Fleet docs v0.2.0-rc.448

Machines

Machine health has one place: the Machines page at /machines. It has one card for each box in the fleet: the control server and each node. The dashboard's health strip and its folded Boxes section link there and do not repeat the detail. The Boxes section keeps only the monitor alerts.

How to read a card

Part Shows Look for
Name and badges Hostname, server or node, elastic or fixed, the fleet release, the Claude Code version
CPU Percent of all cores busy since the last sample, and the number of cores Amber at 90 %, red at 98 %
load 1, 5 and 15 minute load Amber above the number of cores, red above twice that
Sparkline CPU busy, the last 15 minutes
RAM Used (total less available) of total Amber at 88 %, red at 95 %
RAM dot Memory pressure: the share of time tasks waited for memory (/proc/pressure/memory, 10 s) Green below 5 %, amber below 20 %, red above. Grey: the kernel has no pressure data
Disks One bar for each mount that holds /, the node's state dir or the projects root Amber at 85 %, red at 95 %
turns Running turns of the node's max_running
  • Click a card to see the agents on that box and their process trees.
  • The page updates live. A box that has sent nothing for 30 s turns grey with "last seen … ago".
  • Only users see the page. Agent and node tokens get a 403.

When it looks wrong

You see It means Do this
A grey card, "last seen …" No heartbeat from that node for 30 s Check the node's service and its network to the server. Look at the workers for lost agents.
A node is missing It has not joined, or it runs a release from before the machine view Upgrade the node
"reconnecting" in the header The live stream dropped It retries on its own. If it stays, reload the page.
Memory is red and turns fail The box has too little memory for its turns Lower max_running on that node, or add a node