Querying Compute Nodes
Fuzzball provides visibility into the compute nodes in your Fuzzball cluster. Nodes are physical or virtual machines that execute your workflow’s jobs. Each node provides CPU cores, memory, and specialized devices like GPUs.
You can view detailed information about a node through both the web UI and CLI. Node information includes total and available resources, device counts, jobs currently executing on each node, and the node’s state – its scheduling status, the holds recorded against it, and the job failures attributed to it.
The web UI dashboard counts each cluster’s nodes by hold state – cordoned, evicting, or
deprovisioning – so a node taken out of service is visible without opening the nodes list. A
node with several holds is counted once, under its strongest effect. Across every cluster, the
Nodes ready tile counts the nodes whose status is Ready, and the Nodes not held tile
counts the nodes with no holds. A cordoned node can be ready, so the two counts are reported
separately. For what Fuzzball observes about a node and how a
silent node is detected, see
Node Health Monitoring.