Skip to content
Hyperion

Read server and site statistics

The Stats page answers two questions: how hard is each node working, and what does each site cost it. Use it when a server feels slow, a disk fills up, or you want to find the site behind a spike.

Before you start

Section titled “Before you start”
  • You are an Owner or an Administrator. The page covers the whole cluster, so roles limited to their own sites do not see it.

Steps

Section titled “Steps”
  1. Open Stats in the sidebar, under Overview.
  2. With worker nodes, choose a Node at the top: All nodes, master (this node) or a worker. The page title shows the choice. Choose Refresh to set how often the page updates itself: 5 s, 10 s (the default), 15 s, 30 s, 1 min, 5 min or Paused.
  3. Under Nodes, read one row per node: CPU, Memory, Disk, Network in / out and Sites. A dot shows whether the node’s agent is online. Click a node to open its section.
  4. Read the Health group for the selected node:
Tile What it shows
CPU Busy share over the last minute, the number of cores, and “stalled N%” when tasks waited for CPU
Load · 1 min The one-minute load average, with average and peak over the chart window
Memory Used share and amounts, excluding reclaimable cache, plus swap in use
Network Current rate in and out, with the peak of each
Disk Share of the node’s volume used, with the part held by site files and the part held by backups

Sparklines fill in over time. A tile reads “History fills in after the next 5 min sample” until enough samples exist, and the network chart needs a few seconds of samples. The CPU, Load · 1 min, Memory and Network charts are drawn from samples taken every 15 seconds and cover the last hour, so they move with the figures above them. Hover a chart to read the value and the time (UTC) of the nearest point. A node that has not yet taken four such samples, such as one just updated, or an older node, draws the 5-minute history instead, which covers 4 hours.

  1. Read Where the CPU goes · last minute. The bar splits CPU into Sites (PHP and cron of every site, the total of the table below), Database (MariaDB and Redis, shared by all sites and not part of any site’s figure), Web server (nginx and the PHP-FPM master process), Panel (Hyperion itself), Other (the operating system and everything else, backups included) and Idle. Use it when the sites add up to 8% and the node says 40%.
  2. Read the strip under it: Uptime, Pressure cpu/mem/io (the share of time tasks stalled on each resource), OOM kills · 24 h and Agent (its version). An out-of-memory kill is a process the kernel forcibly ended, and the site may have returned 500 errors briefly. The strip shows when the last one happened.
  3. Read the Sites group: Hostings (active, suspended, failed), Site files, Backups on disk, Traffic out · 24 h and Requests · 24 h. Neither disk figure on the page counts backups, because they sit on the node volume outside every site’s home. On a filling server they are often the reason.
  4. In Per-site usage, pick the period with Live, 30 min, 6 h, 12 h, 24 h or 2 days, and click a column heading to sort by Site, CPU, CPU peak, RAM, Disk, Traffic or Requests. The table lists active sites only. Changing the period, the sort or the node updates the page in place and keeps the address, your scroll position and the browser’s back button.
    • Live shows CPU and RAM over the last minute, and traffic and requests over the last 5 minutes.
    • The other periods show the average and the peak over that period. A note appears when history only reaches back part of the way.
    • CPU is a share of everything the node can do: 100% means all its cores are busy.
  5. If Backups by site appears, read how much each site’s backups take On server and Off-site. Files that no backup record names are counted and called out, such as the archives of a deleted site.

Sampling: per-site CPU comes from a sampler that reads /proc every few seconds. The charts are sampled every 15 seconds, the longer history behind them every 5 minutes, and traffic and requests come from the nginx access logs.

Check that it worked

Section titled “Check that it worked”
  • The Health group says “read <time> ago” and the node’s dot is green. “agent offline” in red means the numbers are stale.
  • The sites in Per-site usage match Hostings under Sites.

If it fails

Section titled “If it fails”
  • “No stats yet”: no sample has been taken. Wait for the next 5-minute sample.
  • “No active sites here yet”: the selected node serves no active site.
  • A node’s tiles show “no sample yet”: the node’s agent has not reported. Check Service health.
  • “Incomplete view.” at the top: a node’s answer failed authentication and was discarded, so its numbers are missing.
Section titled “Related”