12.9.6 Unattended Upgrades

Guix provides a service to perform unattended upgrades: periodically, the system automatically reconfigures itself from the latest Guix. Guix System has several properties that make unattended upgrades safe:

• upgrades are transactional (either the upgrade succeeds or it fails, but you cannot end up with an “in-between” system state);
• the upgrade log is kept—you can view it with guix system list-generations—and you can roll back to any previous generation, should the upgraded system fail to behave as intended;
• channel code is authenticated so you know you can only run genuine code (see 通道);
• guix system reconfigure prevents downgrades, which makes it immune to downgrade attacks.

To set up unattended upgrades, add an instance of unattended-upgrade-service-type like the one below to the list of your operating system services:

(service unattended-upgrade-service-type)

The defaults above set up weekly upgrades: every Sunday at midnight. You do not need to provide the operating system configuration file: it uses /run/current-system/configuration.scm, which ensures it always uses your latest configuration—see provenance-service-type, for more information about this file.

There are several things that can be configured, in particular the periodicity and services (daemons) to be restarted upon completion. When the upgrade is successful, the service takes care of deleting system generations older that some threshold, as per guix system delete-generations. See the reference below for details.

To ensure that upgrades are actually happening, you can run guix system describe. To investigate upgrade failures, visit the unattended upgrade log file (see below).

Scheme Variable: unattended-upgrade-service-type

This is the service type for unattended upgrades. It sets up an mcron job (see 执行计划任务) that runs guix system reconfigure from the latest version of the specified channels.

Its value must be a unattended-upgrade-configuration record (see below).

Data Type: unattended-upgrade-configuration

This data type represents the configuration of the unattended upgrade service. The following fields are available:

schedule (default: "30 01 * * 0")

This is the schedule of upgrades, expressed as a gexp containing an mcron job schedule (see mcron job specifications in GNU mcron).

channels (default: #~%default-channels)

This gexp specifies the channels to use for the upgrade (see 通道). By default, the tip of the official guix channel is used.

operating-system-file (default: "/run/current-system/configuration.scm")

This field specifies the operating system configuration file to use. The default is to reuse the config file of the current configuration.

There are cases, though, where referring to /run/current-system/configuration.scm is not enough, for instance because that file refers to extra files (SSH public keys, extra configuration files, etc.) via local-file and similar constructs. For those cases, we recommend something along these lines:

(unattended-upgrade-configuration
  (operating-system-file
    (file-append (local-file "." "config-dir" #:recursive? #t)
                 "/config.scm")))

The effect here is to import all of the current directory into the store, and to refer to config.scm within that directory. Therefore, uses of local-file within config.scm will work as expected. See G-表达式, for information about local-file and file-append.

services-to-restart (default: '(mcron))

This field specifies the Shepherd services to restart when the upgrade completes.

Those services are restarted right away upon completion, as with herd restart, which ensures that the latest version is running—remember that by default guix system reconfigure only restarts services that are not currently running, which is conservative: it minimizes disruption but leaves outdated services running.

Use herd status to find out candidates for restarting. See 服务, for general information about services. Common services to restart would include ntpd and ssh-daemon.

By default, the mcron service is restarted. This ensures that the latest version of the unattended upgrade job will be used next time.

system-expiration (default: (* 3 30 24 3600))

This is the expiration time in seconds for system generations. System generations older that this amount of time are deleted with guix system delete-generations when an upgrade completes.

注: The unattended upgrade service does not run the garbage collector. You will probably want to set up your own mcron job to run guix gc periodically.

maximum-duration (default: 3600)

Maximum duration in seconds for the upgrade; past that time, the upgrade aborts.

This is primarily useful to ensure the upgrade does not end up rebuilding or re-downloading “the world”.

log-file (default: "/var/log/unattended-upgrade.log")

File where unattended upgrades are logged.