Skip to content

Operator-managed DatabaseServer

A DatabaseServer is a single-instance MySQL 8 server that the operator runs for one StroomCluster. The operator generates its passwords and creates the Stroom databases. It suits evaluation and smaller deployments. For production, consider an external database with replication and managed backups: a DatabaseServer has one replica and no failover.

Create a DatabaseServer

apiVersion: stroom.gchq.github.io/v1
kind: DatabaseServer
metadata:
  name: dev
  namespace: stroom
spec:
  image:
    repository: mysql
    tag: "8.4"
  databaseNames:
    - stroom
    - stats
  additionalConfig:
    - max_connections=1000
  resources:
    requests:
      cpu: "1"
      memory: 4Gi
    limits:
      memory: 4Gi
  volumeClaim:
    storageClassName: fast-ssd
    accessModes:
      - ReadWriteOnce
    resources:
      requests:
        storage: 100Gi

See also samples/operator-database/database-server.yaml.

Field Description
image A MySQL 8 server image. The operator relies on the image's standard entrypoint (MYSQL_USER, MYSQL_PASSWORD and /docker-entrypoint-initdb.d).
databaseNames Databases to create on first start. Include the StroomCluster's appDatabaseName and statsDatabaseName (by default stroom and stats).
additionalConfig Lines appended to the [mysqld] section of /etc/my.cnf, each a single option such as max_connections=500. Options the operator sets (datadir, port, user), and options that give access to the server's files or bypass authentication (such as secure_file_priv, local_infile, init_file and skip_grant_tables), are rejected.
resources Container requests and limits.
volumeClaim PersistentVolumeClaim spec for the data directory. Use fast, low-latency storage.
readinessProbeTimings, livenessProbeTimings Probe timings. If omitted, Kubernetes' probe defaults apply.
podSecurityContext, securityContext, nodeSelector, tolerations, affinity Pod settings.
podAnnotations Annotations for the database Pod.

Connection limit

MySQL allows 151 connections by default. Each Stroom node opens connection pools to both databases, and the operator also connects, so a cluster of more than a few nodes can exceed the default. Set max_connections in additionalConfig, with headroom.

Deploy and check

kubectl apply -f database-server.yaml
kubectl -n stroom get databaseserver dev
kubectl -n stroom logs stroom-dev-db-0 -f

The State column shows Deployed once the operator has created the database objects. The Pod is stroom-<name>-db-0 and the server is reachable in the cluster at stroom-<name>-db-headless.<namespace>.svc.cluster.local:3306.

Connect a StroomCluster

spec:
  databaseServerRef:
    serverRef:
      name: dev
      # namespace: stroom   # defaults to the StroomCluster's namespace

A DatabaseServer can serve only one StroomCluster. The first cluster to reference it claims it. A second cluster referencing it fails to reconcile, with the error DatabaseServer ... already claimed by StroomCluster ... in the operator log.

While a StroomCluster uses it, the DatabaseServer can't be deleted: the deletion waits until the cluster has gone.

Operator 1.1.5 and earlier

Earlier versions recorded the claim using the DatabaseServer's own name rather than the cluster's, so a StroomCluster whose DatabaseServer had a different name failed with already claimed. The operator now replaces a claim by a StroomCluster that doesn't exist, which repairs these claims when you upgrade.

Credentials

The operator generates a Secret, stroom-<name>-db, with passwords under the keys root and stroomuser. Stroom and the operator connect as stroomuser, which is granted all privileges on each of databaseNames.

Back up this Secret

The Secret isn't deleted with the DatabaseServer, but if it's lost the data can't be reached with the generated credentials. Keep a copy somewhere secure:

kubectl -n stroom get secret stroom-dev-db -o yaml > stroom-dev-db-secret.yaml

Changing a DatabaseServer

Changes to a DatabaseServer's image, resources, probe timings, Pod settings and additionalConfig are applied to the running server. Changing anything in its Pod restarts MySQL, so Stroom loses its database connection briefly; plan changes for a quiet period. A change to additionalConfig alone updates the configuration, which MySQL reads when it next starts.

While the server's Pod is being replaced, and until it's ready again, the operator doesn't create or change the nodes of the StroomCluster using it, as a node stopping then couldn't drain. If you change both at once, or an operator upgrade changes both, MySQL restarts first, then the nodes are rolled. The operator logs Waiting for the DatabaseServer before creating or updating the nodes while it waits. A new StroomCluster's nodes are likewise created once its DatabaseServer is ready.

Some changes need more:

  • volumeClaim. Kubernetes doesn't allow a StatefulSet's volume claim template to change, so the operator keeps the original. To change it, delete the DatabaseServer (its PersistentVolumeClaim data-stroom-<name>-db-0 and Secret are kept), resize or replace the volume claim yourself, and apply the DatabaseServer again.
  • databaseNames. The databases are created when MySQL first initialises its data directory. Adding a name later doesn't create the database; create it, and grant stroomuser access to it, yourself.
  • The passwords. The generated Secret is never changed by the operator.

Take a backup before changing the image. Upgrading MySQL to a new major or minor version also needs the usual MySQL upgrade checks.

Operator 1.1.5 and earlier

Earlier versions created these resources once and didn't update them afterwards. After upgrading the operator, the database server's Pod is updated to match the DatabaseServer, which restarts MySQL once.

Deleting a DatabaseServer

kubectl -n stroom delete databaseserver dev

If a StroomCluster still uses it, the command waits (press Ctrl+C to stop waiting; the deletion continues in the background) until that cluster is deleted. The data volume claim and the Secret are kept. Delete them yourself if you no longer need the data.