Database

The Blueprints Processor stores CBAs, model types, resource dictionaries, resolution results, configuration snapshots and workflow audit records in a PostgreSQL database (tested with PostgreSQL 13, the version deployed by OOM, and 17). MariaDB is no longer supported as the CDS database.

The schema is created and updated by Hibernate at start-up (blueprintsprocessor.db.hibernateHbm2ddlAuto=update); there are no separate SQL scripts to run.

Configuration

Property

Value

blueprintsprocessor.db.url

jdbc:postgresql://<host>:5432/<database>

blueprintsprocessor.db.username

database user

blueprintsprocessor.db.password

database password

blueprintsprocessor.db.driverClassName

org.postgresql.Driver

blueprintsprocessor.db.hibernateHbm2ddlAuto

update

Hibernate detects the dialect from the database connection. Set blueprintsprocessor.db.hibernateDialect only to override it.

processor-db and CBA database connections

The processor-db resource source and database connections defined inside a CBA can still target other databases. Supported type values are maria-db, mysql-db, mssql and postgres-db. In OOM, processor-db points at the SDNC MariaDB by default.

Upgrading from a MariaDB-based release

The new release does not migrate data by itself, and the OOM cds chart no longer deploys the CDS MariaDB: upgrading the chart removes it. Take a dump before upgrading, then choose one of the following.

Option A: re-onboard the CBAs. Upload or publish the CBAs again after the upgrade. Stored resource resolutions, configuration snapshots and workflow audit records are lost.

Option B: copy the data with pgloader. This keeps everything.

  1. Before upgrading, dump the CDS database. With the default chart values the MariaDB pod is cds-db-0 and the password of the sdnctl user is in the <release>-cds-db-secret secret:

    PW=$(kubectl get secret -n onap <release>-cds-db-secret -o jsonpath='{.data.password}' | base64 -d)
    kubectl exec -n onap cds-db-0 -c mariadb -- \
      mariadb-dump -usdnctl -p"$PW" --single-transaction --no-tablespaces sdnctl > cds-dump.sql
    
  2. Upgrade the cds chart. Wait until the Blueprints Processor is ready: it creates the tables in PostgreSQL and loads the initial model types and resource dictionaries. Then scale it to 0.

  3. Import the dump into a temporary MariaDB that pgloader can reach, for example with Docker on the machine running pgloader:

    docker network create cds-migration
    docker run -d --name cds-mariadb --network cds-migration -e MARIADB_ROOT_PASSWORD=<password> \
      -e MARIADB_DATABASE=sdnctl -e MARIADB_USER=sdnctl -e MARIADB_PASSWORD=<password> mariadb:11
    docker exec -i cds-mariadb mariadb -usdnctl -p<password> sdnctl < cds-dump.sql
    
  4. Make the PostgreSQL primary reachable, for example with kubectl port-forward -n onap svc/cds-pg-primary 5432:5432, and save the following command file as cds.load. Run pgloader as the CDS database user; no superuser is needed.

    LOAD DATABASE
         FROM mysql://sdnctl:<password>@<mariadb-host>/sdnctl
         INTO postgresql://sdnctl:<password>@<postgres-host>/sdnctl
    WITH create no tables, data only, truncate
    CAST type datetime to timestamp drop default using zero-dates-to-null
    INCLUDING ONLY TABLE NAMES MATCHING 'BLUEPRINT_MODEL', 'BLUEPRINT_MODEL_CONTENT',
         'BLUEPRINT_WORKFLOW_AUDIT_STATUS', 'ERROR_DOMAINS', 'ERROR_DOMAINS_ERROR_MESSAGES',
         'ERROR_MESSAGES', 'MODEL_TYPE', 'RESOURCE_CONFIG_SNAPSHOT', 'RESOURCE_DICTIONARY',
         'RESOURCE_RESOLUTION', 'TEMPLATE_RESOLUTION';
    

    truncate replaces the initial data loaded at start-up with the migrated rows. pgloader writes into the PostgreSQL schema named after the MySQL database, sdnctl, which is where the Blueprints Processor creates its tables on the OOM PostgreSQL. If your PostgreSQL has the CDS tables in the public schema instead, add this line before CAST:

    ALTER SCHEMA 'sdnctl' RENAME TO 'public'
    

    Run it, for example with docker run --rm --network cds-migration -v $PWD/cds.load:/cds.load dimitri/pgloader pgloader /cds.load. pgloader lower-cases the MariaDB table and column names, which matches the names the Blueprints Processor creates on PostgreSQL.

  5. pgloader does not move the identity sequence of the workflow audit table. Reset it as the CDS database user:

    SELECT setval(pg_get_serial_sequence('blueprint_workflow_audit_status', 'workflow_audit_id'),
                  coalesce(max(workflow_audit_id), 0) + 1, false)
      FROM blueprint_workflow_audit_status;
    
  6. Compare the row count of every table listed above in both databases, scale the Blueprints Processor back up and check that the CBAs are listed by GET /api/v1/blueprint-model. Remove the temporary MariaDB afterwards.