[ac-pplx]

Summary

The nodes page now shows how full each node is.

It listed nodes with include_occupancy=false, so it never showed:

• how many sandboxes a node holds;
• how many snapshots exist only on its disk;
• whether its network slots are degraded or quarantined.

The admin API returns all of these on request. An operator choosing a node to drain, or looking for the full one, had to open each node's sandbox drawer.

Node reads stay metadata-only (include_occupancy=false), as on main: the one-second refresh, the initial load, Refresh and the reload after a node action. The per-node sandbox and snapshot counts come from a separate best-effort include_occupancy=true read, made after the page loads, on Refresh, after a node action and every 30 seconds while the page is visible. The page merges them into the rows by node ID and drops any response for an older request or another role filter. A new Sandboxes column shows:

• the active sandbox count;
• a "local snaps" tag when the node holds node-local snapshots;
• an "unhealthy slots" tag when slots are degraded or quarantined.

A tooltip gives the full slot counts, from each row's own node response, and when the sandbox counts were taken. The counts are marked "stale" when the last occupancy read failed or is older than 60 seconds, and a failed read never fails the page or an action.

Stack

#1256 (the node page) is stacked on this PR. It edits ui/src/app/dashboard/nodes/page.tsx, which #1242 and #1250 also touch; whichever merges later may need a small rebase.

Validation

• npm ci, npm test (50 pass) and npm run build in ui/.
• A Playwright loop against a local Kind stack with synthetic data (two workers with 2 and 4 sandboxes, two idle builders), on main (production build) and on this branch (the Kind space-ui image). The nodes step checks that each node shows its active count.
• No new console errors, page errors or failed requests compared with main.

Visual confirmation

Before	After

Severity and risk

Severity: medium. Capacity is the first thing to check before draining or placing work.

Risk: low. The page adds one occupancy query (per-node sandbox and snapshot counts) every 30 seconds per open nodes page, plus one on load, on Refresh and after each node action. Plain space-admin-cli list-nodes reads metadata only (include_occupancy=false). Drain ranking (list-nodes --sort drainability) and get-node --diagnostics are the other consumers of this counting query. Roll back by reverting the commit.

AI assistance: Claude Code wrote the code and this description, and ran the validation; a maintainer reviews it.

[github-actions]

Thanks for opening this pull request! A SPACE maintainer will take a look.

CONTRIBUTING.md explains how we review outside contributions.

Withdrawing this approval for now. Any feedback above still stands.

[johnlouie-pplx]

Approving with one nonblocking follow-up. At 27077d3, I checked the counts against the API contracts; the UI build passed, and mocked-API checks distinguished zero, missing and failed counts.

In ui/src/app/dashboard/nodes/page.tsx, the Sandboxes cell returns early when the sandbox count is unavailable, hiding the unhealthy-slot warning even though slot data comes from a separate read. Could we keep that warning beside "could not count" or "not computed", and say "sandbox counts are stale" so the tooltip doesn't imply that the slot sample has the same age?