Database backups¶
A DatabaseBackup runs mysqldump on a schedule, as a Kubernetes CronJob, and writes compressed dumps to a volume
you provide.
Operator 1.1.5 and earlier
Earlier operator versions looked up the CronJob through the batch/v1beta1 API, which Kubernetes 1.25 removed,
so on current clusters no CronJob was created and no backups ran. If you created a DatabaseBackup with one of
those versions, check that kubectl -n <namespace> get cronjob stroom-<name>-db-backup exists after upgrading.
Create a DatabaseBackup¶
apiVersion: stroom.gchq.github.io/v1
kind: DatabaseBackup
metadata:
name: dev
namespace: stroom
spec:
image:
repository: mysql
tag: "8.4"
schedule: "0 2 * * *" # every day at 02:00
databaseServerRef:
serverRef:
name: dev
databaseNames:
- stroom
- stats
volume:
nfs:
server: backup.example.com
path: /exports/stroom-backups
See also samples/operator-database/database-backup.yaml.
| Field | Description |
|---|---|
image |
An image with bash, mysqldump and gzip. Use the official mysql:8.4 image, which can back up MySQL 8.0 servers too. The mysql/mysql-server images have no gzip, and the mysqldump in 8.0.32 needs the RELOAD privilege, which the database user doesn't have. |
schedule |
A cron schedule. |
databaseServerRef |
A DatabaseServer (serverRef) or an external server (serverAddress and userName), as for a StroomCluster. If both are set, serverRef is used. |
databaseNames |
Databases to back up. If omitted, all databases are backed up (--all-databases). |
volume |
Any Kubernetes volume source, such as nfs or persistentVolumeClaim, where the dumps are written. |
To run a backup now, outside the schedule:
kubectl -n stroom create job --from=cronjob/stroom-dev-db-backup stroom-dev-db-backup-manual
kubectl -n stroom logs -f job/stroom-dev-db-backup-manual
What a backup contains¶
Each run connects as the database user (by default stroomuser, with the password from the database Secret) and
runs:
The file is written to <volume>/<YYYY-MM>/<backup name>_<YYYY-MM-DD_HH-MM-SS>.sql.gz and made read-only, and the
log ends with Backup successful. If mysqldump fails, the partial file is removed, the log ends with
Backup failed, and the Job fails, so failed backups show up as failed Jobs. Runs never overlap
(concurrencyPolicy: Forbid), and finished Job Pods are removed after five minutes, so check the outcome soon after
a run, or rely on your monitoring.
Things to bear in mind:
- No retention. Old backups are never deleted. Prune the backup volume yourself.
- Consistency.
--single-transactiongives a consistent snapshot without locking tables, but conflicts with schema changes. Don't upgrade Stroom, which may migrate the schema, while a backup is running. - Changes. Changes to a
DatabaseBackup, such as its schedule, are applied to its CronJob. A backup already running isn't affected. - The Secret. The dumps don't include the database Secret. Back it up separately; see Credentials.
Restoring a backup¶
A restore replaces the contents of the Stroom databases. Test the procedure before you need it.
- Stop Stroom: delete the
StroomClusterand wait for it to go. -
Load the dump into the database server. For an operator-managed
DatabaseServer: -
Apply the
StroomClusteragain.