Automatic security updates now have their own "Turn on" / "Turn off"
action with a confirmation, instead of a field that the Save bar
applied. The confirmation explains what can stop:
- Docker from the operating system packages (docker.io, moby-engine)
restarts on a security update, so all containers stop and start.
- Docker CE from the Docker repository is not updated: its repository
has no security origin (apt) and no security advisories (dnf).
Checked on Debian 12 (unattended-upgrade --dry-run skips docker-ce)
and CentOS Stream 9.
- Services that use an updated library can restart; a new kernel is
installed without a reboot.
The on/off state is locked, so only the confirmed action can change it.
A new maintenance window is applied to the server on Save only when the
updates are on.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The check frequency, automatic security updates, and maintenance window
are now fields of the first section, in one row, instead of two extra
boxes. The "not supported" note moved into the existing callout.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- ui.server.packages_update_requested: user, server, packages (or all),
package manager, and the activity of the update.
- ui.server.packages_update_finished: written when the remote process
finishes (also when it fails), with the result and the exit code of
the activity. The update gets a fresh type_uuid (an indexed activity
property), so the listener finds exactly this activity. The actor
comes from the request, because the listener runs in the queue.
- ui.server.reboot_requested with result started or failed.
- ui.server.auto_security_updates_updated (enabled, time, package
manager, result) and ui.server.patch_settings_updated for a changed
check frequency.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- ServerSetting::patchStatus() summarizes the saved patch check
(updates, security updates, reboot required, error). Results saved
before security and reboot detection count as 0 and false.
- Grid cards show "N security updates", "N updates", "Patch check
failed", and "Reboot required" badges; the table gets a Patches
column from lg. The badge text is searchable.
- The toolbar shows a one-line team summary, for example "Patches: 2
servers with updates · 1 with security updates · 1 needs a reboot".
- Badges and the summary are shown only for servers where the user can
view the security pages (viewSecurity on the server), so members do
not see them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- New server settings: is_auto_security_updates_enabled and
auto_security_updates_time (maintenance window, HH:MM in the time zone
of the server). Both move with a server transfer.
- Debian/Ubuntu: install unattended-upgrades and write
52coolify-unattended-upgrades. `#clear` drops the distro origin lists,
so only the security origins stay (Debian <codename>-security, Ubuntu
<codename>-security, and the Ubuntu Pro ESM security origins). A
drop-in for apt-daily-upgrade.timer sets the time.
https://manpages.debian.org/bookworm/apt/apt.conf.5.en.htmlhttps://github.com/mvo5/unattended-upgrades/blob/master/README.md
- RHEL family: install dnf-automatic (dnf5-plugin-automatic on dnf5),
set upgrade_type = security and apply_updates = yes in
/etc/dnf/automatic.conf, and add a drop-in for dnf-automatic.timer
(dnf4) or dnf5-automatic.timer (dnf5).
https://dnf.readthedocs.io/en/latest/automatic.htmlhttps://dnf5.readthedocs.io/en/latest/dnf5_plugins/automatic.8.html
- Off removes the Coolify files, so the operating system default comes
back. Other package managers show "not supported".
- The schedule and the automatic updates share one form and one
unsaved bar. The remote commands run only when the automatic update
settings change; a failure keeps the old setting.
- Commands are written for non-root SSH users (tee and single sh -c
scripts) and were run in Debian 12, Ubuntu 24.04, Rocky 9, and Fedora
containers as root and as a sudo user.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Security updates: apt marks a package from a "-security" suite (Debian
12 now reports oldstable-security). dnf reads `dnf updateinfo list
--security` (dnf5 keeps updateinfo as an alias of advisory). zypper
uses `zypper -x patch --dry-run --category security`, because
`zypper -x list-patches` ignores --category and has no package names.
pacman and apk cannot tell, so they show no security count.
https://dnf5.readthedocs.io/en/latest/commands/advisory.8.htmlhttps://github.com/openSUSE/zypper/blob/master/doc/zypper.8.txt
- Reboot required: /var/run/reboot-required, `needrestart -b` (KSTA 2
or 3), `dnf needs-restarting -r` (matched by text, because a missing
plugin also exits 1), and `zypper needs-rebooting` (exit 102).
https://github.com/liske/needrestart/blob/master/README.batch.mdhttps://dnf-plugins-core.readthedocs.io/en/latest/needs_restarting.html
- An explicit risk list per package manager (Docker engine, kernels,
libc/glibc/musl, systemd, dbus, openssh) replaces the "name contains
docker/kernel" guess on the page and in every notification channel.
- The page shows security and reboot badges, a security count, and the
risk of each package. "Reboot server" (with confirmation, update
policy on the loaded server) schedules the reboot with systemd-run and
checks the server again five minutes later. Archived (transferred)
servers are never touched.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- server_settings gets patch_check_result (JSON), patch_check_error,
patch_checked_at, and patch_check_frequency.
- The patches page shows the saved result and "Last checked" on load.
"Check for updates" queues ServerPatchCheckJob (manual) instead of
running SSH in the Livewire request. The job saves the result and
broadcasts ServerPatchCheckFinished on the server team channel; the
page reloads the saved result (with a 10 s poll as a fallback).
- A finished package update (callEventOnFinish) queues a new check on
the server side, so the page no longer depends on the browser.
ServerPackageUpdated is removed.
- The check frequency is a server setting (cron or preset, at most once
per hour). Empty keeps the staggered weekly default.
- The failure notification is sent only when the state changes from ok
to error, not on every check.
- Page access needs viewSecurity and every action needs update on the
loaded server; a member gets 403. Updates accept only a package from
the saved result.
- Server transfer: the frequency moves, the last result stays on the
source (runtime).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- apt: use apt-get with DEBIAN_FRONTEND=noninteractive and keep changed
configuration files (--force-confdef/--force-confold). A single package
uses install --only-upgrade, so it is never installed when missing.
- pacman: do not run `pacman -Sy` on the live database during the check.
Use checkupdates (with fakeroot) or a temporary sync database, the same
method checkupdates uses. pacman 7 downloads as the alpm user, so the
temporary directory is 755. Arch allows only "Update all" (-Syu), because
partial upgrades are unsupported.
https://man.archlinux.org/man/checkupdates.8https://wiki.archlinux.org/title/System_maintenance#Partial_upgrades_are_unsupported
- dnf: list upgrades with `repoquery --upgrades --latest-limit 1 --qf`
(dnf5 needs the explicit \n, dnf4 prints empty lines) and read the
current versions from `rpm -qa`. The package name has no .arch suffix.
https://dnf5.readthedocs.io/en/latest/changes_from_dnf4.html
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Rollback now also lists the commit images of the last 10 successful
deployments when the application pushes to a Docker registry. The
deployment pulls such an image when the server no longer has it.
The page and the API share GetRollbackImages, and the API returns a
source field (local or registry).
Resolve conflicts between main's security fixes and v4.5 features:
- Keep v4.5 features (log redaction, multi-S3 backups, transfer flow,
scheduled task modal, sidebar) and port main's fixes into them.
- Reuse RequiresDeployAbility for the Hostinger create endpoint.
- Align MCP tool abilities and descriptions with main's new deploy
requirements, and use deploy tokens in the affected MCP tests.
- Drop the duplicate DockerCleanupJobTest helper and tests, and the
exportBundle test for a method v4.5 removed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Require deploy permission for API changes that trigger server side effects, including resource deletion, storage access, provider operations, and database imports. Also escape Docker cleanup filters and sanitize log drain server names.
Require deploy permission when setting a post-install script or extra SSH keys, document the 403 response, and cover both denial and allowed deployment cases.
- Database variants: use the same database name in Forgejo and the database
container (`forgejo` by default); Postgres stayed unhealthy and MySQL/MariaDB
installs failed with access denied.
- SSH: publish ${FORGEJO_SSH_PORT:-22222} and set SSH_PORT, DOMAIN and
SSH_DOMAIN, so clone URLs are correct and several instances fit on a server.
- Runner variants: start Forgejo from /usr/bin/s6-svscan, lock the install
with a generated admin, register runner 13.2.0 with a generated shared
secret, and enable the templates.
Closes#11924
oldRawParser made the bind directory of a long-syntax source like
../data at {workdir}./data, next to the path that Docker Compose uses.
Resolve the source with replaceLocalSource(), as the compose parsers do,
so Coolify makes the directory that the container mounts. A source that
resolves above / now stops the deployment before remote work.
Closes#4055
ServiceParserReferenceTest parses reference fixtures for serviceParser and
applicationParser (one service per feature, with setting variations and error
cases) and every template, and compares the output with expected files.
scripts/parser-snapshot-diff shows how a code change alters the output of all
cases (save / baseline). Template expected files are gitignored.
A service that declares SERVICE_FQDN_X_PORT while the named service declares
SERVICE_URL_X_PORT (same value) now counts as a copy, like an exact copy, so it
does not take the named service's domain. Also replace the Postgres-only
regex in the magic variable loop with the same check in PHP.
Generate now stores the new secret at once. The unsaved bar compares
fields with the saved secrets, so other drafts stay marked as unsaved.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Call updateCompose() after saving an application's fqdn in the services
API urls handler so the domain is written back to the compose file.
Without this, StartService re-parsing the compose before deploy would
drop the domain and its SERVICE_URL_*/SERVICE_FQDN_* variables.
Add a feature test covering persistence after a second parse.
The template port 3003 sent API and realtime traffic to a port that the
containers do not listen on (502). Use port 80 as upstream does, update
images and env to the official 2.5.0 compose, drop the removed assistant
service, and default _APP_OPTIONS_FORCE_HTTPS to disabled (upstream
default) so installs on http domains reach the API.
When a compose service copies another service's SERVICE_URL_/SERVICE_FQDN_
declaration (old Twenty template: worker repeats SERVICE_URL_TWENTY_3000),
the parser and updateCompose wrote the variable from the copier's generated
domain, so a custom domain on the named service was lost on every deploy.
Skip exact copies; declarations with another path or port are unchanged.
Closes#11924
GNU tar exits with 1 when a file changes while it is read. The archive
is still valid, so only exit codes above 1 fail the backup.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
All instances share the same default update cron, so they request the
CDN at the same second and Cloudflare rejects the burst (error 1200).
Queue the CDN fetching jobs with a random 0-600s delay to spread the load.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Add a per-task overlap lock to ScheduledTaskJob that outlives the worker
timeout, so a long or killed run cannot overlap the next cron run.
- Mark the dropped occurrence as skipped instead of releasing the job.
Closes#11648
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A single service deploy with an unchanged config keeps the running
container, which is already on the network. The connect then logged a
false "already exists" error and a warning. Check the container's
networks first.
Closes#11295
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Andras BacsaiandClaude Opus 5.5
