Skip to main content

​
Containers

Containers lists every container that is up this second: the task or box it runs, its app and function, how long it has been alive, and its worker. Open one for its live output, which streams straight from the container (the same lines that reach your terminal during python app.py), and a Stop button. Sandboxes and notebooks are containers too, so they appear here. A container disappears from this list when it exits; its logs stay with the task under Apps.

​
Workers

Workers shows the worker processes, one per node and pool: the pool name (default, build, l40s), the node’s CPU, memory and GPUs, and how much of each is in use by containers right now. A cordoned or draining worker is marked. On a Spot GPU pool, a node that is reclaimed disappears here and its tasks are retried on another worker. levers9 worker list, cordon, uncordon and drain do the same from the CLI; levers9 machine list shows the hardware inventory and on-demand offers.

​
Workspace metrics

The Workspace metrics button in the top bar opens the totals across workers: CPU, memory and GPUs in use, containers and tasks per minute, with 24 hours of history. The landing page’s live section reads the same series.

​
How many workers you can get right now

Each pool chip on the Workers page says how many more workers the pool can start at this moment, with the reason on hover. The scheduler holds a request (it never fails it) when a pool already has its cap of workers pending, when workers were added faster than its per-minute budget, or, on an attached cluster, when every node already has a worker waiting to start. levers9 pool list shows the same as the Workers now column. The limits are worker.provisioningGuard in the controlplane config: 20 pending and 30 additions a minute per pool by default.