Server Status

Last updated

/jabali-admin/server-status. Live status of every watched service plus host vitals.

What is rendered

Errgroup-aggregated polling every 5 seconds. Each card returns its own status; one slow card does not block the others.

Services card

One row per watched unit: jabali-panel, jabali-agent, jabali-kratos, nginx, mariadb, redis-server and ssh always, plus jabali-stalwart, jabali-webmail, pdns, postgresql and docker when their module is on. See Services for the full list and the rules for each unit.

Each row shows:

  • The unit’s state: active, inactive, failed, activating or deactivating. An inactive unit that is not enabled at boot shows a grey idle tag.
  • Start, Restart, Reload, Stop, Enable at boot and Disable at boot buttons, depending on the unit’s state.

Host vitals

  • CPU usage and 1-minute load average
  • Memory: used / free / cache, swap usage
  • Disk: per-mount used / free, including /var/lib/mysql, /var/lib/stalwart, /home. The numbers are the ones df shows. Used is what files take and Free is what non-root can still write. Blocks the filesystem reserves for root (5% on ext4 by default) count as neither, so Used + Free is less than the size. The percent is df’s Use%: Used / (Used + Free), rounded up. A mount turns amber at 80% and red at 95%, and the same marks raise the Server Status disk alerts and the disk.full.warn / disk.full.crit notifications.
  • Network: in / out per interface

Queues card

Placeholder, not built yet. Will surface mail queue depth, backup queue depth, reconciler tick lag, and notification dispatch lag.

Recent panel requests

Top 10 panel-api requests in the last 60 seconds by latency. Drill-in shows the full route, status code, and request-id (correlate with journalctl -u jabali-panel).

Health badge

The admin header carries a lightweight Health badge backed by a cheap ?view=health projection of Server Status (single-flight + short TTL so the badge poll can’t stampede the box). It gives an at-a-glance green / degraded signal without opening the full Server Status page.

Service controls

The buttons are always on for admins. The agent runs each action and the panel logs it with the admin who ran it.

The panel refuses Stop and Disable on the units that serve the panel itself (jabali-panel, jabali-agent, jabali-kratos, mariadb, nginx, redis-server) with 403 self_destruct_blocked. The card does not show those buttons for those units. See Units you cannot stop.

Polling, not WebSocket

The page polls at 5 seconds via a single endpoint that returns the aggregated state JSON. WebSocket was considered and rejected: polling works behind every reverse-proxy and corporate-firewall combination tested, and the page does not require millisecond-scale updates.

Live-verified

The page surface was live-verified on 192.168.100.150 as the system’s primary vitals view.

  • Services — the watched units and what each button does.
  • Notifications — the service.down event sends a notification when a watched unit goes down while no one is watching the page. It checks its own list of units; see Alerts when a unit goes down.
  • Updates: running jabali update is the most common reason a service briefly disappears from this page.