Troubleshooting a slow or failing Linux server
A read-only first-response checklist for finding service, disk, memory, and network problems on Linux.

A slow or failing server can have several causes. The disk may be full, a service may have stopped, memory may be tight, or a network path may have failed. Start by collecting evidence. Do not change several settings at once, because that can hide the cause.
This guide uses common Linux tools. Some commands or output fields vary by distribution. htop is optional and may need installation; use top if it is not available. Run read-only checks first. Use sudo only when a command needs elevated access.
Record the symptoms first
Write down what users see and when it started. Note the affected service, URL or port, recent deployments, package updates, configuration changes, and whether the problem affects one user or everyone. Record the server's timezone if you need to compare logs with reports from another system.
Do not restart services, clear logs, delete files, or change resource limits before capturing the current state. If the server is unstable, collect the most useful evidence first and save it somewhere with enough free space. Redact passwords, tokens, private IPs, and personal data before sharing output.
Check the service with systemd
Find the service name, then inspect its state:
systemctl status nginx
systemctl is-active nginx
Replace nginx with the affected unit. status shows recent service information and some journal output. It may not show the full failure history.
Read the unit's recent logs:
journalctl -u nginx --since "30 minutes ago" --no-pager
For a broader view of recent system messages, use:
journalctl --since "30 minutes ago" --no-pager
Look for the first error near the time the problem began. Repeated later errors may be consequences of the first failure. If the service failed during boot, review the current boot's messages with journalctl -b.
Do not immediately run systemctl restart. First save the status and relevant logs. A restart can restore service, but it can also remove useful process state or cause a second outage. If you do restart after collecting evidence, note the exact time and result.
Check disk space and inodes
A full filesystem can stop applications from writing logs, temporary files, or database data. Check mounted filesystems:
df -hT
df -i
df -hT reports space and filesystem type. df -i reports inode use. A filesystem can run out of inodes even when it still has free space.
If a mount is nearly full, identify the directory responsible before removing anything. Use a targeted command such as du -xhd1 /var if your version of du supports it. The -x option keeps the scan on one filesystem. Do not delete database files, container volumes, or logs just because they look large. Find out what owns them and whether a service is still writing to them.
Check the system journal size with journalctl --disk-usage. Do not vacuum logs during an active incident before saving the messages you need. If space is critical, use your distribution's documented cleanup process and confirm the files are safe to remove.
Check memory and process load
Start with a summary:
free -h
Linux uses spare memory for caches. A low free number alone does not prove memory exhaustion. Check available memory, swap use, and whether the system reports out-of-memory events in the journal.
Use top to see active processes and system load:
top
Press q to exit. If htop is already installed, it provides an interactive view, but it is not present on every server. Record the process name, CPU use, memory use, and time observed before taking action. High load can come from CPU work or from tasks waiting on storage or other resources, so do not change CPU or memory limits based on one snapshot.
Check recent kernel messages for memory pressure or process termination:
journalctl -k --since "30 minutes ago" --no-pager
Look for messages about out-of-memory kills, device errors, or repeated driver warnings. Keep the surrounding timestamps and context.
Check network state without changing it
First check the interface addresses and routes:
ip address
ip route
Then check whether the expected service is listening:
ss -lntup
The process details may require sudo. Compare the listening address and port with the service configuration. A service bound only to 127.0.0.1, for example, is not listening on every network interface. Do not change the bind address until you know the intended exposure and firewall rules.
Test name resolution and the remote endpoint separately:
getent hosts example.com
curl -I --max-time 10 https://example.com/
Replace example.com with a hostname you are authorized to test. A failed curl can result from DNS, routing, TLS, a proxy, or the application itself. Test the server's gateway or a known endpoint only when appropriate for your network. ping can help check basic reachability, but many hosts block it, so no ping response does not prove that a service is down.
Avoid using commands that alter routes, interfaces, firewall rules, or DNS settings during initial diagnosis. First capture the current output and confirm whether the issue is local to the server or on the path to it.
Compare the evidence
Use timestamps to line up service logs, system messages, and user reports. Ask:
Did one service fail, or are several services affected?
Did the problem begin after a deployment, reboot, or configuration change?
Is a filesystem full, or are inodes exhausted?
Is memory available, and did the kernel kill a process?
Is the service listening on the expected address and port?
Can the server resolve the hostname and reach the intended endpoint?
A single command rarely identifies the cause. Compare observations before you change settings.
Incident checklist
Record the symptom, start time, affected users, and affected service.
Save
systemctl statusand relevantjournalctloutput.Check
df -hT,df -i, andfree -h.Capture
toporhtopoutput and relevant kernel messages.Capture
ip address,ip route, andss -lntupoutput.Test DNS and the target endpoint separately.
Note every change, command, timestamp, and result.
Make one controlled change at a time, then repeat the same checks.
If the server hosts critical data or the evidence suggests disk or filesystem damage, stop before running repair or cleanup commands. Preserve the current state and use the system or application recovery procedure. Some repair actions can make data loss worse.
What to collect before asking for help
Provide the distribution and version, the affected unit or application, the time the issue began, and the exact command output that relates to the failure. Include relevant logs with timestamps, disk and inode use, memory and swap use, and network address and route details. Redact secrets and public-facing addresses if they should not be shared.
Keep the evidence from before and after any change. That makes it easier to see what changed and gives the next person a reliable starting point.
Join the discussion
Have a thought or a correction? Send it over — your message opens in your email app.