Skip to content

Release v1.6.0

Latest
Latest

Choose a tag to compare

View all tags
@hippogr hippogr released this 30 Apr 03:48
· 40 commits to main since this release
v1.6
e24afaa
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

We are pleased to announce the official release of Lucia Training Platform v1.6.0!

Lucia Training Platform v1.6.0 Release Notes

This release focuses on security hardening, Docker image optimization, infrastructure upgrades, and bug fixes across the platform.

Platform Features & Bug Fixes

  • Upgraded webportal to Node.js 24 and removed the separate webportal-dind service — webportal now runs directly without Docker-in-Docker, simplifying deployment and reducing image size
  • Fixed job-detail page error handling for permission denied errors — now shows a clear message instead of infinite loading
  • Fixed job YAML and output log display issues on the webportal
  • Added support for tagging different types of GPUs
  • Skipped validation job submission for CPU nodes
  • Made Prometheus retention size configurable per service to prevent disk full issues
  • Added tool to preserve application tokens when revoking all tokens
  • Removed cronjob of abnormal-detector when stopping the service
  • Fixed exception when no name exists in filter

Docker Image Optimization

  • Reduced Docker image sizes for cluster-local-storage, copilot-chat, dashboard-data-backup, utilization-reporter, abnormal-detector, cert-expiration-checker, cluster-utilization, reverse-proxy, and model-proxy
  • Upgraded metrics-cleaner base image from Python 3.7 to 3.12-slim
  • Cleaned up job-exporter Docker image

Infrastructure & Networking

  • Updated Cilium from 1.18.6 to 1.18.9
  • Updated Go version to 1.25 across all Go-based components
  • Homebrew build for kube-scheduler and Grafana container images
  • Downgraded kube-scheduler version to match service Kubernetes version
  • Added IPoIB subnet route in init.sh to fix InfiniBand TCP connectivity on NetworkManager-managed nodes
  • Fixed DNS problem for cluster-local-storage
  • Fixed zlib 1.3.1 missing issue for pylon
  • Added Managed Identity support for build scripts
  • Made imagePullSecrets conditional to eliminate FailedToRetrieveImagePullSecret warnings
  • Removed secret deployment for image pull in favor of ACR credentials

Alert Manager & Node Management

  • Fixed KeyError when alert-parser processes validating nodes with no alerts
  • Downgraded hardware issues without Azure FaultCode to triaged_unknown to avoid broken OFR pipeline
  • Prevented node-recycler from submitting duplicate OFR tickets for the same node
  • Skipped classification for cordoned nodes with empty NodeId to prevent OFR pipeline stalling

Security

  • Updated Go toolchain and packages across all Go-based services
  • Updated Node.js packages for rest-server, alert-handler, job-status-change-notification, database-controller, and webportal
  • Updated Python packages for copilot-chat
  • Fixed S360 vulnerabilities across 13 container images including openssl, axios, follow-redirects, lodash, nodemailer, and minimatch
  • Updated go-ntlmssp to 0.1.1 for reverse proxy
  • Updated k8s-rdma-shared-dev-plugin to adapt to latest gRPC package

CI/CD

  • Updated CI workflow to filter dev-box from changed services detection
  • Removed all existing statefulsets in the system during cleanup instead of only config-defined ones
Assets 2
Loading