Servers

Server health

The cards, the status badges, the six numbers, crashed apps, websites, and what every warning means.

The Servers page shows one card per server. It is designed to be readable in three seconds: a colour, a sentence, and the numbers behind it.

Servers with problems are always sorted to the top. Within each status, a server with something else to look at comes first: firing alerts, websites missing uptime, No alert configured, or a reboot needed. A line above the list reads something like "3 servers · 1 with problems · checked 4 min ago".

The status badge

BadgeMeaningWhat to do
HealthyEverything we check looks fineNothing
Needs attentionSomething will become a problem if ignored — a disk filling up, updates waitingDeal with it this week
ProblemSomething is wrong now — a service or app has stopped, a disk is nearly fullDeal with it today
Can't connectWe could not log in on the last trySee connection problems
Checking…A check is running right nowWait a few seconds
Not runningAn EC2 instance that is stopped (or starting, or stopping). Its disk, memory and services aren't checked; its state isStart it in AWS. The card picks it up at the next check, or click Check now
No SSH loginAn EC2 instance with no username and .pem saved yet, so it can't be opened or checkedClick Add SSH login on its card — see EC2 instances

Click Needs attention or Problem to see what to fix. A window lists only the checks that didn't pass, each with a heading (Disk, Inodes, Updates…), what was found and what it means. Each one has its own Ask AI chatbot button: it opens a new chat with the server, that issue and what it means filled in, asking the bot why it is happening and how to solve it. Nothing is sent until you press Send.

An EC2 instance opened through AWS Session Manager shows SSM under its status: green while the SSM agent is online and the last check got through, grey when it isn't connected.

The six numbers

Every card shows the same six, so you can compare servers at a glance. An EC2 instance's card starts with its State (running, stopped…) next to Updates, and every card has a Websites tile.

The table view packs the same facts into fewer columns:

ColumnShows
StatusThe badge, then anything that needs a look: crashed apps, websites missing uptime, stopped services, and which agents aren't set up
StateEC2 only: running, stopped…
AlertsOnly when a server has the monitoring agent: its firing alerts
Usage %Memory and disk as bars: green, then yellow, then red (disk yellow from 80%, red from 95%; memory yellow from 80%, red from 90%)
Updates / RebootUpdates waiting, and under it whether a reboot is needed

Disk

How full the fullest disk is.

  • 80% or more — the figure turns red and the server shows Needs attention. Plan to clean up.
  • 95% or more — the server shows Problem. Act now. A full disk stops databases writing, stops logs being written and can take a website offline outright.

Clicking it opens Logs, because log files are the most common cause and the easiest thing to trim safely.

Inodes are counted separately and warn at 90%. Inodes are the number of files a disk can hold, regardless of their size. Millions of tiny files — cached sessions, mail queues — can exhaust them while gigabytes remain free.

Memory

How much RAM is in use. Amber at 90%.

Linux deliberately uses spare memory for caching, so a high number is not automatically bad. It matters when it is sustained and there is no swap.

You will also see "No swap on a small server" on any machine with less than 2 GB of RAM and no swap file. That combination is how apps get killed at random under load. Adding swap is one of the guided tasks.

Load

The load average, compared against the number of processor cores. Amber at 2 per core — load 8 on a 4-core machine.

Load counts tasks waiting for the processor, so Load 3.50 on 4 cores is a busy but healthy machine, while the same number on a 1-core machine means everything is queuing.

Services

Programs that should be running. "All running" is what you want. Otherwise you get the names: "1 service stopped: nginx".

This is a Problem, not a warning — something that was meant to be running is not. Clicking it opens Services with the stopped one already selected.

One exception is certbot, which renews Let's Encrypt certificates. It is not meant to keep running: it starts twice a day, renews anything close to expiry and exits. It only shows here when its last run failed, and as a warning rather than a Problem: certbot tries again within hours, and starts renewing a month before a certificate expires. Certificate expiry itself is watched by Uptime, which reads the certificate your visitors get and emails you 14 and 3 days ahead — see Websites.

Background jobs are a warning too: cloud-init, the network wait at boot and multipathd (they stay failed until the next reboot, often over a hiccup at boot), and apt's daily runs (security updates aren't installing by themselves). They show as "1 background job failed: cloud-init".

A few Ubuntu jobs are never counted, since they fail on most cloud servers and change nothing: fwupd-refresh (firmware updates, which VMs don't have), motd-news, apt-news, esm-cache and update-notifier-* (login-banner news and Ubuntu Pro ads, which need outbound internet), and secureboot-db (VMs that don't boot with UEFI). They still show as failed on the Services tab.

Updates

Security updates waiting to be installed. Amber whenever there is at least one.

These are the fixes for publicly known ways of breaking into your server, so they matter more than the number suggests. Clicking opens Packages.

Reboot

"Reboot needed" appears after an update replaces something that only a restart can finish loading — typically the kernel. Until you restart, the old, unpatched version is still the one running.

The card shows it as its own tile; the table shows it under the update count in the Updates / Reboot column, for example "Reboot needed for 12 days". It never changes the server's status to Needs attention: a server keeps running fine on the old version, so reboot when it suits you, at a quiet moment. The Packages tab can schedule one 1–60 minutes ahead so you can warn people first.

Apps and memory

These have no tile of their own. When one finds something, the card gets an extra row under the six numbers, and the table shows it under the status. Healthy servers look exactly the same as before.

Apps

Apps run under PM2 or in Docker don't show up in Services, so they are checked separately:

  • Problem — a PM2 app that crashed and PM2 gave up restarting it, or one that restarted 3 or more times since the last check (a crash loop). A container that keeps restarting, fails its health check, or crashed and stayed down.
  • Needs attention — a PM2 app someone left stopped. Start it, or delete it if it's no longer used.

A container you stop yourself with docker stop is not counted as crashed. A crashed container stays on the card until it runs again or is removed. Clicking the row opens the PM2 or Docker tab.

Out of memory

When a server runs out of memory, Linux kills a process to free some, and PM2 or Docker often restarts it before anyone notices. The card shows each kill from the last 24 hours, for example "Out of memory in the last 24 h: killed node ×3". This is a warning. Ask the AI bot which app is growing and whether the server needs more memory or swap.

Websites

The check also reads which websites nginx serves on the server. When some of them have no uptime monitor, the card shows a Websites row, for example "2 missing uptime", and the table shows the same under the status.

Click it to see the list of those sites. Add monitor next to each opens a new monitor with its address, name and server already filled in. Add all at the top right adds a monitor for every site in the list in one go, with the usual settings: checked every minute, alerts to your default channels, linked to the server. If your plan's monitor limit is reached part-way, it stops there and says how many were added. Change any of them later on the Uptime page.

Add all always adds the https:// address, even for a site nginx serves only over HTTP, because HTTPS is often handled in front of nginx (Cloudflare, a load balancer). If a site really is HTTP only, edit its monitor on the Uptime page.

An uptime monitor checks the site from outside every few minutes and emails you when it goes down or its certificate is about to expire, which the server check alone can't do.

  • Each site is counted once. A site served over both HTTP and HTTPS is listed at its https:// address, and counts as covered once a monitor checks that address. A site served only over HTTP is listed at its http:// address, and is covered by a monitor on either address.
  • example.com and www.example.com are one site, even when nginx has a separate block for each. A monitor on either name covers it. The https:// address uses whichever name has the certificate.
  • Every other subdomain, such as api.example.com, is its own site.
  • The port is added when it isn't the usual one. You can change the address in the form, for example when a load balancer or Cloudflare handles HTTPS in front of the server.
  • Catch-all names (_), wildcards (*.example.com, .example.com), regex names, IP addresses, local names and sites that only listen on the server itself are left out. They are never listed as missing uptime.

The count is amber so you notice it, but it never changes the server's status: your plan limits how many monitors you can have, so a site without one is a suggestion, not a fault.

Agents not set up

Under the status, a server can also show one or two amber lines:

  • Monitoring agent not set up: the server has no monitoring agent, so nothing records CPU, memory, disk and network between checks, and nothing alerts you when they spike.
  • Traffic Insights agent not set up: nginx serves sites on the server, but Traffic Insights isn't installed, so there are no charts of requests, errors, slow pages or visitors. A server without nginx never shows this line.

Each line goes away by itself once that agent is installed; with both, neither shows. Click a line to see why the agent is worth having, what you get from it, and how to install it, with a button that opens the right tab on the server. Installing needs the Owner, Admin or Operator role.

Like missing uptime, these lines never change the server's status.

Monitoring alerts

A server with the monitoring agent (set up from its Monitor tab) shows its firing alerts: under the status on its card, and in the Alerts column of the table, for example 2 alerts firing in red. Click it to see each alert: what it is (CPU, memory, disk and which mount, swap), its value now and its peak, the alarm it crossed ("above 90% for 5 min"), when it started and how long it has been firing. Open in Monitoring goes to the server's charts. In the table, None firing means the agent is there and all is well; a dash means the server has no agent.

These are the Monitoring page's own alerts, not a second set of rules. When an alert clears there, it disappears here too, within about a minute. Alerts don't change the server's status.

Where alerts go

The bell icon on each card and table row opens the server's Alert notifications: send its alerts to your team's default channels, or choose any of your email, Slack and Teams channels. A server whose alerts go nowhere shows No alert configured in red under its status (or under its name in the table); click it to choose a channel. Only Owners and Admins can change it. See Notifications.

How often checks run

  • Every 15 minutes in the background.
  • Again when you open the page, if the last check is more than 5 minutes old.
  • Immediately when you press Check all now.

Checks are read-only. They log in, read numbers, and log out. They never change anything, never restart anything, and write nothing to your server.

They use the same measurements as the Monitor and Packages tabs, so the numbers always agree. PM2 apps are read as the user who runs them, never as root.

Every server with a saved login is checked: servers you added by hand, and, in an AWS project, EC2 instances whose SSH login is saved, while they are running. Hidden servers are the exception.

Hide a server

Old test boxes or EC2 instances someone else looks after can clutter the page. Press the eye icon on a card (or table row) to hide it. A hidden server:

  • disappears from Servers home and the Dashboard, for everyone in the project
  • isn't checked in the background or by Check all now
  • keeps its login, notes and history — nothing is removed

Show hidden at the top right brings hidden servers back into view, dimmed and marked Hidden. From there you can open one, check it, or press the eye icon again to unhide it. Hiding and unhiding are written to the server's Activity log.

Only Owners and Admins can hide a server.

Ask AI bot

A card with problems gets an Ask AI bot button. It opens a new chat with the server already attached and its problems written into the message — for example "Services details: 1 service stopped: nginx" — and waits for you to type what you want.

Ask it to investigate. It starts in Read mode, so it can look but not change anything until you switch it. See The AI assistant.

Notes

Anything you wrote in the server's Notes field appears in red on the card. Use it for the things someone needs to know before acting:

Hosts example.com · do not reboot 9–6 · DB backup runs 02:00

Edit them with ⋮ → Edit. Notes live in DevOps Agent only; nothing is written to your server.

"Last change" line

Under the metrics you may see "Last change 3 min ago · Vaibhav: Restarted nginx". That is the most recent entry from the Activity log, and it links to the full history. It is the fastest way to answer "did someone just touch this?".