Slinky Architecture

Slinky Architecture An architecture diagram generated by Archify. Approved user access · authentication / scope · External Services Approved user access authentication / scope ArgoCD · GitOps deployment · External Services ArgoCD GitOps deployment Approved image registry · pinned Slurm / job images · External Services Approved image registry pinned Slurm / job images JupyterHub · Notebook job submission · Amazon EKS Cluster JupyterHub Notebook job submission LoginSet / login · bound user identity · Amazon EKS Cluster LoginSet / login bound user identity slurmrestd · RestApi / auth · Amazon EKS Cluster › Slurm Control Plane slurmrestd RestApi / auth slurmctld · Controller / StateSave · Amazon EKS Cluster › Slurm Control Plane slurmctld Controller / StateSave slurmdbd · Accounting / external DB · Amazon EKS Cluster › Slurm Control Plane slurmdbd Accounting / external DB NodeSet / slurmd · StatefulSet / DaemonSet · Amazon EKS Cluster NodeSet / slurmd StatefulSet / DaemonSet Shared data storage · permissions / mount / lifecycle · Amazon EKS Cluster Shared data storage permissions / mount / lifecycle Karpenter · Auto scaling · Amazon EKS Cluster Karpenter Auto scaling GPU NodePool · Provisions GPU compute nodes · Amazon EKS Cluster GPU NodePool Provisions GPU compute nodes SSH submit jobs REST API accounting dispatch jobs mount /shared GitOps deployment image pull scale GPU nodes External Services Amazon EKS Cluster Slurm Control Plane Legend Frontend Backend Database Cloud External

Actual APIs

  • • Controller / NodeSet / Accounting / LoginSet / RestApi / Token
  • • slinky.slurm.net/v1beta1 (1.2.2)

Execution and recovery

  • • slurmctld state and accounting DB cannot replace each other.
  • • NodeSet defaults to StatefulSet-style; DaemonSet mode ignores replicas.
  • • Validate draining, job completion/requeue and node reclamation.