SC26AI agents lab + dinner on the river · Nov 17
Parallel Works
View as Markdown
Share:

Use a Managed Cluster

Use this guide to open terminals and sessions, run workflows, submit scheduler jobs, and understand the managed-cluster status shown in ACTIVATE.

Before connecting, you need an ACTIVATE account, permission to use the cluster, and a Linux account recognized by the nodes you will use. Your administrator can configure the agent to create Linux accounts automatically or provide them through the site's existing account system. See How Access Works for the account requirements and connection methods.

Open a Terminal

Select Open Terminal from the cluster page to connect to the configured login address. The SSH username is your ACTIVATE username.

From a local terminal with the authenticated ACTIVATE CLI, run:

pw ssh <cluster-name>

The explicit managed-cluster form is also supported:

pw ssh pw://<cluster-name>

See pw ssh for port forwarding and ProxyCommand examples.

For a registered login node, ACTIVATE uses the agent tunnel when connected and may attempt direct SSH to the node's hostname when the tunnel is unavailable. Your administrator configures the login address.

Start a Desktop or VS Code Session

An organization administrator must first select a tunnel-connected Session Node. It can differ from the login node.

Use Quick Launch on the cluster's Details tab, or create a session from Sessions. A session remains assigned to the node selected when it was created even if the cluster's Session Node setting changes later.

  • Follow Desktop Sessions for a browser-based graphical desktop on the selected session node.
  • Follow VS Code Sessions for browser VS Code or a local VS Code Remote-SSH connection.
  • To run a desktop or VS Code session on a scheduler-allocated compute node instead of the session node, choose one of the cluster's environments as the target.

Starting a session requires a connected agent tunnel on the session node. If a session stays pending, check that node's tunnel status or contact the cluster administrator.

Run Workflows and Submit Jobs

Users with cluster access can select the managed cluster as a workflow compute resource. See Running Workflows.

You can also open a terminal and submit jobs using the site's scheduler commands. For Slurm examples, see Submitting Jobs via Slurm. Follow your Linux administrator's instructions for partitions, queues, accounts, storage, and software modules.

Understand What ACTIVATE Displays

  • Details summarizes online nodes, CPU, memory, storage, current and historical utilization, and active sessions. It also displays partitions for a detected Slurm controller.
  • Jobs lists active Slurm or PBS jobs. Slurm jobs can be filtered by partition.
  • Nodes shows each registered node's status, type, operating system, agent version, tunnel state, and last heartbeat.
  • A node detail page shows CPU, memory, disk, load, uptime, process count, network totals, IP addresses, filesystems, and metric history.

Scheduler data comes from the registered node running slurmctld or pbs_server.

Access Problems

If you cannot sign in:

  1. Confirm you have Login, Sudo, or Admin access to the cluster.
  2. Confirm your ACTIVATE username is the Linux username expected by the site.
  3. If you connect with a personal SSH key and the site uses ACTIVATE key lookup, confirm its public key is listed under Account > Authentication > SSH Keys. Platform and managed workspace connections can use platform-managed keys.
  4. Tell the administrator whether browser terminal, pw ssh, or both fail and whether the selected login node appears online.

Administrators can continue with operations and troubleshooting.