Skip to content

Release Notes - Version 1.19.0

Release Date: April 28, 2026

🎉 Highlights

This release delivers Okta Single Sign-On authentication, automatic AMD GPU detection in containers, and stronger resource governance controls. Users can now sign in with their Okta credentials through a secure server-side flow, with per-environment feature flag support for phased rollouts. Containers automatically detect and configure AMD ROCm hardware at startup, eliminating manual setup for GPU-accelerated workloads. Administrators gain new controls over per-team storage quotas and workload limits, and the platform has been strengthened with a Python 3.10 runtime upgrade and a Django security patch.

🚀 New features

🔐 Okta Single Sign-On

Users can now authenticate through Okta using a secure server-side authorization code exchange, improving security by keeping tokens server-side rather than exposing them in the browser. SSO can be enabled or disabled independently for each deployment environment, allowing phased rollouts across development, staging, and production.

What's included:

  • Okta SSO login using server-side code exchange (OAuth 2.0 best practices)
  • Per-environment feature flag for controlled rollout
  • Okta UAT configuration applied to the OSSCI environment

🖥️ AMD GPU detection in containers

Containers now automatically detect AMD GPUs at startup and configure the environment accordingly, so GPU-accelerated workloads run without manual setup.

What's included:

  • ROCm device detection runs automatically when a container starts
  • Plexus environment variables are persisted inside containers, making them available to all running jobs
  • Eliminates manual configuration steps for teams running AMD GPU workloads

📊 Resource governance enhancements

Administrators gain finer control over how teams consume shared platform resources, with new per-team limits on both storage and active workloads.

What's included:

  • New team_storage_size field in team-queue limits enables per-team storage quotas
  • Queue configuration now enforces a maximum number of concurrent workloads per team, ensuring fair distribution across multi-tenant deployments
  • A new is_customer_use flag on queues distinguishes customer-facing queues from internal ones

🔍 User invitation lookup

A new API endpoint (find-user-invitation) allows platform administrators to look up a user's existing invitation record directly, without additional manual lookups.

🔧 Improvements

  • Python 3.10 upgrade: The platform runtime has been upgraded from Python 3.8.13 to 3.10.20, bringing language improvements, better performance, and extended security support
  • Shared memory allocation: The /dev/shm shared memory size is now tuned to fit queue resource constraints, reducing the risk of memory-related failures in containerized workloads
  • Distributed workload fields: The num_replicas and num_nodes fields have been harmonized across the codebase for consistent naming and behaviour when specifying distributed jobs

🐛 Bug fixes

  • Fixed workload cancellation triggering unnecessary job status lookups, improving reliability under high load
  • Fixed group name filtering in the team list view on the DRF browsable API
  • Fixed the invitation expiry async task to correctly save the invitation record when it expires

🔒 Security

  • Django upgrade: Django upgraded from 4.2.29 to 4.2.30, incorporating the latest security patches

🧪 Testing

  • Migrated the test framework from the legacy nose runner to pytest and pytest-django, modernizing testing infrastructure and improving compatibility with current Python tooling

🗑️ Removals

  • Removed legacy code related to the Unipath library, reducing unused dependencies

← Back to What's New | Previous Release (v1.18.0) →