Upgrading¶
Upgrading Seldon Deploy
Important
When upgrading Seldon Deploy, please, always use matching version of installation resources that can be downloaded here. This is especially important for the Seldon Deploy helm charts as these may contain updates.
Important
Seldon always recommend to first validate upgrade path in development cluster before rolling out changes to production.
Hint
Model artifacts trained to work with pre-packaged model servers may require retraining when upgrading. This is due to changes in their dependencies or in their Python versions. Advanced monitoring components – model explainers, drift detectors, outlier detectors, etc. – are the most likely to be affected.
Seldon recommends testing these artifacts in a development environment to determine whether or not they require retraining. Alternatively, you can modify the Helm values to specify previous versions of the relevant model servers.
Warning
Seldon only support upgrading to the next released version of Seldon deploy. This means that if you for example are upgrading from Seldon Deploy 1.3 to Seldon Deploy 1.5 you need to first upgrade to version 1.4 and only then to version 1.5
Hint
Keep your configuration files, e.g. Helm values of Seldon components, under VCS. This will allow you to easily rollback the installation if required.
Upgrading Process Summary¶
Before proceeding with upgrading Seldon Core and Seldon Deploy check the corresponding subsection with detailed notes for each version of Seldon Deploy, available as subpages listed in References section. Upgrading notes for Seldon Core can be found here.
The general process of upgrading Seldon product can be summarized in following steps:
Obtain installation resources corresponding to your version of Seldon Deploy from here.
Follow upgrading notes dedicated to the version you are upgrading to
Upgrade Seldon components executing executing
helm upgrade ...
commands as described on corresponding documentation pages:
Important
Always, first follow the upgrade procedure in your development cluster, validate that all your models and deployments work as expected, and only then follow with upgrading your production cluster.