11 November 2026 · 6 MIN READ
A Linux server is misbehaving: the first five things I check
A short, repeatable triage sequence for the moments when a server is slow, unreachable, or just wrong.
One: what changed?
Before touching anything: what was the last change? A deploy, a config edit, a certificate renewal, a package update. The majority of sudden misbehavior follows a change, and “nothing changed” usually means nobody knows yet.
uptime tells you how recently the box rebooted. A quick look at recent logins and command history answers whether a human was here before the alarm.
Two: the resources
Load, memory, and disk. Disk is the classic: full disks take down databases, logging, and TLS renewals at once, and they fill at the worst possible time.
uptime # load relative to core count
free -m # memory pressure
df -h # disk space
df -i # inodes (they run out too)Three: what is actually running
A process list sorted by resource answers who is consuming what, and often reveals the real story: a worker count that grew, a script that runs every minute but never finishes, or the OOM killer having made a choice for you overnight (check dmesg).
- One process at 100% CPU is different from fifty processes fighting — they have different owners and different fixes.
- A missing process matters as much as a hungry one: the service you expected to see might be the one that died.
Four: logs, targeted
Logs answer the “why” once the “what” is known. The discipline is targeting: last hour, one unit, one pattern — not an open-ended scroll through everything.
# Everything one service logged in the last hour
journalctl -u nginx --since "1 hour ago"
# Follow while reproducing
journalctl -u app -fFive: the network path
If the server is unreachable or half-reachable, separate the layers: is the process listening on the expected port, is the local firewall passing it, and does the request fail the same way from outside the host as from inside it? A curl from your laptop versus a curl from the server splits the problem in half.
ss -tlnp # what is listening, on which interface
curl -I http://localhost # inside the host
curl -I https://example.com # from outsideThen: write down what you did
A triage sequence only becomes an asset when it survives the incident. What was checked, what was found, what was changed — written into the runbook, so the next person starts from the checklist instead of from zero.
That is the principle behind the whole sequence: make the process understandable, then automate the parts that repeat.