Check a node's load, swap and busiest processes
CPU and memory percentages don't always explain a slow node. A node can sit at modest CPU while work queues for it, or swap to disk while memory looks fine. The Metrics card on a node's Overview tab shows the figures that do: load against the cores, swap in use, how much work waited, and the processes using the most. Only a node that reports these figures shows them. A node that does not report them shows neither the Signals nor the Processes section.
Who can do this: anyone who can view nodes.
Read the signals
- In the sidebar, select Directory, then the Machines tab.
- Select the node's name. Its details open on the Overview tab.
- Find the Metrics card. Signals and Processes sit under its latest figures and always show the node's last report. The range list at the card's top right (1h, 6h, 24h or 7d) changes only the charts below them.
- Under Signals, read Load · N cores, where N is the node's core count. It gives three numbers: the average over the last 1, 5 and 15 minutes. A load well above the core count means work is waiting for the processor.
- Read Swap used, such as 1.2 GB of 8.0 GB. None means the machine has no swap. Swap that keeps growing means memory is short, even when the memory percentage looks fine.
- Read CPU pressure, Memory pressure and IO pressure. Each is the share of the last minute in which some work was held up waiting for that resource. Near 0% is healthy. A high IO pressure points to a slow or busy disk.
- Under Processes, read the busiest processes, highest CPU first, then highest RAM. Each row gives the process Name, the User it runs as, its CPU and its RAM. CPU is a share of one core, so a process using four cores reads 400%.
Only the process name and user are shown, never the command line, so no secrets leak through this list. A figure the node doesn't report is left out rather than shown as zero.
To find which task's session is behind the load, and what to do about it, see Find what is overloading a node. To judge from a week of history whether the machine is the right size, see Check whether a node is big enough.