On an unmanaged server the provider is responsible only for hardware, network and virtualization; OS, PHP and database updates, backups, monitoring and incident response are yours. On our test bench that adds up to roughly 2-7 hours per month per server. Managed hosting takes the server side off your hands, semi-managed splits it between you and the provider, and none of the six providers whose terms we read will fix your code, plugins or themes.
How we checked
On 10 November 2026 we read the official support scope pages and documentation of DigitalOcean, Hetzner, Kinsta, WP Engine, SiteGround and Namecheap (links in the table below). We reproduced an unmanaged server owner’s routine on a test bench: a clean Ubuntu 24.04 LTS built from the April 2024 release packages in a separate chroot, nginx 1.24, PHP-FPM 8.3, MariaDB 10.11, unattended-upgrades, certbot 2.9 and Pebble, the test ACME server from the Let’s Encrypt team. PHP 8.4 came from Ubuntu 25.10 because the third-party PHP repository is not reachable from the bench. Command run times are measured, human time is estimated. These are bench figures, not RichHost server figures.
The three models in one minute
Unmanaged. You get a server with root access; the provider runs the hardware, network and hypervisor. DigitalOcean states plainly that its Droplets are “completely unmanaged” and that support cannot access the server itself. Everything above the hypervisor is yours, from kernel patches to the SSL certificate.
Managed. The provider updates the OS and server software, takes backups, watches services and brings them back after a crash. Examples: Hetzner Managed Server (system updates, service monitoring, security fixes, daily backups) or managed WordPress at Kinsta or WP Engine.
Semi-managed. There is no standard definition: the provider takes one layer and the rest is yours. In practice you see two setups. One is your own VPS plus a managed database that the provider updates and backs up. The other is shared hosting, where the server, PHP and mail belong to the provider and the site, CMS and plugins belong to you. Shared hosting trades control for convenience, so check in advance which hidden shared hosting limits hit first.
Responsibility matrix: who does what
The table is compiled from the provider support pages in the next section. “Provider” means the provider does the task, “you” means you or your admin do it.
| Task | Unmanaged VPS | VPS + managed DB | Shared hosting | Managed WordPress |
|---|---|---|---|---|
| Hardware, network, hypervisor | provider | provider | provider | provider |
| OS updates and security patches | you | server: you; DB: provider | provider | provider |
| Service restarts and reboots after updates | you | server: you; DB: provider in the maintenance window | provider | provider |
| PHP version and extensions | you | you | provider installs, you pick the version and test the site | provider installs, you test the site |
| Database version and updates | you | provider, cannot be disabled | provider | provider |
| Backups | you (disk snapshots as a paid option) | DB: provider; files: you | provider tool; restore testing: you | provider; restore testing: you |
| Monitoring and alerts | you | you; DB metrics: provider | server: provider; site: you | platform: provider; site: you |
| Response when the server goes down | you | server: you; DB: provider | provider | provider |
| Code, plugins, themes, hacked site | you | you | you | you, with limited help from support |
The last row is the same in every column. Managed does not mean someone will update your plugins and fix the conflict an update causes: Kinsta and WP Engine list exactly that among their exclusions.
What providers say about the limits of support
Quotes are in the original wording, checked on 10 November 2026.
| Provider and product | What it takes on | What it does not do | Source |
|---|---|---|---|
| DigitalOcean Droplet | Hardware, network, platform; free ticket support with replies within 24 hours | “Droplets are completely unmanaged”; support “cannot access your Droplets directly” | Droplet Policies |
| DigitalOcean Managed MySQL | Database updates in a four-hour window, daily backups kept for seven days, point in time restore | Updates are mandatory: “you can’t disable them”; connections may drop briefly during the window | Updates, backups |
| Hetzner Cloud | Host and microcode patches; daily server backups, seven slots (separate option) | Backups “do not include any Volumes”; no Windows support | FAQ, backups |
| Hetzner: dedicated server and Managed Server | Dedicated: “Hardware and network support only”. Managed: system updates, service monitoring, security fixes, daily backups | On dedicated, OS updates “Must be managed independently” | Managed vs bare metal |
| Kinsta, managed WordPress | Site backups, switching PHP versions, investigating reproducible performance issues | “Any code editing”; “Updating WordPress core, plugins, or themes on your behalf”; “Code audits” | Scope of support |
| WP Engine, managed WordPress | Automatic WordPress updates, backups, server maintenance and security | “cannot assist with custom code”; third-party plugins and themes: only “limited support” | Scope of support |
| SiteGround, shared | Server software (Apache, PHP, NGINX), backup tool, PHP version manager | “Writing or debugging custom code”; “Website security audits and malicious code clean-up” | Scope of support |
| Namecheap, shared | Standard hosting support | “Configuration of third-party scripts and software”; “manual backup restorations” | Support boundaries |
Two takeaways. First, the line almost always runs along your site’s code. Even on managed WordPress, Kinsta offers plugin updates only as a paid add-on, and support does not fix what an update breaks. Second, provider backups come with conditions: DigitalOcean keeps database copies for seven days, Hetzner Cloud does not back up attached volumes, Namecheap does not do manual restores under standard support. That is why your own backup restore testing matters under any model.
How many hours an unmanaged server takes: the bench
We built a server that nobody had updated since installation: Ubuntu 24.04 with April 2024 packages, nginx, PHP-FPM 8.3, MariaDB with a test shop database (300,000 orders and 20,000 products, 165 MB) and a simple PHP script that counts orders. Then we ran the owner’s five routine tasks.
1. Security updates
Once the update repositories were enabled, 146 packages were pending and unattended-upgrades picked 109 from the security pocket. The dry run took 86 seconds, the installation 158 seconds. Then the details showed up.
# unattended-upgrade --dry-run -v
Allowed origins are: o=Ubuntu,a=noble, o=Ubuntu,a=noble-security, o=UbuntuESMApps,a=noble-apps-security, o=UbuntuESM,a=noble-infra-security
Packages that will be upgraded: apt apt-utils bsdutils ... (109 packages)
# unattended-upgrade -v (wall 158.5 s)
# mariadb -N -e "select version()"
10.11.7-MariaDB-2ubuntu2
# apt list --upgradable 2>/dev/null | grep -c upgradable
51
- The packages were updated, but the old MariaDB 10.11.7 was still running in memory. The new code only took effect after a service restart, and restarting nginx, PHP-FPM and MariaDB produced 3.1 seconds of errors: 27 refused connections and one PHP fatal error. On a regular systemd server the package scripts restart some services themselves, so the short outage happens whenever the automatic update runs.
- By default unattended-upgrades installs security updates only. The other 51 packages, including the MariaDB 10.11.14 bugfix release, were left to us (another 17 seconds of
apt-get upgradeplus a restart). - The Ubuntu documentation confirms that automatic reboot is off by default, so an updated kernel waits until you reboot.
In a normal month, with everything already up to date, the dry run takes 1.4 seconds. The time goes into reading the package list and checking the site after the restart, not into the commands.
2. PHP 8.3 → 8.4
The Ubuntu 24.04 archive only ships PHP 8.3. Security updates patch it, but the version string stays “8.3.6” (our package was built on 19 October 2026), so vulnerability scanners that only look at the version number get it wrong. PHP 8.3 gets security fixes until 31 December 2027, PHP 8.4 until 31 December 2028 (php.net). To switch, we installed PHP-FPM 8.4 next to the old version (16 seconds), started a separate pool and pointed nginx at it.
# curl -s -o /dev/null -w "%{http_code}" http://127.0.0.1:12881/
500
PHP Fatal error: Uncaught Error: Class "mysqli" not found in /var/www/shop/index.php:2
# apt-get install php8.4-mysql (wall 2.5 s), pool reload
PHP 8.4.11 | orders 300000
# memory_limit / upload_max_filesize under 8.4: 128M 2M (8.3 had 256M 64M)
Two mistakes in five minutes: a forgotten mysqli extension took the site down, and the custom php.ini settings did not carry over to the new version. Rolling back took 0.4 seconds because the old pool was still running. The step by step order is in our guide on how to switch PHP versions safely.
3. Database dump and restore
# mariadb-dump --single-transaction --routines --triggers --databases shop | gzip > shop.sql.gz
wall 2.6 s, 12 212 619 bytes
# restore into a separate database wall 6.1 s
CHECKSUM TABLE orders: 824665296 = 824665296, products: 197893451 = 197893451
# app user on the restored database
ERROR 1142 (42000): SELECT command denied to user 'shop'@'localhost' for table `shop_restore`.`orders`
# UPDATE of one row during a dump without --single-transaction: 2.400 s, with it: 0.017 s
The restore worked and the checksums matched. But a single database dump contains no users or grants, so on a new server the application was denied access. And a dump without --single-transaction locks the tables: a one-row write waited 2.4 seconds, which becomes minutes on a large database.
4. Log rotation
On a fresh server logrotate -d answered “log does not need rotating” for all 11 patterns, which means it checked nothing. A forced rotation exposed the real problem: nginx kept writing to access.log.1, the new access.log stayed empty, and logrotate printed no error because the post-rotation command sends its output to /dev/null. On the bench the cause was a blocked service command; on a live server the same happens whenever that command breaks. The fix: nginx -s reopen.
5. Certificate renewal
certbot renew --dry-run always talks to the Let’s Encrypt staging server, so it cannot run offline: on the bench it failed with NameResolutionError for acme-staging-v02.api.letsencrypt.org. We pointed certbot at a local Pebble instead.
# certbot renew --dry-run (wall 2.4 s)
Congratulations, all simulated renewals succeeded
# same with nothing listening on the HTTP port (wall 1.4 s)
Type: connection
Detail: ... dial tcp 127.0.0.1:12880: connect: connection refused
Let’s Encrypt certificates are valid for 90 days today, 64 days from 10 February 2027 and 45 days from 16 February 2028 (Let’s Encrypt announcement). Renewal has to be automatic and checked regularly.
Monthly check script
We rolled the five checks into one script for Ubuntu 24.04. Early versions raised false alarms (shared memory of nginx and PHP-FPM looked like a “deleted library”); the version below was tested on the bench against real problems: stale binaries after an update, the nginx log, the certificate. A full run takes 11-14 seconds.
#!/bin/bash
# monthly-check.sh: monthly self-check for an unmanaged Ubuntu 24.04 web server
DB=shop
echo "== updates"
apt-get update -qq
echo "upgradable: $(apt list --upgradable 2>/dev/null | grep -c upgradable)"
unattended-upgrade --dry-run -v 2>&1 | grep -E "will be upgraded|No packages found" | cut -c1-120
[ -f /run/reboot-required ] && echo "REBOOT REQUIRED" || echo "no reboot flag"
echo "== services still running replaced libraries"
for p in $(pgrep -x mariadbd) $(pgrep -x php-fpm8.3) $(pgrep -x nginx); do
n=$(grep '(deleted)' /proc/$p/maps | grep -cE '[.]so|/usr/s?bin/')
[ "$n" -gt 0 ] && echo "restart $(cat /proc/$p/comm)"
done | sort -u
echo "== restore test"
mariadb-dump --single-transaction --routines --triggers $DB | gzip > /tmp/$DB.sql.gz
mariadb -e "DROP DATABASE IF EXISTS check_$DB; CREATE DATABASE check_$DB"
zcat /tmp/$DB.sql.gz | mariadb check_$DB
for t in $(mariadb -N -e "SHOW TABLES" $DB); do
a=$(mariadb -N -e "CHECKSUM TABLE $DB.$t" | cut -f2)
b=$(mariadb -N -e "CHECKSUM TABLE check_$DB.$t" | cut -f2)
[ "$a" = "$b" ] && echo "ok $t" || echo "MISMATCH $t"
done
mariadb -e "DROP DATABASE check_$DB"; rm -f /tmp/$DB.sql.gz
echo "== logs"
for p in $(pgrep -x nginx); do
ls -l /proc/$p/fd | grep -q 'access.log.1' && echo "nginx writes to access.log.1: run nginx -s reopen"
done | sort -u
echo "== certificates"
certbot renew --dry-run --no-random-sleep-on-renew 2>&1 | grep -E "succeeded|failure|Detail"
echo "== disk"
df -hP / | awk 'NR==2 {print "used " $5 ", free " $4}'
Hours per month: an estimate from the bench
Method: machine time is measured on the bench; human time is our estimate of reading output, deciding and checking the site after each step, multiplied by frequency. It is an estimate, not a time study of an admin’s work. Monitoring and incidents were not reproduced on the bench and are listed separately.
| Task | Frequency | Machine time on the bench | Human time per month (estimate) |
|---|---|---|---|
| Security updates and a planned restart | weekly review, monthly install | 1.4 s dry run; 158 s to install 109 packages; 3.6 s restart | 35-50 min |
| Moving to a new PHP version | about once a year | 16 s install, 0.4 s rollback; two mistakes | 10-20 min (2-4 h per upgrade) |
| Database restore test | monthly | 2.6 s dump and 6.1 s restore for 165 MB | 15-30 min |
| Log rotation | monthly | under 0.1 s | 5-10 min |
| Certificates | monthly | 2.4 s | 5-10 min |
| Monitoring and alerts (not on the bench) | ongoing | not measured | 30-60 min |
| Routine subtotal | 1.7-3 h | ||
| Incidents (not on the bench) | unpredictable | not measured | 0-4 h |
| Total | roughly 2-7 h per server |
Multiply the upper bound by your admin’s hourly rate and compare it with the price gap between unmanaged and managed. Without proper hosting monitoring the incident line usually grows, because you learn about outages from customers.
How to choose: if X, pick Y
| If | Pick |
|---|---|
| One or a few WordPress or PHP sites, no admin | Shared hosting: the server is the provider’s job, the site, plugins and backup checks are yours |
| The site earns money every hour, no team, budget allows | Managed WordPress or a managed server; before paying, find out from the contract who updates plugins and fixes what updates break |
| You need root and your own services (queues, Node.js, Docker) and someone has 2-7 hours a month | Unmanaged VPS |
| The database is critical and you can run the server yourself | Semi-managed: VPS plus a managed database; plan for mandatory updates in the maintenance window |
| An hour of your time costs more than the managed vs unmanaged price gap divided by 2-7 | Managed |
| You have never restored a database from a dump | Do not go unmanaged until you pass a restore test like the one above |
| Your VPS is running out of resources | First compare VPS vs dedicated server, then pick the management model separately |
Where RichHost shared hosting fits
RichHost sells virtual (shared) hosting: the provider runs the server side and the owner manages the site. Plans start at $8 per month and include our own control panel with an AI assistant, free Let’s Encrypt certificates, continuous cloud backups and servers in Europe and the USA. Support works from 10:00 to 20:00 Kyiv time and replies in about 30 minutes. RichHost does not sell VPS or dedicated servers, so if you need root access, choose a management model with another provider using the tables above. If you do not need root, see the RichHost shared hosting plans.
FAQ
Does managed hosting update WordPress plugins?
It depends on the provider. WP Engine lists automatic WordPress updates among its features; Kinsta updates plugins and themes only as a paid add-on and does not fix conflicts after an update. Look for this in the support scope page before you pay.
Are Ubuntu security updates enabled automatically?
Yes. unattended-upgrades is installed by default and applies security updates only, without rebooting. Bugfix updates, like MariaDB 10.11.14 on our bench, and service restarts are left to you.
How much time per month does an unmanaged VPS take?
Our bench estimate is 1.7-3 hours of routine work per server plus up to 4 hours for incidents, roughly 2-7 hours in total. The check script cuts the machine part to 11-14 seconds, but you still have to read the result and act on it.
Do I need my own backup if the provider makes one?
Yes. Provider backups have conditions: DigitalOcean keeps database copies for seven days, Hetzner Cloud excludes attached volumes, Namecheap does not do manual restores under standard support. Your own copy and a tested restore close those gaps.



