doltclusterctl

README

doltclusterctl

A small control plane job for interacting with a dolt cluster running within kubernetes.

This binary perform certain operational tasks on the state of the sql-servers and the labels for the pods in a dolt cluster. Designed to be run within the Kubernetes cluster with a service account that has permissions to list and edit pods and StatefulSets.

Usage

Generally run within a cluster using kubectl, and run with a serviceAccount that has permissions to read StatefulSets and to read and edit pods.

For example, something like:

kubectl run -i --tty \
    --image localhost:5000/doltclusterctl:latest \
    --image-pull-policy Always \
    --restart=Never \
    --rm \
    --overrides '{"apiVersion": "v1","kind": "Pod","spec": {"serviceAccountName": "doltclusterctl"}}' \
    -n dolt \
    doltclusterctl -- \
    doltclusterctl -n dolt applyprimarylabels dolt

The -n NAMESPACE flag tells the binary which namespace the StatefulSet lives in. The default is default.

The next parameter is the operation to run. The operations are:

• applyprimarylabels
• gracefulfailover
• promotestandby
• rollingrestart

The last parameter is the name of the stateful set on which to operate.

Operations

applyprimarylabels is the simplest operation. It queries the state of the dolt servers running within the stateful set and looks for the server instance which is currently the primary. If it finds it, it applys appropriate labels to the Pods in the StatefulSet so that traffic will be routed to the primary.

The traffic routing should be setup to route write traffic to the Pod which is labeled with dolthub.com/cluster_role=primary.

gracefulfailover can be used to shift traffic away from the current primary. It:

1. Removes routing of traffic to the current primary.
2. Makes the current primary assume role standby.
3. Makes a different chosen standby server assume role primary.
4. Starts routing traffic to the new primary.

For a cluster with more than one standby, a useful option is -min-caughtup-standbys N, which can be given to gracefulfailover. In that case, gracefulfailover will succeed as long as at least N standbys can be caught up. The standby which gets promoted to primary will be one of the standbys which was successfully caught up. This option requires Dolt 1.6.0 or higher.

promotestandby is more aggressive in its behavior. Without causing the existing primary to assume role standby, it makes a server in the cluster which is currently a standby into the new primary and begins routing traffic to it. This should only be used if gracefulfailover cannot succeed because the current primary or one of the standbys is currently down.

rollingrestart will perform a graceful rolling restart of all the Pods in the StatefulSet. It will first identify every Pod which is a standby and will delete it, relying on the ReplicaController to bring it back. Once it is back, it will move on to the next Pod. Once all standby Pods have been restarted in this way, it will perform a graceful failover of the primary. It will then delete the old primary Pod, which is now a standby, and will wait for it come back up.

rollingrestart should be run every time StatefulSet spec.template.spec.image: is changed in order to perform a dolt upgrade. It can also be run in order to pick up new config.yaml settings across the cluster, for example.

Authentication

Set the environment variables DOLT_USERNAME and DOLT_PASSWORD to control the credentials the tool uses to connect to the sql-server instances. By default, it uses root with no password.

TODO

• There should be a way of manually selecting a replica to become a primary in the case of something like detected_broken_config.