Upgrading a DSS instance

Note

This does not apply to Mac OS X. Instance upgrade is automatic on OS X.

Prerequisites

The upgrade procedure allows you to upgrade between revisions (1.2.X) and minor versions (1.Y) of Data Science Studio. For example, you can upgrade from 1.1.2 to 1.1.3 or from 1.1.3 to 1.2.0

This procedure does not cover major version upgrades, for example from version 1.2.3 to 2.0.0. For major version upgrades, see the release note of the target major version.

In the rest of this procedure, DATA_DIR denotes the location of the Data Science Studio Data directory.

Notes and limitations

For each version of DSS, we publish Release notes, which indicate the detailed limitations, attention points and notes about release. We strongly advise that you read all release notes for the new DSS versino before starting the upgrade.

The following list of migration notes and limitations apply to all minor versions of DSS (for example from 3.0 to 3.1):

  • Due to frequent upgrades of the underlying libraries, the saved models deployed in the Flow may not be usable for scoring anymore and may need to be retrained (relaunching the train recipe)
  • For the same reason, models previously trained in analysis will generally not be deployable directly after upgrade. You will generally need to retrain them before being able to deploy them.
  • The packages deployed in an API node generally need to be regenerated on DSS and redeployed on the API node.

Some people prefer to keep an old instance running and create a new DSS instance with the new version. This is complicated and requires several migration operations (and keep in mind that each instance needs its own block of 10 consecutive TCP ports). In any case, the path would be “duplicate the instance, migrate ports and DATA_DIR, upgrade the new instance” (copying DATA_DIR between DSS instances of distinct versions is not supported).

Pre-upgrade tasks

Warning

Before upgrading, it is very highly recommended to backup the whole content of the data directory.

Stop the old version of DSS

DATA_DIR/bin/dss stop

Unpack the new software

Unpack the distribution tarball in the location you have chosen for the new installation directory.

cd SOMEDIR
tar xzf /PATH/TO/dataiku-dss-NEWVERSION.tar.gz
# This creates installation directory SOMEDIR/dataiku-dss-NEWVERSION for the new version

Perform the upgrade

dataiku-dss-NEWVERSION/installer.sh -d DATA_DIR -u

Like for normal install, DSS will check for missing system dependencies, and ask you to run a dependencies installation command with superuser privileges if needed.

DSS will ask you to confirm migration of the existing data directory

Update R installation

If R installation has been performed (see: doc:/installation/r), you must perform again the “install-R-integration” step after upgrade.

DATA_DIR/bin/dssadmin install-R-integration

Start the new version of DSS

To start DSS, run the following command:

DATA_DIR/bin/dss start