Updates
WHost Updates
The Overview tab draws the version card from the agent's last vendor check (the scheduler asks daily and five minutes after every agent start); when that answer is more than an hour old, opening the tab asks the vendor again in the background and the card follows the answer. The version card shows the current version, the version this server is offered (the next release it can install — one release at a time between stable releases, the newest between betas of one line), the channel and, when a newer release exists, the release notes in the Changelog card, grouped by kind (new features, improvements, security, fixes, changes) and written in the panel's language when the vendor published them in it, in English otherwise. "Check for Updates" asks the vendor again. A check that could not be answered — the vendor unreachable, a limiter answer — is shown on the card as "Failed to check for updates" with the reason; the card never reads "You're up to date" for a check that did not run. "Install Update" opens a confirmation naming the versions; the install runs on the server, the card turns into the eight-step progress stepper (backup → downloading → verifying → extracting → migrating → applying → restarting → health_check), and when the agent has restarted on the new version the page reloads itself. A refused install (already on the latest version, no check in this agent process, a limiter answer — six starts per hour, the toast names the seconds left, or a backup, a restore or a scheduled backup running on the server — the install ends in a restart that would cut it, so the toast names that work and when it started; install once it has finished) is shown as an error toast with the agent's reason and the card stays where it was. While an install runs, backups and restores are refused and scheduled backups wait for its end (see Backups — scope and settings). A failed install shows the failure card with the agent's reason (for example a package whose code does not compile, or one that carries another version than the one offered) and, when the agent put the previous version back, the rollback notice; the service keeps running the version it had, nothing is restarted, and the History tab lists the run as rolled back. A release that installs but does not start is put back by the agent's post-update check about two minutes after the restart; the page then reports the failure and the History row reads rolled back. There is no roll-back button: the previous agent tree stays beside the new one on the server. Choosing the beta channel asks for beta releases; it does not admit the licence to the beta programme. A licence that is not admitted is answered from the stable channel: the card lists no beta release, and because the check carries that refusal the version card shows the note "Beta programme: this licence is not admitted, the stable channel was used". To receive beta releases, ask WISECP to admit the licence holder to the programme. The "Update available" banner in the topbar reads the result of the scheduler's last check (daily, and five minutes after every agent start) and never asks the vendor itself; "Check for Updates" refreshes it.
The History tab lists every install attempt — from/to version, date, manual or automatic, status (completed, failed, rolled back) with the error text on hover — ten per page.
Below it, while the guarded OS update run is on (or has runs to show), Automatic OS updates lists its last ten runs: the date, the kind of run (night run, manual run, undo, database update) with the choice it ran under, the result (installed, nothing to install, put back, failed; undone for an undo) and the packages counted by result. The eye button opens a run in full: each package with its group, its from and to versions and its result (a package left for a manual run with its reason), the groups put back with the checks that failed after them, the versions the run held back, and the units a package changed (below). The newest run that installed something other than the kernel offers Undo, behind a confirmation: its packages go back to the versions they had before the run, from the copies that run kept, and the versions removed are held back until a newer one is published; the kernel is not put back. Older runs are not offered — their packages may have moved on since. The undo runs on the server in a unit of its own (it goes on if the agent restarts); the card reads In progress while a run of any kind runs, the buttons wait for it, and the result appears as a new row.
The Settings tab saves every change at once, one field per save: the update channel (stable / beta; a server installed from a beta release starts on beta), the WHost Updates switch — automatic installs of new WHost releases (agent and panel), not of operating system packages — with its type (all, security fixes only, patch only, minor), six preferences — backup before update, notify when available, notify when installed, check OS packages periodically, notify on OS security updates, allow the vendor to push critical updates — each with a line under it saying what it does, and, last on the tab, the Operating System Security Updates card with its switch, the choice of what it installs, and the night restart (next two sections). Backup before update archives the agent code alone — not the configuration — under /var/whost/backups/update_<time>/, and the newest three archives are kept; undoing a failed update does not need it, because the run and the post-update check put the previous version back themselves. The two notify switches decide whether the scheduler sends the update-available notice and the notice after an automatic install; a failed automatic install is always reported, an install started from the panel sends neither, and the notification settings still choose panel and e-mail. Security mode without the vendor override never installs anything by itself; the page says so under the switch. An automatic install that finds a backup running waits and tries again 30 minutes later, without a failure notification. A save the agent refuses (a limiter answer, an invalid value) returns the control to its stored value and shows the reason.
OS Security Updates
The Operating System Security Updates card on the Settings tab decides whether the server installs operating system updates on its own, also when nobody signs in to the panel for weeks. Switched on, WHost's own guarded run installs them every night, an hour before the night restart time (03:30 with the default 04:30, server time), whether the agent runs or not: it is a script on the system's own tools (/usr/local/sbin/whost-os-update, started by whost-os-update.timer), so it still works when an update broke the agent or its Python. The operating system's own installing run (unattended-upgrades on Ubuntu; dnf-automatic and WHost's PHP patch timer on AlmaLinux, Rocky Linux and CentOS Stream) is switched off while the guarded run is in place. A run that finds a backup, a restore or a round of scheduled backups running waits until the work has finished (up to three hours; after that the night's run is skipped), and the night restart waits for the run.
What it installs — the row above the night restart, shown while the switch is on:
- Security updates (the default) — what the OS Packages list classifies as security: the operating system's security pocket (Ubuntu's
-security) or security advisories (the RHEL family), the patch releases of the PHP versions WHost installed — 8.4.x to 8.4.y, from Ondřej Surý's PPA on Ubuntu and Remi's repository on the RHEL family; the PHP version itself never changes — and, on Ubuntu 22.04, the patch releases of the agent's Python (the deadsnakes PPA). - All updates — every pending update, under the same checks and rules.
Whatever the choice, MariaDB takes only the security fixes of its installed series on its own (its other updates wait for Back up and update MariaDB on the OS Packages tab, and a move to the next long-term series is started from the MariaDB card there — never by the night run), the kernel takes only security updates — they take effect at the night restart and are never put back on their own (the previous kernel stays installed) — and an update that brings in a new package waits for you. A new installation starts on security updates.
How a run goes:
- Keep the installed version. Before anything is installed, the package file of each package's installed version is kept (
/var/lib/whost/os-update/archive/): from that archive, the package cache, the repositories, or rebuilt from the installed files (dpkg-repack/rpmrebuild, which the agent installs). An update whose installed version cannot be kept is not installed — it could not be put back — and is listed on the OS Packages tab for a manual run. - Group by group. Web server and WAF module, PHP, mail, FTP and DNS, database, the agent's Python, other packages, kernel — each group with its exact versions. A group whose installation would also change a package without a kept file, remove a package or bring in a new one waits for a manual run.
- Check after each group.
whost-os-checktests the configuration of every installed web server, the monitored services, a hidden test site of WHost's own (whost-canary.invalid, answering the server itself only) through the WAF and every installed PHP version along the path a hosted site's request takes, the panel and the agent, the SMTP / IMAP / FTP greetings, a DNS question and a database query. Only a check that passed before the run is judged, so a problem that was already there is not blamed on the update. - Put back and hold back. When a check that passed before fails after a group, that group goes back to the kept files, the check runs again, and the version that broke it is held back until a newer one is published — the hold then lifts itself. The other groups keep their updates.
- Services stay as you set them. A package's scripts may switch on a service you keep off — OpenLiteSpeed's start its server and enable it at boot, where it would take the web ports from nginx at the next restart — or switch off one you keep on. The run notes before its first install which services, sockets and timers are enabled and running, and after every install, put-back or undo — before the check — puts each one a package changed back as it was: switched off again (and stopped if it was not running), or enabled again. It does not start a service a package stopped (the check covers WHost's services) and leaves a service a package brought as it came; both are listed. The run's details on the History tab show these units (state before → after the package → after WHost put it back, as enabled/running) and the run's notice names them; a unit WHost could not put back makes the run failed (critical notice). WHost's own units and a mask a package ships are left out.
The health check also runs three minutes after every boot against the last clean check, so a problem that appears after the night restart is reported too. A configuration file changed on the server keeps its version (the package's new copy is saved beside it as .dpkg-dist / .rpmnew), and a manual apt / dnf run behaves as before. rspamd, OpenLiteSpeed and Node.js packages are covered by All updates only.
The card shows, while the switch is on: a run in progress; a last run that put a group back (warning) or did not finish (error), with a button to the History tab; a health check that found a check failing that passed before; then the facts — the last automatic install with its packages, the next run with the tool, the last health check (how many checks, how many failing) and, with the night restart on, the last restart that came back clean — and Restart required when the server asks for one. Every time on the card is server time, written with the server's time zone (for example Nov 2, 2026, 03:30 AM UTC), whatever time zone the browser is in.
- Off — nothing installs on its own, Ubuntu's own automatic security updates included; you are only notified when security updates are pending (Notify on OS security updates) and apply them from the OS Packages tab. The card marks this as not recommended.
- Never saved — a server updated from a release without this setting keeps the operating system's default until the switch is saved: on Ubuntu its own security updates install on their own (without WHost's checks), PHP patch releases do not; on the RHEL family nothing installs on its own. The card says which one applies and recommends switching on. A new installation saves the switch on.
Notifications go to the panel and by e-mail (the update category, per the notification settings): updates installed and checked; an update put back after a failed check, naming the packages and the checks (critical); a run that did not finish (critical); a run put back from the panel; a health check that found a check failing that passed before (critical; one found while the agent was stopped is told when it starts); and for a MariaDB series upgrade (OS Packages tab, below): upgraded and checked, the old series put back after a failed check, naming the checks and sites (critical), an upgrade that did not start and why (warning), an upgrade that did not finish cleanly (critical). A run notice also names the units a package changed and what WHost did about them (step 5 above). Opening the health check notice goes to the Settings tab, the run notices to the History tab.
Saving the switch or the choice applies it in the background; the card shows Applying… and, when something could not be written or installed, the error. The agent checks its files every 10 minutes and after every start and puts back what is missing: /usr/local/sbin/whost-os-update and whost-os-check, /etc/whost/os-update/update.conf and check.conf, whost-os-update.service / .timer, whost-os-check.service / .timer, the test site (/var/www/whost-canary, /etc/nginx/conf.d/whost-canary.conf, Apache zz-whost-canary.conf); the runs and checks are kept under /var/lib/whost/os-update/. Endpoints: GET /system/update/os-security (mode, running_since, runs, manual, locks, last_check, database), PUT /system/update/settings (os_packages_auto_security, os_packages_auto_mode as security / all).
Night Restart
A kernel, microcode, the C library and a few system services (dbus, systemd-logind) take effect only after the server restarts; the other services an update touches are restarted by the package tools themselves. The Night restart switch at the bottom of the Operating System Security Updates card restarts the server at the time chosen beside it — server time, 04:30 by default, the time zone shown next to the field — and only when the operating system asks for a restart: /var/run/reboot-required on Ubuntu (the packages that set it are listed in /var/run/reboot-required.pkgs), dnf needs-restarting -r on the RHEL family. A night on which nothing asks passes without a restart. It is a setting of its own: it covers updates installed from the OS Packages tab as well, whatever the security updates switch says.
- WHost's own timer runs it (
whost-os-reboot.timerstartswhost-os-reboot.service, which runs with the agent's interpreter on the agent's files), so it works while the agent is stopped too. A window missed while the server was off is not made up at the next boot. - The night is skipped, and the next night tries again, while one of these runs: a backup, a restore or a round of scheduled backups; a WHost update (and the check after its restart) or an OS package upgrade started from the panel; an account transfer; the guarded OS update run (also while it waits for backup work to finish), an undo or a database update started from the panel, or the operating system's own automatic update run.
- Before going down it records the packages that asked, the running kernel and the services that run. When the same packages, at the same versions, still ask after that restart, it makes no new attempt for them — the restart did not take; check the boot loader and the server console — and says so once.
- After the boot the agent asks the services that ran before the restart (it waits up to five minutes for slow ones) and compares the running kernel with the one the update brought.
Notifications go to the panel and by e-mail (the update category, per the notification settings): the restart the next window will make, once per boot and set of packages; a skipped night and why; how the server came back — services not running, by name (critical), or another kernel than the expected one (warning); packages that still ask after a restart made for them (warning); a restart that could not be started (critical). While the switch is on, the "Server restart required" notice after an automatic install is not sent: the scheduled one replaces it.
The card shows the switch, the time field (saved when it loses focus) and a note only when there is something to say: the restart the next window makes with its packages, a skipped night with its reason, a restart that came back with a problem, or packages that still ask after a restart. A restart that came back clean is one of the card's facts at the top (with the security updates switch off, it shows as a line under the switch). While the switch is on, the "Restart required" note at the top of the card gives way to it.
- Never saved — a server updated from a release without this setting writes nothing and restarts nothing; the card recommends switching it on. A new installation saves it on.
- Off — the unit and the timer are removed; the server restarts only when you restart it (OS Packages tab).
The files: /etc/systemd/system/whost-os-reboot.service and .timer; the record of the last attempt and the last restart is /var/lib/whost/os_auto_reboot.json. Endpoints: GET /system/update/os-security (reboot), PUT /system/update/settings (os_auto_reboot, os_auto_reboot_time as HH:MM).
OS Packages
Lists pending apt / dnf upgrades with the installed and available versions and a category (security, feature, bug fix, other); the tab badge and the topbar banner carry the same counts. Security covers what the automatic run installs: an update of the operating system's security pocket (on Ubuntu also when the same version is published to -updates), a security advisory on the RHEL family, and the PHP patch releases of the PHP repositories WHost uses (and the agent's Python on Ubuntu 22.04); "Apply Security Updates" installs exactly these packages. The list is the agent's last package scan (daily, and after every upgrade); opening the tab scans again in the background only when that scan is more than an hour old, and the refresh button beside the card title scans at once. A scan reads the package index from the mirrors at most every 10 minutes (in between it lists from the index already on the server) and answers the previous list again at once while neither that index nor the installed packages have changed; scans asked for at the same time share one run. "Apply Security Updates" (disabled when no security update is pending) and "Apply All Updates" each open a confirmation; the upgrade runs in the background and the card above the list shows its progress and the time it started, then its result with the start and end times and the number of packages upgraded — the page can be left while it runs. A failed upgrade names what stopped it: for each repository whose files could not be downloaded, its host, the HTTP status it answered and the number of files (for example "ppa.launchpadcontent.net answered 503 — 9 files could not be downloaded"), followed by one hint — a 5xx answer means the repository is unavailable for a while and nothing on the server needs fixing (try again later), no answer at all means the server could not reach it (check its network and DNS); a failure that was not a download shows the package manager's last line instead. The finished card stays until the next upgrade or until it is closed with the × button beside its title: closing it clears the result for every administrator (the record stays in the server's upgrade history) and is refused while an upgrade runs. A list that could not be read (a limiter answer, the package manager busy) is shown as an error with a Refresh button, never as "no updates available". When the host asks for a reboot (kernel, libc) a banner offers "Reboot Now" behind its own confirmation. The last 100 upgrade attempts are kept on the server. Package names sent to the API are checked against the pending list — a name that is no longer pending is refused straight away with Unknown package(s) — not in the pending update list, so a stale request cannot occupy the single upgrade slot. While the guarded OS update run is working — the night run, an undo, a database update or a MariaDB series upgrade — both apply buttons are refused with "An OS package upgrade is running. Try again when it finishes." (409 OS_UPGRADE_IN_PROGRESS): packages changed between the run's steps would escape its check and its put-back. They are refused the same way while other package work started from the panel runs, and the message names it: a PHP version install or removal ("A PHP version is being installed or removed. Try again when it finishes.", 409 PHP_PACKAGE_JOB_IN_PROGRESS) or a web server plugin's activation, deactivation or update ("A web server plugin is being activated, deactivated or updated. Try again when it finishes.", 409 PLUGIN_TASK_IN_PROGRESS) — one such job runs at a time. A run that starts while an upgrade from this tab, a PHP version job or a web server plugin's work is in progress waits for it to finish before it plans (an hour at most; a run whose wait runs out installs nothing and is reported as a run that did not finish), and an upgrade that finds a run started a moment before it ends at once, failed, with nothing installed. While any of this work runs, the MariaDB card's series upgrade lists an OS update run in progress among what stops it.
While the guarded run is on, the MariaDB card follows the list, and two more cards when they have something to show:
-
MariaDB — the installed series and version, and the next long-term series still supported with its end-of-support date (or that the installed one is the newest still supported; a series whose support has ended is skipped, so 10.5 goes to 10.11). Upgrade series opens a dialog that reads the plan first (the source is looked up on the network): the target series; the source — the distribution's repositories when they carry it (a module stream on AlmaLinux, Rocky Linux and CentOS Stream), otherwise MariaDB's repository for that series, added for the upgrade only; the data folder and its size; the free disk against what the copy of the data folder, the dump and 1.5 GB to spare need; about how long the databases stay read-only; how many sites are checked (the front page of the active accounts that have a database — 25 picked at random when there are more); and what stops the upgrade now — no MariaDB server package, already the newest supported series, no source for this release, not enough disk, no
/root/.my.cnfto sign in with, MariaDB not answering, an OS update run or backup work in progress. Start the upgrade runs it in a unit of its own (whost-os-update db-series, so it goes on if the agent restarts; the night run and the night restart wait for it):- The package file of every installed database package is kept, as for the night run (a version that cannot be kept stops the upgrade before anything changes), and the new series is downloaded.
- The databases turn read-only. Their state is measured — every table and view opens, the health check, the database lookups of Postfix, Dovecot, PowerDNS, Pure-FTPd and the panel, the sample of sites — every database is dumped (
/var/lib/whost/os-update/db-backups/db-series-<run>.sql.gz), MariaDB stops, and its data folder and configuration are copied. - The new series is installed, started read-only, upgraded (
mariadb-upgrade) and restarted, and the same measurements are taken again: a table that opened before and does not now, a new error line in the server's log, or a check, a database lookup or a site that passed before and fails now (a lookup that fails is asked again for a minute, a site twice more — a pooled connection the restart cut reconnects) is a failure. - Everything passes: the databases open for writing on the new series, the copy of the data folder is deleted and the dump is kept for 7 days. Something fails: the data folder and the configuration come back from the copy, the old series' packages from their kept files and a source added for the upgrade is removed; the old series starts read-only, the same checks run, and the databases open for writing — nothing written is lost, because nothing could be written meanwhile.
While the databases are read-only (usually under a minute), sites and services can read but not write, and the agent refuses the database changes it makes as root (a new database or database user, a password change) with
409 DATABASE_UPGRADE_IN_PROGRESS. The card shows the stage while it runs, and the History tab the run: series before → after, the server after a put-back, how long the databases were read-only, the sites checked, the dump, and the checks that failed on the new series and, if any, after the put-back. After a success there is no way back from the panel — going back would lose what was written since: the run cannot be undone, nor can a run from before it, and the dump stays for 7 days. When the new series came from MariaDB's repository, the repository stays pinned to that series and its patch releases count as security fixes for the night run. -
Left for a manual run — the updates the last automatic run did not install, each with its reason: the installed version's package could not be kept (so the update could not be put back); a move to another MariaDB series (never automatic); a MariaDB update outside security; a kernel update outside security; an update that brings in a new package; an installation that would also change packages whose installed version could not be kept. Apply them from the list above when it suits you. When MariaDB updates of the installed series are among them, Back up and update MariaDB (behind a confirmation) dumps every database first — the last three dumps are kept in
/var/lib/whost/os-update/db-backups/, and without a dump nothing is touched — then installs those updates as a run of the database group: checked, and put back if a check fails. A move to the next MariaDB series is made from the MariaDB card. The run appears on the History tab as a database update. -
Held-back versions — the versions held back after a failed check or an undo, with the reason. Lift (behind a confirmation) removes the hold, so the next run — or an update you apply — may install that version again. On the Debian family the holds are apt pins in
/etc/apt/preferences.d/90whost-os-update-locks; on the RHEL family versionlock excludes.
Endpoints used: GET /system/update/check, POST /system/update/install, GET /system/update/status/stream, GET /system/update/history, GET/PUT /system/update/settings, GET /system/update/os-security, POST /system/update/os-security/runs/{run_id}/undo, DELETE /system/update/os-security/locks/{name}, POST /system/update/os-security/database, GET/POST /system/update/os-security/database/series, GET /system/packages/summary, GET /system/packages/updates, POST /system/packages/upgrade, GET /system/packages/upgrade/status, POST /system/reboot. The update preferences live in /etc/whost/update_settings.json on the server (configuration.md).
Our support team is here around the clock for anything you can't find above.