UbiOps version 2.22.0¶
Client library version
Python client library version for this release: 3.14.0
CLI version for this release: 2.14.0
On the 23rd of February 2023 we have released new functionality and made improvements to our UbiOps SaaS product. An overview of the changes is given below.
Pipeline variables allow you to create fixed variables that you can use in your pipeline. To use pipeline variables, simply navigate to your pipeline editor and go to the variables tab in the left sidebar. Here you can create new variables, or drag previously created variables into your canvas. Pipeline variables support all of the data types that can also be used for input and output fields of deployments and pipelines.
Previous runs of a request schedule¶
It is now possible to easily list all the previous runs of a specific request schedule. In the WebApp you can simply navigate to your request schedule and click the previous runs tab. If you are using the API directly, the client library, or the CLI, you can pass the request schedule name to the list call. This will make sure only the requests related to that request schedule will be listed.
Improved logging page (WebApp)¶
We did an overhaul of our logging page to improve its performance and design. You can now view more logs at once, and functions like the auto loading are more performant.
UbiOps limits the resources that a project can use, such as the amount of memory and GPUs. These limits can be raised upon request. Until now, these resources were limited by the total sum that all deployment versions in a project could potentially use. Deployment versions that were not active counted towards the limit as well, which could prevent you from creating new deployment versions even if other versions were not active.
This approach has been replaced by quotas, which only limit the resources that are actively used by running deployment versions. It allows you to define more deployment versions in your project.
For example, if your project has access to 5 GPUs, you will now be able to define as many GPU versions with as many instances as you like. But only 5 GPUs will be able to run simultaneously at any given time.
Project quotas are visible via the "List quotas" endpoint on the API.
The previously deprecated field
memory_allocation has been removed from the API and the client libraries. The
instance_type field can be used instead.