Swap
A VPS with 1–2 GB of RAM usually comes without swap. As long as memory suffices, you don’t notice. When it
doesn’t — an image build, apt, a database spike — the kernel calls the OOM killer, which kills the biggest
process. Most often that’s the database or Docker itself. With swap, the kernel moves rarely used pages to disk,
and the server gets through the spike slower but alive.
| With | How it works | When to use |
|---|---|---|
| a swap file on disk | memory pages go to /swapfile |
almost always: created in a minute, resized without repartitioning |
| zram | compressed swap in memory itself, about 2–3 to 1 (zram-tools) | a slow disk or one with a write limit; 1 GB of RAM or less |
| no swap | memory is plentiful, or every container has a limit | 8 GB of RAM or more, a known load |
The size is 2 GB with 1–4 GB of RAM. A server doesn’t need more: if it lives in swap all the time, it’s short on memory, and swap only drags that out.
Create the swap file
Check there’s no swap
swapon --show # empty — no swap free -h # total used free shared buff/cache available # Mem: 1.9Gi 612Mi 230Mi 4.0Mi 1.2Gi 1.2Gi # Swap: 0B 0B 0BCreate and enable it
sudo fallocate -l 2G /swapfile sudo chmod 600 /swapfile sudo mkswap /swapfile sudo swapon /swapfilechmod 600is required: swap receives whatever was in memory, process passwords included. On ext4 — the file system of Ubuntu cloud images —fallocatecreates the file instantly. On btrfs, see “If it doesn’t work”.Enable it at boot
echo '/swapfile none swap sw 0 0' | sudo tee -a /etc/fstabCheck the line before rebooting: a mistake in
/etc/fstabcan stop the boot at the emergency console.sudo findmnt --verifychecks the file without a reboot.Make the kernel less eager to swap
vm.swappinessis how readily the kernel moves pages to disk. The default of 60 is a desktop setting. On a server, set 10: swap kicks in when memory is in fact short.echo 'vm.swappiness=10' | sudo tee /etc/sysctl.d/99-swappiness.conf sudo sysctl --system | grep swappiness # vm.swappiness = 10
The systemd journal
journald keeps the journal in /var/log/journal and by default takes up to 10% of the file system, capped at
4 GB. On a 25 GB disk that’s 2.5 GB — a tenth of the server for logs. Set the limit explicitly.
rsyslog runs alongside: it writes the same events to /var/log/syslog and /var/log/auth.log, which
logrotate rotates. Don’t remove it — CrowdSec reads login attempts from /var/log/auth.log (CrowdSec against SSH brute force).
See how much it takes
journalctl --disk-usage # Archived and active journals take up 1.4G in the file system.Set the limit
sudo mkdir -p /etc/systemd/journald.conf.d sudo tee /etc/systemd/journald.conf.d/00-size.conf >/dev/null <<'EOF' [Journal] SystemMaxUse=500M SystemKeepFree=1G EOF sudo systemctl restart systemd-journaldSystemMaxUse=500Mcaps the journal at 500 MB; on a small server that’s a few weeks of history.SystemKeepFree=1G: if less than 1 GB is left on disk, the journal gives way before hitting its cap. A separate file injournald.conf.dsurvives package upgrades, just as withsshdandapt.Remove the excess right away
sudo journalctl --vacuum-size=500M # Vacuuming done, freed 912.0M of archived journals from /var/log/journal/…
Docker container logs are a separate story; their limit goes in daemon.json (Docker on Ubuntu).
Time
An accurate clock isn’t cosmetic. A browser rejects a certificate whose notBefore is in the future, two-factor
codes live 30 seconds, and CrowdSec and logs from different servers are matched by time.
| Ubuntu 24.04 | Ubuntu 26.04 | |
|---|---|---|
| service | systemd-timesyncd |
chrony on new installs; an upgrade from 24.04 keeps timesyncd |
| protocol | NTP, UDP 123 | NTS — NTP with server authentication: TCP 4460 for key exchange plus UDP 123 |
| your own servers | /etc/systemd/timesyncd.conf.d/*.conf, NTP= |
/etc/chrony/sources.d/*.sources, then sudo chronyc reload sources |
| check | timedatectl timesync-status |
chronyc tracking, chronyc -N authdata |
NTS protects against time being spoofed on the way: plain NTP can be forged, and the server will trust someone else’s clock.
Check the sync
timedatectl # Local time: Sun 2026-10-18 19:40:11 UTC # Universal time: Sun 2026-10-18 19:40:11 UTC # Time zone: Etc/UTC (UTC, +0000) # System clock synchronized: yes # NTP service: activeyesandactive— the clock is fine, and you’re done.If it says no — find which server is unreachable
On 24.04:
timedatectl timesync-status # Server: 185.125.190.56 (ntp.ubuntu.com) # Poll interval: 34min 8s (min: 32s; max 34min 8s) # Offset: -1.204msNo
Serverline, or anOffsetin seconds — look for what blocks outbound UDP 123.On 26.04:
chronyc -N authdata # Name/IP address Mode KeyID Type KLen Last Atmp NAK Cook CLen # ========================================================================= # 1.ntp.ubuntu.com NTS 4 30 128 24d 0 0 8 64 # 2.ntp.ubuntu.com NTS 4 30 128 24d 0 0 8 64 # ntp-bootstrap.ubuntu.com NTS 4 30 128 24d 0 0 8 64This is the working state:
KeyIDisn’t zero,Cookis the stock of keys for requests,Lastis how long ago chrony got keys over TCP 4460. The default sources are Ubuntu’s pools from/etc/chrony/sources.d/ubuntu-ntp-pools.sources, all withnts.If outbound TCP 4460 is closed, there are no keys, and chrony sends no time requests at all, even though UDP 123 is open:
chronyc sources # MS Name/IP address Stratum Poll Reach LastRx Last sample # =============================================================================== # ^? ntp-nts-2.ps5.canonical.> 0 6 0 - +0ns[ +0ns] +/- 0ns # ^? ntp-nts-3.ps5.canonical.> 0 6 0 - +0ns[ +0ns] +/- 0ns chronyc -N authdata # 1.ntp.ubuntu.com NTS 0 0 0 - 2 0 0 0 # 2.ntp.ubuntu.com NTS 0 0 0 - 2 0 0 0 journalctl -u chrony --since "-10min" | grep NTS-KE # chronyd[442730]: NTS-KE session with 203.0.113.123:4460 (2.ntp.ubuntu.com) timed outThe signs: every source
^?withReach 0, zeros inauthdataand a growingAtmpattempt counter,NTS-KE session … timed outin the journal.A closed port doesn’t show right away. chrony keeps its keys in
/var/lib/chrony, reuses them after a restart, and fetches new ones only every few weeks. So a server whose egress firewall closed 4460 keeps syncing for a while, and then the clock stops being checked for no visible reason.
If it doesn’t work
swapon: /swapfile: swapon failed: Invalid argument or skipping - it appears to have holes
- Cause
- The file was created with
fallocateon btrfs or an old XFS — there it ends up with “holes”, and the kernel won’t use it as swap. - Fix
- Create the file with
dd:sudo dd if=/dev/zero of=/swapfile bs=1M count=2048 status=progress, then the samechmod,mkswap,swapon. On btrfs, usebtrfs filesystem mkswapfile.
swapon: /swapfile: insecure permissions 0644, 0600 suggested
- Cause
- Everyone can read the swap file, and it can hold passwords and keys from process memory.
- Fix
sudo chmod 600 /swapfile.
The journal still takes several gigabytes
- Cause
- The cap applies to new writes;
journaldremoves old files only when it writes new ones. - Fix
sudo journalctl --vacuum-size=500Mremoves the old data right away.
timedatectl: System clock synchronized: no
- Cause
- Outbound UDP 123 is closed (at the provider or in your own egress firewall), and on 26.04 TCP 4460 for NTS as well.
- Fix
- Look at
timedatectl timesync-status(24.04) orchronyc sources -v(26.04); open the outbound ports or set the provider’s NTP server.
26.04: chronyc sources shows every source as ^? with Reach 0, though UDP 123 is open
- Cause
- Outbound TCP 4460 is closed: chrony gets no NTS keys and without them sends no time requests.
- Fix
- Check
journalctl -u chrony | grep NTS-KE— it showsNTS-KE session … timed out; open TCP 4460 outbound and runsudo systemctl restart chrony.
Failed to set ntp: NTP not supported, or chronyd won’t start
- Cause
- A VPS on container virtualization (LXC, OpenVZ): the clock is shared with the host, and the guest can’t change it.
- Fix
- Do nothing — the provider keeps the clock. Only check that the time is right:
date -uagainst an accurate clock.
How to check
swapon --show
# NAME TYPE SIZE USED PRIO
# /swapfile file 2G 0B -2
cat /proc/sys/vm/swappiness
# 10
journalctl --disk-usage
# Archived and active journals take up 488.0M in the file system.
timedatectl show -p NTPSynchronized --value
# yes
After a reboot, the same: swap came up from /etc/fstab, and swappiness is still 10.