Upgrade notes

Substra 0.34.0

This version upgrades the integrated PostgreSQL databases from versions 14 to 16. This upgrade must be performed manually.

If you do not use the integrated databases, then you have nothing to do.

Otherwise, you must update the data format. There are many ways to do this; here we give you a simple way which is acceptable if:

Have small amounts of data

Have access to the clusters

Downtime is acceptable

Warning

Before going further, make sure no operations are in progress and cordon the node.

Gather required info:
- database credentials (here $PG_USER and an associated password)
- Kubernetes namespace (here $NAMESPACE)
- Helm release (hereafter $RELEASE_NAME, can be obtained from helm get release -A)
- hostname of the Postgres service (hereafter $HOST, can be obtained from kubectl get svc -n $NAMESPACE)
- name of the database in Postgres: by default orchestrator for the orchestrator, and substra for the backend (hereafter $DB)

Deploy a psql client (we will call it postgres-backup) in the namespace

kubectl apply -f -n $NAMESPACE - << 'EOF'
apiVersion: v1
kind: Pod
metadata:
  name: postgres-backup
spec:
  containers:
    - name: postgres
      image: postgres
      command: ["sleep", "infinity"]
EOF

Launch a shell from within the postgres-backup pod and dump the DB:

pg_dump $DB --host=$HOST -U $PG_USER -f /dump.sql --clean --create

(OPTIONAL) Retrieve the dump:

kubectl --retries 10 cp $NAMESPACE/postgres-backup:/dump.sql dump.sql
sha1sum dump.sql # check it matches with the source (use `shasum` on Mac OS)

Delete the Postgres StatefulSet and PVC

This depends on your particular set-up but it should look like this:

kubectl delete -n substra statefulset $RELEASE_NAME-postgresql
kubectl delete -n substra pvc data-$RELEASE_NAME-postgresql-0

Perform the database upgrade

Note versions 8.0.0 and 23.0.0 are used: they contain the database upgrade. The apps upgrade are done in 8.0.1 and 23.0.2.

You can get values with helm get values

Orchestrator:

helm upgrade -n $NAMESPACE $RELEASE_NAME https://github.com/Substra/charts/raw/main/orchestrator-8.0.0.tgz --values orc-values.yaml

Backend:

helm upgrade -n $NAMESPACE $RELEASE_NAME https://github.com/Substra/charts/raw/main/substra-backend-23.0.0.tgz --values backend-values.yaml

Delete the applicative deployments and statefulset to avoid them polluting the database (orchestrator-server, backend-server, backend-worker, …)
Launch a shell from within the postgres-backup pod and load the dump:
```
psql --host=$HOST -U $DB_USER < /dump.sql
```
Perform final upgrade as normal

Warning

It must be repeated on each Substra Backend and on the Orchestrator (if standalone).

Substra 0.28.0

This version now allows external database connections, and database setup info and connection info are no longer the same setting.

If you changed some database settings such as credentials in the orchestrator or backend values, like this:

postgresql:
  auth:
   username: my-username
   password: my-password
   database: my-substra-db

Then you’ll need to copy them over to a new database key:

postgresql:
  auth:
   username: my-username
   password: my-password
   database: my-substra-db

database:
  auth:
   username: my-username
   password: my-password
   database: my-substra-db
   # you could also use YAML anchors for this

Substra 0.23.1

This version ships Redis with persistence (AOF) activated. As this component is used as a message broker and not as a cache, the previous redis deployment can be removed before lauching the upgrade.

No task should be running on the clusters, then, for each cluster where substra-backend is deployed, run the following command before upgrading: - kubectl delete statefulsets BACKEND_NAME-redis-master -n NS_NAME, where:

BACKEND_NAME is the release name as defined in Helm

NS_NAME the namespace name where your pods are deployed