OKE RDMA Quickstart Resource Manager template v26.2.0

What's Changed

• Update readme by @OguzPastirmaci in #87
• Update path for OCI CLI in helm-deployment.tf by @OguzPastirmaci in #89
• Fix ubuntu repo by @robo-cap in #92
• Issue number: 90 - fss mount on all worker nodes by @subburamoracle in #91
• Install OKE node client packages from local repo if it exists by @OguzPastirmaci in #93
• Improve ons-webhook resiliency by @robo-cap in #94
• Add retry function to cloud init by @OguzPastirmaci in #95
• Module fixes and improvements by @robo-cap in #96
• Use NSGs instead of SLs for Lustre Service by @robo-cap in #100
• Update NPD values file by @OguzPastirmaci in #102
• Add NCCL tests manifest for BM.GPU.GB200-v3.4 and update the other manifests to use NCCL 2.29 by @OguzPastirmaci in #103
• Add Terratest tests by @OguzPastirmaci in #101
• Add the document for replacing the boot volumes of self-managed nodes by @OguzPastirmaci in #106
• Update NCCL/RCCL images by @OguzPastirmaci in #107
• Add check to wait until kubeconfig exists by @OguzPastirmaci in #108
• Add MI355 manifest and update other manifests by @OguzPastirmaci in #109
• Move GPU Fryer active health checks to Python by @OguzPastirmaci in #110
• Update BM.GPU.MI355X-v1.8.yaml by @OguzPastirmaci in #111
• added support for VM.DenseIO shapes by @shethdhvani in #114
• Update replacing node using BVR guide by @OguzPastirmaci in #115
• Fix pod logs mount by @robo-cap in #118
• Replace Nginx Ingress controller with Contour by @robo-cap in #117
• Fix: Set to retentionSize for Prometheus by @sam-andaluri in #119
• Update contour helm values by @robo-cap in #120
• Add NCCL tests manifest for BM.GPU.GB300.4 by @OguzPastirmaci in #121
• Update BM.GPU.GB300.4.yaml by @OguzPastirmaci in #122
• Add cloud-shell support to the BVR script by @robo-cap in #123
• Remove BV high storage class by @OguzPastirmaci in #126
• Add option to change services CIDR by @OguzPastirmaci in #127
• Add NCCL tests 2.29.3 images by @OguzPastirmaci in #124
• Update Node Problem Detector checks by @OguzPastirmaci in #130
• Add an option to the OKE stack to use an existing Dynamic Group by @subburamoracle in #105
• Bump chart versions by @OguzPastirmaci in #131
• Add per-pool kubernetes version, max pods, and node cycling by @OguzPastirmaci in #128
• Larger CIDR to accomodate more nodes by @OguzPastirmaci in #129
• BugFix: Fix alert webhook to reduce chances of duplicate alerts by @sam-andaluri in #133
• set kubeproxy to use ipvs & several small tweaks by @robo-cap in #132
• Increase DCGM Exporter memory limits by @OguzPastirmaci in #134

New Contributors

• @subburamoracle made their first contribution in #91
• @shethdhvani made their first contribution in #114
• @sam-andaluri made their first contribution in #119

Full Changelog: v25.11.0...v26.2.0

OKE RDMA Quickstart Resource Manager template v25.11.0

• Add option to install OCIR credential helper
• Fix for Metrics Server
• Add support to use image URIs

Full Changelog: v25.10.0...v25.11.0

OKE RDMA Quickstart Resource Manager template v25.10.0

• Kubernetes upgrade: Added support for Kubernetes v1.34
• Documentation: New guide — Deploying Prometheus & Grafana Stack with Dashboards and Alerts manually
• Health checks:
  • Added RCCL tests
  • Added RocM Validation Suite (RVS) gst_single for AMD validation
• Grafana access link: Default domain updated to endpoint.oci-hpc.ai, configurable for custom domains
• Component updates: Refreshed dependencies and minor fixes across the stack

Full Changelog: v25.9.0...v25.10.0

OKE RDMA Quickstart Resource Manager template v25.9.0

• Option to provision a shared Lustre file system and a PV backed by the Lustre file system
• Fully private clusters using Resource Manager Private Endpoint for deployment
• Same dashboards and notifications with the Slurm stack
• Option to use Oracle Linux for non-RDMA pools
• Component updates

OKE RDMA Quickstart Resource Manager template v25.5.1

This is a hotfix release to fix the breaking Helm provider change.

More info about the change here: hashicorp/terraform-provider-helm#1637

OKE RDMA Quickstart Resource Manager template v25.5.0

• Added AMD Device Metrics Exporter
• Added AMD dashboards

OKE RDMA Quickstart Resource Manager template v25.4.0

• Added Kubernetes v1.32
• Changed the default number of maximum pods per node to 110

OKE RDMA Quickstart Resource Manager template v25.3.1

• OKE AMD GPU device plugin is enabled for BM.GPU.MI300X.8 shape
• OKE DCGM Exporter is disabled (upstream DCGM Exporter is deployed)
• Helm fix for Grafana load balancer not being deleted properly on Terraform destroy
• Updated the health checks for Node Problem Detector
• Updated Grafana dashboards
• Added the required policies for Oracle Cloud Agent GPU/RDMA monitoring

OKE RDMA Quickstart Resource Manager template v25.3.0

• VCN-native pod networking is now the default option for pod networking instead of Flannel.
• Node Problem Detector is now deployed part of the stack and integrated with the Prometheus/Grafana stack for alerting.
• Switched to using the upstream OKE Terraform module.

OKE RDMA Quickstart Resource Manager template v25.3.0-beta

• VCN-native pod networking is now the default option for pod networking instead of Flannel.
• Node Problem Detector is now deployed part of the stack.
• Fixed a Node Exporter issue preventing metrics from being streamed from bare metal GPU nodes.