three-zone-ha-7885d77d9c-8f7qb
   gk3-gkemulti-west1-nap-25b73chq-739a9d40-4csr us-west1-c prepare-three-zone-ha-7885d77d9c-98fpn
   gk3-gkemulti-west1-default-pool-160c3578-bmm2 us-west1-a prepare-three-zone-ha-7885d77d9c-phmhj
   

Standard

1. In Cloud Shell, create a GKE Standard cluster in the us-west1 region.

   gcloud container clusters create $CLUSTER_NAME \
     --region=$REGION \
     --machine-type="e2-standard-2" \
     --disk-type="pd-standard" \
     --num-nodes="5"
   

2. Get the GKE cluster credentials.

   gcloud container clusters get-credentials $CLUSTER_NAME \
     --region=$REGION
   

Deploy MySQL StatefulSets

In this section, you deploy one MySQL StatefulSet. Each StatefulSet consists of three MySQL replicas.

To deploy the MySQL StatefulSet, follow these steps:

1. Create a namespace for the StatefulSet.

   kubectl create namespace mysql1
   

2. Create the MySQL secret.

   apiVersion: v1
   kind: Secret
   metadata:
     name: mysql-secret
   type: Opaque
   data:
     password: UGFzc3dvcmQkMTIzNDU2 # Password$123456
     admin-password: UGFzc3dvcmQkMTIzNDU2 # Password$123456

   kubectl apply -n mysql1 -f secret.yaml
   

   The password is deployed with each Pod, and is used by management scripts and commands for MySQL InnoDB Cluster and ClusterSet deployment in this tutorial.

3. Create the StorageClass.

   apiVersion: storage.k8s.io/v1
   kind: StorageClass
   metadata:
     name: fast-storageclass
   provisioner: pd.csi.storage.gke.io
   volumeBindingMode: WaitForFirstConsumer
   reclaimPolicy: Retain
   allowVolumeExpansion: true
   parameters:
     type: pd-balanced

   kubectl apply -n mysql1 -f storageclass.yaml
   

   This storage class uses the pd-balanced Persistent Disk type that balances performance and cost. The volumeBindingMode field is set to WaitForFirstConsumer meaning that GKE delays provisioning of a PersistentVolume until the Pod is created. This setting ensures that the disk is provisioned in the same zone where the Pod is scheduled.

4. Deploy the StatefulSet of MySQL instance Pods.

   apiVersion: apps/v1
   kind: StatefulSet
   metadata:
     name: dbc1
     labels:
       app: mysql
   spec:
     replicas: 3
     selector:
       matchLabels:
         app: mysql
     serviceName: mysql
     template:
       metadata:
         labels:
           app: mysql
       spec:
         topologySpreadConstraints:
         - maxSkew: 1
           topologyKey: "topology.kubernetes.io/zone"
           whenUnsatisfiable: DoNotSchedule
           labelSelector:
             matchLabels:
               app: mysql
         affinity:
           podAntiAffinity:
             requiredDuringSchedulingIgnoredDuringExecution:
             - labelSelector:
                 matchExpressions:
                 - key: app
                   operator: In
                   values:
                   - mysql
               topologyKey: "kubernetes.io/hostname"
         containers:
         - name: mysql
           image: mysql/mysql-server:8.0.28
           command:
           - /bin/bash
           args:
           - -c
           - >-
             /entrypoint.sh
             --server-id=$((20 +  $(echo $HOSTNAME | grep -o '[^-]*$') + 1))
             --report-host=${HOSTNAME}.mysql.mysql1.svc.cluster.local
             --binlog-checksum=NONE
             --enforce-gtid-consistency=ON
             --gtid-mode=ON
             --default-authentication-plugin=mysql_native_password
           env:
           - name: MYSQL_ROOT_PASSWORD
             valueFrom:
               secretKeyRef:
                 name: mysql-secret
                 key: password
           - name: MYSQL_ADMIN_PASSWORD
             valueFrom:
               secretKeyRef:
                 name: mysql-secret
                 key: admin-password
           - name: MYSQL_ROOT_HOST
             value: '%'
           ports:
           - name: mysql
             containerPort: 3306
           - name: mysqlx
             containerPort: 33060
           - name: xcom
             containerPort: 33061
           resources:
             limits:
               cpu: "500m"
               ephemeral-storage: "1Gi"
               memory: "1Gi"
             requests:
               cpu: "500m"
               ephemeral-storage: "1Gi"
               memory: "1Gi"
           volumeMounts:
           - name: mysql
             mountPath: /var/lib/mysql
             subPath: mysql
           readinessProbe:
             exec:
               command:
               - bash
               - "-c"
               - |
                 mysql -h127.0.0.1 -uroot -p$MYSQL_ROOT_PASSWORD -e'SELECT 1'
             initialDelaySeconds: 30
             periodSeconds: 2
             timeoutSeconds: 1
           livenessProbe:
             exec:
               command:
               - bash
               - "-c"
               - |
                 mysqladmin -uroot -p$MYSQL_ROOT_PASSWORD ping
             initialDelaySeconds: 30
             periodSeconds: 10
             timeoutSeconds: 5
     updateStrategy:
       rollingUpdate:
         partition: 0
       type: RollingUpdate
     volumeClaimTemplates:
     - metadata:
         name: mysql
         labels:
           app: mysql
       spec:
         storageClassName: fast-storageclass
         volumeMode: Filesystem
         accessModes:
         - ReadWriteOnce
         resources:
           requests:
             storage: 10Gi

   kubectl apply -n mysql1 -f c1-mysql.yaml
   

   This command deploys the StatefulSet consisting of three replicas. In this tutorial, the primary MySQL cluster is deployed across three zones in us-west1. The output is similar to the following:

   service/mysql created
   statefulset.apps/dbc1 created
   

   In this tutorial, the resource limits and requests are set to minimal values to save cost. When planning for a production workload, make sure to set these values appropriately for your organization's needs.

5. Verify the StatefulSet is created successfully.

   kubectl get statefulset -n mysql1 --watch
   

   It can take about 10 minutes for the StatefulSet to be ready.

6. When all three pods are in a ready state, exit the command using Ctrl+C. If you see PodUnscheduleable errors due to insufficient CPU or memory, wait a few minutes for the control plane to resize to accommodate the large workload.

   The output is similar to the following:

   NAME   READY   AGE
   dbc1   1/3     39s
   dbc1   2/3     50s
   dbc1   3/3     73s
   

7. To inspect the placement of your Pods on the GKE cluster nodes, run this script:

   bash ../scripts/inspect_pod_node.sh mysql1 mysql
   

   The output shows the Pod name, the GKE node name, and the zone where the node is provisioned, and looks similar to the following:

   gke-gkemulti-west-5-default-pool-4bcaca65-jch0 us-west1-b dbc1-0
   gke-gkemulti-west-5-default-pool-1ac6e8b5-ddjx us-west1-c dbc1-1
   gke-gkemulti-west-5-default-pool-1f5baa66-bf8t us-west1-a dbc1-2
   

   The columns in the output represent the hostname, cloud zone, and Pod name, respectively.

   The topologySpreadConstraints policy in the StatefulSet specification (c1-mysql.yaml) directs the scheduler to place the Pods evenly across the failure domain (topology.kubernetes.io/zone).

   The podAntiAffinity policy enforces the constraint that Pods are required to not be placed on the same GKE cluster node (kubernetes.io/hostname). For the MySQL instance Pods, this policy results in the Pods being deployed evenly across the three zones in the Google Cloud region. This placement enables high availability of the MySQL InnoDB Cluster by placing each database instance in a separate failure domain.

Prepare the primary MySQL InnoDB Cluster

To configure a MySQL InnoDB Cluster, follow these steps:

1. In the Cloud Shell terminal, set the group replication configurations for the MySQL instances to be added to your cluster.

   bash ../scripts/c1-clustersetup.sh
   

   POD_ORDINAL_START=${1:-0}
   POD_ORDINAL_END=${2:-2}
   for i in $(seq ${POD_ORDINAL_START} ${POD_ORDINAL_END}); do
     echo "Configuring pod mysql1/dbc1-${i}"
     cat <<'  EOF' | kubectl -n mysql1 exec -i dbc1-${i} -- bash -c 'mysql -uroot -proot --password=${MYSQL_ROOT_PASSWORD}'
   INSTALL PLUGIN group_replication SONAME 'group_replication.so';
   RESET PERSIST IF EXISTS group_replication_ip_allowlist;
   RESET PERSIST IF EXISTS binlog_transaction_dependency_tracking;
   SET @@PERSIST.group_replication_ip_allowlist = 'mysql.mysql1.svc.cluster.local';
   SET @@PERSIST.binlog_transaction_dependency_tracking = 'WRITESET';
     EOF
   done

   The script will remotely connect into each of the three MySQL instances to set and persist the following environment variables:

   • group_replication_ip_allowlist: allows the instance within the cluster to connect to any instance in the group.
   • binlog_transaction_dependency_tracking='WRITESET': allows parallelized transactions which won't conflict.

   In MySQL versions earlier than 8.0.22, use group_replication_ip_whitelist instead of group_replication_ip_allowlist.

2. Open a second terminal, so that you do not need to create a shell for each Pod.

3. Connect to MySQL Shell on the Pod dbc1-0.

   kubectl -n mysql1 exec -it dbc1-0 -- \
       /bin/bash \
       -c 'mysqlsh --uri="root:$[e-mail]"'
   

4. Verify the MySQL group replication allowlist for connecting to other instances.

   \sql SELECT @@group_replication_ip_allowlist;
   

   The output is similar to the following:

   +----------------------------------+
   | @@group_replication_ip_allowlist |
   +----------------------------------+
   | mysql.mysql1.svc.cluster.local   |
   +----------------------------------+
   

5. Verify the server-id is unique on each of the instances.

   \sql SELECT @@server_id;
   

   The output is similar to the following:

   +-------------+
   | @@server_id |
   +-------------+
   |          21 |
   +-------------+
   

6. Configure each instance for MySQL InnoDB Cluster usage and create an administrator account on each instance.

   \js
   dba.configureInstance('[e-mail]', {password: os.getenv("MYSQL_ROOT_PASSWORD"),clusterAdmin: 'icadmin', clusterAdminPassword: os.getenv("MYSQL_ADMIN_PASSWORD")});
   dba.configureInstance('[e-mail]', {password: os.getenv("MYSQL_ROOT_PASSWORD"),clusterAdmin: 'icadmin', clusterAdminPassword: os.getenv("MYSQL_ADMIN_PASSWORD")});
   dba.configureInstance('[e-mail]', {password: os.getenv("MYSQL_ROOT_PASSWORD"),clusterAdmin: 'icadmin', clusterAdminPassword: os.getenv("MYSQL_ADMIN_PASSWORD")});
   

   All instances must have the same username and password in order for the MySQL InnoDB Cluster to function properly. Each command produces output similar to the following:

   ...
   
   The instance 'dbc1-2.mysql:3306' is valid to be used in an InnoDB cluster.
   
   Cluster admin user 'icadmin'@'%' created.
   The instance 'dbc1-2.mysql.mysql1.svc.cluster.local:3306' is already
   ready to be used in an InnoDB cluster.
   
   Successfully enabled parallel appliers.
   

7. Verify that the instance is ready to be used in a MySQL InnoDB Cluster.

   dba.checkInstanceConfiguration()
   

   The output is similar to the following:

   ...
   
   The instance 'dbc1-0.mysql.mysql1.svc.cluster.local:3306' is valid to be used in an InnoDB cluster.
   
   {
       "status": "ok"
   }
   

   Optionally, you can connect to each MySQL instance and repeat this command. For example, run this command to check the status on the dbc1-1 instance:

   kubectl -n mysql1 exec -it dbc1-0 -- \
       /bin/bash \
       -c 'mysqlsh --uri="root:$[e-mail]" \
       --js --execute "dba.checkInstanceConfiguration()"'
   

Create the primary MySQL InnoDB Cluster

Next, create the MySQL InnoDB Cluster using the MySQL Admin createCluster command. Start with the dbc1-0 instance, which will be the primary instance for the cluster, then add two additional replicas to the cluster.

To initialize the MySQL InnoDB Cluster, follow these steps:

1. Create the MySQL InnoDB Cluster.

   var cluster=dba.createCluster('mycluster');
   

   Running the createCluster command triggers these operations:

   • Deploy the metadata schema.
   • Verify that the configuration is correct for Group Replication.
   • Register it as the seed instance of the new cluster.
   • Create necessary internal accounts, such as the replication user account.
   • Start Group Replication.

   This command initializes a MySQL InnoDB Cluster with the host dbc1-0 as the primary. The cluster reference is stored in the cluster variable.

   The output looks similar to the following:

   A new InnoDB cluster will be created on instance 'dbc1-0.mysql:3306'.
   
   Validating instance configuration at dbc1-0.mysql:3306...
   
   This instance reports its own address as dbc1-0.mysql.mysql1.svc.cluster.local:3306
   
   Instance configuration is suitable.
   NOTE: Group Replication will communicate with other instances using
   'dbc1-0.mysql:33061'. Use the localAddress
   option to override.
   
   Creating InnoDB cluster 'mycluster' on
   'dbc1-0.mysql.mysql1.svc.cluster.local:3306'...
   
   Adding Seed Instance...
   Cluster successfully created. Use Cluster.addInstance() to add MySQL
   instances.
   At least 3 instances are needed for the cluster to be able to withstand
   up to one server failure.
   

2. Add the second instance to the cluster.

   cluster.addInstance('[e-mail]', {password: os.getenv("MYSQL_ROOT_PASSWORD"), recoveryMethod: 'clone'});
   

3. Add the remaining instance to the cluster.

   cluster.addInstance('[e-mail]', {password: os.getenv("MYSQL_ROOT_PASSWORD"), recoveryMethod: 'clone'});
   

   The output is similar to the following:

   ...
   The instance 'dbc1-2.mysql:3306' was successfully added to the cluster.
   

4. Verify the cluster's status.

   cluster.status()
   

   This command shows the status of the cluster. The topology consists of three hosts, one primary and two secondary instances. Optionally, you can call cluster.status({extended:1}).

   The output is similar to the following:

   {
       "clusterName": "mysql1",
       "defaultReplicaSet": {
           "name": "default",
           "primary": "dbc1-0.mysql:3306",
           "ssl": "REQUIRED",
           "status": "OK",
           "statusText": "Cluster is ONLINE and can tolerate up to ONE failure.",
           "topology": {
               "dbc1-0.mysql:3306": {
                   "address": "dbc1-0.mysql:3306",
                   "memberRole": "PRIMARY",
                   "mode": "R/W",
                   "readReplicas": {},
                   "replicationLag": null,
                   "role": "HA",
                   "status": "ONLINE",
                   "version": "8.0.28"
               },
               "dbc1-1.mysql:3306": {
                   "address": "dbc1-1.mysql:3306",
                   "memberRole": "SECONDARY",
                   "mode": "R/O",
                   "readReplicas": {},
                   "replicationLag": null,
                   "role": "HA",
                   "status": "ONLINE",
                   "version": "8.0.28"
               },
               "dbc1-2.mysql:3306": {
                   "address": "dbc1-2.mysql:3306",
                   "memberRole": "SECONDARY",
                   "mode": "R/O",
                   "readReplicas": {},
                   "replicationLag": null,
                   "role": "HA",
                   "status": "ONLINE",
                   "version": "8.0.28"
               }
           },
           "topologyMode": "Single-Primary"
       },
       "groupInformationSourceMember": "dbc1-0.mysql:3306"
   }
   

   Optionally, you can call cluster.status({extended:1}) to obtain additional status details.

Create a sample database

To create a sample database, follow these steps:

1. Create a database and load data into the database.

   \sql
   create database loanapplication;
   use loanapplication
   CREATE TABLE loan (loan_id INT unsigned AUTO_INCREMENT PRIMARY KEY, firstname VARCHAR(30) NOT NULL, lastname VARCHAR(30) NOT NULL , status VARCHAR(30) );
   

2. Insert sample data into the database. To insert data, you must be connected to the primary instance of the cluster.

   INSERT INTO loan (firstname, lastname, status) VALUES ( 'Fred','Flintstone','pending');
   INSERT INTO loan (firstname, lastname, status) VALUES ( 'Betty','Rubble','approved');
   

3. Verify that the table contains the three rows inserted in the previous step.

   SELECT * FROM loan;
   

   The output is similar to the following:

   +---------+-----------+------------+----------+
   | loan_id | firstname | lastname   | status   |
   +---------+-----------+------------+----------+
   |       1 | Fred      | Flintstone | pending  |
   |       2 | Betty     | Rubble     | approved |
   +---------+-----------+------------+----------+
   2 rows in set (0.0010 sec)
   

Create a MySQL InnoDB ClusterSet

You can create a MySQL InnoDB ClusterSet to manage replication from your primary cluster to replica clusters, using a dedicated ClusterSet replication channel.

A MySQL InnoDB ClusterSet provides disaster tolerance for MySQL InnoDB Cluster deployments by linking a primary MySQL InnoDB Cluster with one or more replicas of itself in alternate locations, such as multiple zones and multiple regions.

If you closed MySQL Shell, create a new shell by running this command in a new Cloud Shell terminal:

  kubectl -n mysql1 exec -it dbc1-0 -- \
      /bin/bash -c 'mysqlsh \
      --uri="root:$[e-mail]"'

To create a MySQL InnoDB ClusterSet, follow these steps:

1. In your MySQL Shell terminal, obtain a cluster object.

   \js
   cluster=dba.getCluster()
   

   The output is similar to the following:

   <Cluster:mycluster>
   

2. Initialize a MySQL InnoDB ClusterSet with the existing MySQL InnoDB Cluster stored in the cluster object as the primary.

   clusterset=cluster.createClusterSet('clusterset')
   

   The output is similar to the following:

   A new ClusterSet will be created based on the Cluster 'mycluster'.
   
   * Validating Cluster 'mycluster' for ClusterSet compliance.
   
   * Creating InnoDB ClusterSet 'clusterset' on 'mycluster'...
   
   * Updating metadata...
   
   ClusterSet successfully created. Use ClusterSet.createReplicaCluster() to add Replica Clusters to it.
   
   <ClusterSet:clusterset>
   

3. Check the status of your MySQL InnoDB ClusterSet.

   clusterset.status()
   

   The output is similar to the following:

   {
       "clusters": {
           "mycluster": {
               "clusterRole": "PRIMARY",
               "globalStatus": "OK",
               "primary": "dbc1-0.mysql:3306"
           }
       },
       "domainName": "clusterset",
       "globalPrimaryInstance": "dbc1-0.mysql:3306",
       "primaryCluster": "mycluster",
       "status": "HEALTHY",
       "statusText": "All Clusters available."
   }
   

   Optionally, you can call clusterset.status({extended:1}) to obtain additional status details, including information about the cluster.

4. Exit MySQL Shell.

   \q
   

Deploy a MySQL Router

You can deploy a MySQL Router to direct client application traffic to the proper clusters. Routing is based on the connection port of the application issuing a database operation:

• Writes are routed to the primary Cluster instance in the primary ClusterSet.
• Reads can be routed to any instance in the primary Cluster.

When you start a MySQL Router, it is bootstrapped against the MySQL InnoDB ClusterSet deployment. The MySQL Router instances connected with the MySQL InnoDB ClusterSet are aware of any controlled switchovers or emergency failovers and direct traffic to the new primary cluster.

To deploy a MySQL Router, follow these steps:

1. In the Cloud Shell terminal, deploy the MySQL Router.

   kubectl apply -n mysql1 -f c1-router.yaml
   

   The output is similar to the following:

   configmap/mysql-router-config created
   service/mysql-router created
   deployment.apps/mysql-router created
   

2. Check the readiness of the MySQL Router deployment.

   kubectl -n mysql1 get deployment mysql-router --watch
   

   When all three Pods are ready, the output is similar to the following:

   NAME           READY   UP-TO-DATE   AVAILABLE   AGE
   mysql-router   3/3     3            0           3m36s
   

   If you see a PodUnschedulable error in the console, wait a minute or two while GKE provisions more nodes. Refresh, and you should see 3/3 OK.

3. Start MySQL Shell on any member of the existing cluster.

   kubectl -n mysql1 exec -it dbc1-0 -- \
       /bin/bash -c 'mysqlsh --uri="root:$[e-mail]"'
   

   This command connects to the dbc1-0 Pod, then starts a shell connected to the dbc1-0 MySQL instance.

4. Verify the router configuration.

   clusterset=dba.getClusterSet()
   clusterset.listRouters()
   

   The output is similar to the following:

   {
     "domainName": "clusterset",
     "routers": {
       "mysql-router-7cd8585fbc-74pkm::": {
           "hostname": "mysql-router-7cd8585fbc-74pkm",
           "lastCheckIn": "2022-09-22 23:26:26",
           "roPort": 6447,
           "roXPort": 6449,
           "rwPort": 6446,
           "rwXPort": 6448,
           "targetCluster": null,
           "version": "8.0.27"
       },
       "mysql-router-7cd8585fbc-824d4::": {
         ...
       },
       "mysql-router-7cd8585fbc-v2qxz::": {
         ...
       }
     }
   }
   

5. Exit MySQL Shell.

   \q
   

6. Run this script to inspect the placement of the MySQL Router Pods.

   bash ../scripts/inspect_pod_node.sh mysql1 | sort
   

   The script shows the node and Cloud Zone placement of the all of the Pods in the mysql1 namespace, where the output is similar to the following:

   gke-gkemulti-west-5-default-pool-1ac6e8b5-0h9v us-west1-c mysql-router-6654f985f5-df97q
   gke-gkemulti-west-5-default-pool-1ac6e8b5-ddjx us-west1-c dbc1-1
   gke-gkemulti-west-5-default-pool-1f5baa66-bf8t us-west1-a dbc1-2
   gke-gkemulti-west-5-default-pool-1f5baa66-kt03 us-west1-a mysql-router-6654f985f5-qlfj9
   gke-gkemulti-west-5-default-pool-4bcaca65-2l6s us-west1-b mysql-router-6654f985f5-5967d
   gke-gkemulti-west-5-default-pool-4bcaca65-jch0 us-west1-b dbc1-0
   

   You can observe that the MySQL Router Pods are distributed equally across the zones; that is, not placed on the same node as a MySQL Pod, or on the same node as another MySQL Router Pod.

Manage GKE and MySQL InnoDB Cluster upgrades

Updates for both MySQL and Kubernetes are released on a regular schedule. Follow operational best practices to update your software environment regularly. By default, GKE manages cluster and node pool upgrades for you. Kubernetes and GKE also provide additional features to facilitate MySQL software upgrades.

Plan for GKE upgrades

You can take proactive steps and set configurations to mitigate risk and facilitate a smoother cluster upgrade when you are running stateful services, including:

• Standard clusters: Follow GKE best practices for upgrading clusters. Choose an appropriate upgrade strategy to ensure the upgrades happen during the period of the maintenance window:

  • Choose surge upgrades if cost optimization is important and if your workloads can tolerate a graceful shutdown in less than 60 minutes.
  • Choose blue-green upgrades if your workloads are less tolerant of disruptions, and a temporary cost increase due to higher resource usage is acceptable.

  To learn more, see Upgrade a cluster running a stateful workload. Autopilot clusters are automatically upgraded, based on the release channel you selected.

• Use maintenance windows to ensure upgrades happen when you intend them. Before the maintenance window, ensure your database backups are successful.

• Before allowing traffic to the upgraded MySQL nodes, use Readiness Probes and Liveness Probes to ensure they are ready for traffic.

• Create Probes that assess whether replication is in sync before accepting traffic. This can be done through custom scripts, depending on the complexity and scale of your database.

Set a Pod Disruption Budget (PDB) policy

When a MySQL InnoDB Cluster is running on GKE, there must be a sufficient number of instances running at any time to meet the quorum requirement.

In this tutorial, given a MySQL cluster of three instances, two instances must be available to form a quorum. A PodDisruptionBudget policy allows you to limit the number of Pods that can be terminated at any given time. This is useful for both steady state operations of your stateful services and for cluster upgrades.

To ensure that a limited number of Pods are concurrently disrupted, you set the PDB for your workload to maxUnavailable: 1. This ensures that at any point in the service operation, no more than one Pod is not running.

Note: You can also set the minAvailable value to ensure that a minimum number of Pods are running. However, if using minAvailable alone, to guarantee cluster availability, make sure that the value is increased if the size of the cluster increases. In contrast, the maxUnavailable value provides quorum protection for the cluster without any changes; the tradeoff is that only one instance can be disrupted for upgrade at a time.

The following PodDisruptionBudget policy manifest sets the maximum unavailable Pods to one for your MySQL application.

apiVersion: policy/v1
kind: PodDisruptionBudget
metadata:
  name: mysql-pdb
spec:
  maxUnavailable: 1
  selector:
    matchLabels:
      app: mysql

To apply the PDB policy to your cluster, follow these steps:

1. Apply the PDB policy using kubectl.

   kubectl apply -n mysql1 -f mysql-pdb-maxunavailable.yaml
   

2. View the status of the PDB.

   kubectl get poddisruptionbudgets -n mysql1 mysql-pdb -o yaml
   

   In the status section of the output, see the currentHealthy and desiredHealthy Pods counts. The output is similar to the following:

   status:
   ...
     currentHealthy: 3
     desiredHealthy: 2
     disruptionsAllowed: 1
     expectedPods: 3
   ...
   

Plan for MySQL binary upgrades

Kubernetes and GKE provide features to facilitate upgrades for the MySQL binary. However, you need to perform some operations to prepare for the upgrades.

Keep the following considerations in mind before you begin the upgrade process:

• Upgrades should first be carried out in a test environment. For production systems, you should perform further testing in a pre-production environment.
• For some binary releases, you cannot downgrade the version once an upgrade has been performed. Take the time to understand the implications of an upgrade.
• Replication sources can replicate to a newer version. However, copying from a newer to an older version is typically not supported.
• Make sure you have a complete database backup before deploying the upgraded version.
• Keep in mind the ephemeral nature of Kubernetes Pods. Any configuration state stored by the Pod that is not on the persistent volume will be lost when the Pod is redeployed.
• For MySQL binary upgrades, use the same PDB, node pool update strategy, and Probes as described earlier.

In a production environment, you should follow these best practices:

• Create a container image with the new version of MySQL.
• Persist the image build instructions in a source control repository.
• Use an automated image build and testing pipeline such as Cloud Build, and store the image binary in an image registry such as Artifact Registry.

To keep this tutorial simple, you will not build and persist a container image; instead, you use the public MySQL images.

Deploy the upgraded MySQL binary

To perform the MySQL binary upgrade, you issue a declarative command that modifies the image version of the StatefulSet resource. GKE performs the necessary steps to stop the current Pod, deploy a new Pod with the upgraded binary, and attach the persistent disk to the new Pod.

1. Verify that the PDB was created.

   kubectl get poddisruptionbudgets -n mysql1
   

2. Get the list of stateful sets.

   kubectl get statefulsets -n mysql1
   

3. Get the list of running Pods using the app label.

   kubectl get pods --selector=app=mysql -n mysql1
   

4. Update the MySQL image in the stateful set.

   kubectl  -n mysql1 \
       set image statefulset/dbc1 \
       mysql=mysql/mysql-server:8.0.30
   

   The output is similar to the following:

   statefulset.apps/mysql image updated
   

5. Check the status of the terminating Pods and new Pods.

   kubectl get pods --selector=app=mysql -n mysql1
   

Validate the MySQL binary upgrade

During the upgrade, you can verify the status of the rollout, the new Pods, and the existing Service.

1. Confirm the upgrade by running the rollout status command.

   kubectl rollout status statefulset/dbc1 -n mysql1
   

   The output is similar to the following:

   partitioned roll out complete: 3 new pods have been updated...
   

2. Confirm the image version by inspecting the stateful set.

   kubectl get statefulsets -o wide -n mysql1
   

   The output is similar to the following:

   NAME   READY   AGE   CONTAINERS   IMAGES
   dbc1   3/3     37m   mysql        mysql/mysql-server:8.0.30
   

3. Check the status of the cluster.

   kubectl -n mysql1 \
        exec -it dbc1-0 -- \
          /bin/bash \
            -c 'mysqlsh \
            --uri="root:$[e-mail]" \
            --js \
            --execute "print(dba.getClusterSet().status({extended:1})); print(\"\\n\")"'
   

   For each cluster instance, look for the status and version values in the output. The output is similar to the following:

   ...
     "status": "ONLINE",
     "version": "8.0.30"
   ...
   

Rollback the last app deployment rollout

Warning: Some binary versions cannot be downgraded. Understand the implications and constraints before performing a binary upgrade.

When you revert the deployment of an upgraded binary version, the rollout process is reversed and a new set of Pods is deployed with the previous image version.

To revert the deployment to the previous working version, use the rollout undo command:

kubectl rollout undo statefulset/dbc1 -n mysql1

The output is similar to the following:

statefulset.apps/dbc1 rolled back

Scale your database cluster horizontally

To scale your MySQL InnoDB Cluster horizontally, you add additional nodes to the GKE cluster node pool (only required if you are using Standard), deploy additional MySQL instances, then add each instance to the existing MySQL InnoDB Cluster.

Add nodes to your Standard cluster

This operation is not needed if you are using a Autopilot cluster.

To add nodes to your Standard cluster, follow the instructions below for Cloud Shell or the Google Cloud console. For detailed steps, see Resize a node pool

gcloud

In Cloud Shell, resize the default node pool to eight instances in each managed instance group.

gcloud container clusters resize ${CLUSTER_NAME} \
     --node-pool default-pool \
     --num-nodes=8

Console

To add nodes to your Standard cluster:

1. Open the gkemulti-west1 Cluster page in the Google Cloud console.
2. Select Nodes, and click on default pool.
3. Scroll down to Instances groups.
4. For each instance group, resize the Number of nodes value from 5 to 8 nodes.

Add MySQL Pods to the primary cluster

To deploy additional MySQL Pods to scale your cluster horizontally, follow these steps:

1. In Cloud Shell, update the number of replicas in the MySQL deployment from three replicas to five replicas.

   kubectl scale  -n mysql1 --replicas=5 -f c1-mysql.yaml
   

2. Verify the progress of the deployment.

   kubectl -n mysql1 get pods --selector=app=mysql -o wide
   

   To determine whether the Pods are ready, use the --watch flag to watch the deployment. If you are using Autopilot clusters and see Pod Unschedulable errors, this might indicate GKE is provisioning nodes to accommodate the additional Pods.

3. Configure the group replication settings for the new MySQL instances to add to the cluster

   bash ../scripts/c1-clustersetup.sh 3 4
   

   The script submits the commands to the instances running on the Pods with ordinals 3 through 4.

4. Open MySQL Shell.

   kubectl -n mysql1 \
     exec -it dbc1-0 -- \
         /bin/bash \
           -c 'mysqlsh \
           --uri="root:$[e-mail]"'
   

5. Configure the two new MySQL instances.

   dba.configureInstance('root:$[e-mail]', {password: os.getenv("MYSQL_ROOT_PASSWORD"),clusterAdmin: 'icadmin', clusterAdminPassword: os.getenv("MYSQL_ADMIN_PASSWORD")});
   dba.configureInstance('root:$[e-mail]', {password: os.getenv("MYSQL_ROOT_PASSWORD"),clusterAdmin: 'icadmin', clusterAdminPassword: os.getenv("MYSQL_ADMIN_PASSWORD")});
   

   The commands check if the instance is configured properly for MySQL InnoDB Cluster usage and perform the necessary configuration changes.

6. Add one of the new instances to the primary cluster.

   cluster = dba.getCluster()
   cluster.addInstance('[e-mail]', {password: os.getenv("MYSQL_ROOT_PASSWORD"), recoveryMethod: 'clone'});
   

7. Add a second new instance to the primary cluster.

   cluster.addInstance('[e-mail]', {password: os.getenv("MYSQL_ROOT_PASSWORD"), recoveryMethod: 'clone'});
   

8. Obtain the ClusterSet status, which also includes the Cluster status.

   clusterset = dba.getClusterSet()
   clusterset.status({extended: 1})
   

   The output is similar to the following:

   "domainName": "clusterset",
   "globalPrimaryInstance": "dbc1-0.mysql:3306",
   "metadataServer": "dbc1-0.mysql:3306",
   "primaryCluster": "mycluster",
   "status": "HEALTHY",
   "statusText": "All Clusters available."
   

9. Exit MySQL Shell.

   \q
   

Clean up

To avoid incurring charges to your Google Cloud account for the resources used in this tutorial, either delete the project that contains the resources, or keep the project and delete the individual resources.

Delete the project

The easiest way to avoid billing is to delete the project you created for the tutorial.

Caution: Deleting a project has the following effects:

• Everything in the project is deleted. If you used an existing project for the tasks in this document, when you delete it, you also delete any other work you've done in the project.
• Custom project IDs are lost. When you created this project, you might have created a custom project ID that you want to use in the future. To preserve the URLs that use the project ID, such as an appspot.com URL, delete selected resources inside the project instead of deleting the whole project.

If you plan to explore multiple architectures, tutorials, or quickstarts, reusing projects can help you avoid exceeding project quota limits.

Delete a Google Cloud project:

gcloud projects delete PROJECT_ID

What's next

• Learn more about how the Google Cloud Observability MySQL integration collects performance metrics related to InnoDB.
• Learn more about backup for GKE, a service for backing up and restoring workloads in GKE.
• Explore Persistent Volumes in more detail.