Since we run ProxySQL in k8s at work, I needed a test cluster that I could more easily iterate on. Obviously, this isn't exaclty what our real setup looks like, aside from maybe a few names (persona-web, etc), but it's close enough that I can test changes without affecting staging or prod.
This repo uses the MySQL helm charts from Bitnami to setup the mysql backends. The limitation here is we can only have one replica per primary, which doesn't really match our infra, but is good enough for now. In the future perhaps I will find another chart or modify this one to setup two replicas.
This assumes you have a k8s cluster on hand; we're installing some helm charts into it. I'm just using orbstack as my local k8s cluster, but I don't think the implementation really matters a lot for this.
Note: the latest mysql in homebrew is not compatible with the proxysql admin interface; you either need to use [email protected]
or just exec into the contoller pods to access the mysql admin interface.
We've gotten rid of the ruby tooling and we've moved to using the proxysql-agent sidecar. The helm charts now pull the image from the ghcr.io registry, so no extra steps are needed here.
Run ./bin/setup.sh to create the test infra:
- A us1 mysql instance with one read only replica
- A us2 mysql instance with one read only replica
- A
proxysql-core
deployment consisting of 3 pods- This cluster controls the configuration of the rest of the cluster, and does not serve sql traffic
- A
proxysql-satellite
deployment consisting of 3 pods- This cluster gets configuration from the core, and is what serves the sql traffic
We'll create one primary and one secondary mysql instance, and make sure they are replicating. We're using the bitnami mysql chart (and changing values on the command line) for this install.
NB: the mysql charts use PVCs, so don't be surprised when you uninstall/reinstall the chart and the mysql data hasn't reset itself. The ./bin/teardown.sh script will delete everything, if you want a fresh start.
kubectl get namespace mysql > /dev/null 2>&1 \
|| kubectl create ns mysql
kubectl get configmap -n mysql us1-initdb > /dev/null 2>&1 \
|| kubectl create configmap -n mysql us1-initdb --from-file=./helm/data/mysql-us1.sql
kubectl get configmap -n mysql us2-initdb > /dev/null 2>&1 \
|| kubectl create configmap -n mysql us2-initdb --from-file=./helm/data/mysql-us2.sql
helm install mysql-us1 -n mysql ./helm/mysql \
--set nameOverride="mysql-us1" \
--set architecture="replication" \
--set auth.rootPassword="rootpw" \
--set auth.replicationPassword="replication" \
--set auth.database="persona-web-us1" \
--set auth.username="persona-web-us1" \
--set auth.password="persona-web-us1" \
--set initdbScriptsConfigMap="us1-initdb"
helm install mysql-us2 -n mysql ./helm/mysql \
--set nameOverride="mysql-us2" \
--set architecture="replication" \
--set auth.rootPassword="rootpw" \
--set auth.replicationPassword="replication" \
--set auth.database="persona-web-us2" \
--set auth.username="persona-web-us2" \
--set auth.password="persona-web-us2" \
--set initdbScriptsConfigMap="us2-initdb"
For this step, we're creating a proxysql core statefulset cluster and a proxysql satellite deployment cluster. The core cluster is the "leader" and is in charge of distributing the configuration changes to the satellite cluster. The satellites are configured to automatically connect to the core service on boot.
If you want to just test proxysql changes, yuou can run ./bin/reset-proxysql.sh to reinstall it without having to sit thruogh a full teardown/setup. This is mostly useful when I am testing changes to the proxysql-agent docker image.
kubectl get namespace proxysql > /dev/null 2>&1 \
|| kubectl create ns proxysql
helm install proxysql-core -n proxysql ./helm/proxysql/core
helm install proxysql-satellite -n proxysql ./helm/proxysql/satellite
- database: persona-web-us1
- username: persona-web-us1
- password: persona-web-us1
- database: persona-web-us2
- username: persona-web-us2
- password: persona-web-us2
This connects to the database as root, using the root password k8s secret (which is just... rootpw
). You can connect to any service avaiable in the mysql namespace, as long as they have a ClusterIP (ie: not the headless services).
mysql -h$(kubectl get services -n mysql mysql-us1-primary -o jsonpath='{.spec.clusterIP}') -uroot -prootpw persona-web-us1
There is the assumption here that you have the mysql
command available; if not, you can use the docker image from bitnami:
kubectl run mysql-us2-client --rm --tty -i --restart='Never' --image docker.io/bitnami/mysql:8.0.34-debian-11-r56 --namespace mysql --env MYSQL_ROOT_PASSWORD=$MYSQL_ROOT_PASSWORD --command -- bash
Connect to us1 or us2 via proxysql
mysql -h$(kubectl get service proxysql-satellite -n proxysql --output jsonpath='{.spec.clusterIP}') -P6033 -upersona-web-us1 -ppersona-web-us1 persona-web-us1
mysql -h$(kubectl get service proxysql-satellite -n proxysql --output jsonpath='{.spec.clusterIP}') -P6033 -upersona-web-us2 -ppersona-web-us2 persona-web-us2
helm uninstall -n proxysql --ignore-not-found proxysql-core
helm uninstall -n proxysql --ignore-not-found proxysql-satellite
helm uninstall -n mysql --ignore-not-found mysql-us1
helm uninstall -n mysql --ignore-not-found mysql-us2
kubectl delete ns --ignore-not-found proxysql
kubectl delete ns --ignore-not-found mysql