New in version 3004.
A transactional system, like MicroOS, can present some challenges when the user decided to manage it via Salt.
MicroOS provides a read-only rootfs and a tool,
transactional-update
, that takes care of the management of the
system (updating, upgrading, installation or reboot, among others) in
an atomic way.
Atomicity is the main feature of MicroOS, and to guarantee this
property, this model leverages snapper
, zypper
, btrfs
and
overlayfs
to create snapshots that will be updated independently
of the currently running system, and that are activated after the
reboot. This implies, for example, that some changes made on the
system are not visible until the next reboot, as those changes are
living in a different snapshot of the file system.
This model presents a lot of problems with the traditional Salt model, where the inspections (like 'is this package installed?') are executed in order to determine if a subsequent action is required (like 'install this package').
Lets consider this use case, to see how it works on a traditional system, and in a transactional system:
Check if apache
is installed
If it is not installed, install it
Check that a vhost
is configured for apache
Make sure that apache2.service
is enabled
If the configuration changes, restart apache2.service
In the traditional system everything will work as expected. The
system can see if the package is present or not, install it if it
isn't, and a re-check will shows that is already present. The same
will happen to the configuration file in /etc/apache2
, that will
be available as soon the package gets installed. Salt can inspect the
current form of this file, and add the missing bits if required. Salt
can annotate that a change is present, and restart the service.
In a transactional system we will have multiple issues. The first one
is that Salt can only see the content of the snapshot where the system
booted from. Later snapshots may contain different content, including
the presence of apache
. If Salt decides to install apache
calling zypper
, it will fail, as this will try to write into the
read-only rootfs. Even if Salt would call transactional-update pkg
install
, the package would only be present in the new transaction
(snapshot), and will not be found in the currently running system when
later Salt tries to validate the presence of the package in the
current one.
Any change in /etc
alone will have also problems, as the changes
will be alive in a different overlay, only visible after the reboot.
And, finally, the service can only be enabled and restarted if the
service file is already present in the current /etc
.
transactional-update
is the reference tool used for the
administration of transactional systems. Newer versions of this tool
support the execution of random commands in the new transaction, the
continuation of a transaction, the automatic detection of changes in
new transactions and the merge of /etc
overlays.
One prerequisite already present is the support for branching from a different snapshot than the current one in snapper.
With this feature we can represent in transactional-update
the
action of creating a transaction snapshot based on one that is planned
to be the active one after the reboot. This feature removes a lot of
user complains (like, for example, losing changes that are stored in a
transaction not yet activated), but also provide a more simple model
to work with.
So, for example, if the user have this scenario:
+-----+ *=====* +--V--+
--| T.1 |--| T.2 |--| T.3 |
+-----+ *=====* +--A--+
where T.2 is the current active one, and T.3 is an snapshot generated
from T.2 with a new package (apache2
), and is marked to be the
active after the reboot.
Previously, if the user (that is still on T.2) created a new
transaction, maybe for adding a new package (tomcat
, for example),
the new T.4 will be based on the content of T.2 again, and not T.3, so
the new T.4 will have lost the changes of T.3 (i.e. apache2 will not
be present in T.4).
With the --continue
parameter, transactional-update
will
create T.4 based on T.3, and nothing will be lost.
With transactional-update run
we will create a new transaction
based on the current one (T.2), where we can send interactive commands
that can modify the new transaction, and as commented, with
transactional-update --continue run
, we will create a new
transaction based on the last created (T.3)
The run
command can execute any application inside the new
transaction namespace. This module uses this feature to execute the
different Salt execution modules, via call()
. Or even the full
salt-thin
or salt-call
via sls()
, apply()
,
single()
or highstate
.
transactional-update
will drop empty snapshots¶The option --drop-if-no-change
is used to detect whether there is
any change in the file system on the read-only subvolume of the new
transaction will be added. If a change is present, the new
transaction will remain, if not it will be discarded.
For example:
transactional-update --continue --drop-if-no-change run zypper in apache2
If we are in the scenario described before, apache2
is already
present in T.3. In this case a new transaction, T.4, will be created
based on T.3, zypper
will detect that the package is already
present and no change will be produced on T.4. At the end of the
execution, transactional-update
will validate that T.3 and T.4 are
equivalent and T.4 will be discarded.
If the command is:
transactional-update --continue --drop-if-no-change run zypper in tomcat
the new T.4 will be indeed different from T.3, and will remain after the transaction is closed.
With this feature, every time that we call any function of this execution module, we will minimize the amount of transaction, while maintaining the idempotence so some operations.
A change in the system will create a new transaction, that needs to be
activated via a reboot. With pending_transaction()
we can check
if a reboot is needed. We can execute the reboot using the
reboot()
function, that will follow the plan established by the
functions of the rebootmgr
execution module.
/etc
overlay merge when no new transaction is created¶In a transactional model, /etc
is an overlay file system. Changes
done during the update are only present in the new transaction, and so
will only be available after the reboot. Or worse, if the transaction
gets dropped, because there is no change in the rootfs
, the
changes in /etc
will be dropped too!. This is designed like that
in order to make the configuration files for the new package available
only when new package is also available to the user. So, after the
reboot.
This makes sense for the case when, for example, apache2
is not
present in the current transaction, but we installed it. The new
snapshot contains the apache2
service, and the configuration files
in /etc
will be accessible only after the reboot.
But this model presents an issue. If we use transactional-update
--continue --drop-if-no-change run <command>
, where <command>
does not make any change in the read-only subvolume, but only in
/etc
(which is also read-write in the running system), the new
overlay with the changes in /etc
will be dropped together with the
transaction.
To fix this, transactional-update
will detect that when no change
has been made on the read-only subvolume, but done in the overlay, the
transaction will be dropped and the changes in the overlay will be
merged back into /etc
overlay of the current transaction.
With this module we can create states that leverage Salt into this kind of systems:
# Install apache (low-level API)
salt-call transactional_update.pkg_install apache2
# We can call any execution module
salt-call transactional_update.call pkg.install apache2
# Or via a state
salt-call transactional_update.single pkg.installed name=apache2
# We can also execute a zypper directly
salt-call transactional_update run "zypper in apache2" snapshot="continue"
# We can reuse SLS states
salt-call transactional_update.apply install_and_configure_apache
# Or apply the full highstate
salt-call transactional_update.highstate
# Is there any change done in the system?
salt-call transactional_update pending_transaction
# If so, reboot via rebootmgr
salt-call transactional_update reboot
# We can enable the service
salt-call service.enable apache2
# If apache2 is available, this will work too
salt-call service.restart apache2
This module alone is an improvement over the current state, but is easy to see some limitations and problems:
The user needs to know if the system is transactional or not, as not everything can be expressed inside a transaction (for example, restarting a service inside transaction is not allowed).
In the apache2
example from the beginning we can observe the
biggest drawback. If the package apache2
is missing, the new
module will create a new transaction, will execute pkg.install
inside the transaction (creating the salt-thin, moving it inside and
delegating the execution to transactional-update
CLI as part of the
full state). Inside the transaction we can do too the required
changes in /etc
for adding the new vhost
, and we can enable the
service via systemctl inside the same transaction.
At this point we will not merge the /etc
overlay into the current
one, and we expect from the user call the reboot
function inside
this module, in order to activate the new transaction and start the
apache2
service.
In the case that the package is already there, but the configuration
for the vhost
is required, the new transaction will be dropped and
the /etc
overlay will be visible in the live system. Then from
outside the transaction, via a different call to Salt, we can command
a restart of the apache2
service.
We can see that in both cases we break the user expectation, where a change on the configuration will trigger automatically the restart of the associated service. In a transactional scenario we need two different steps: or a reboot, or a restart from outside of the transaction.
Alberto Planas <aplanas@suse.com>
new
None
Linux
Apply an state inside a transaction.
This function will call transactional_update.highstate or transactional_update.sls based on the arguments passed to this function. It exists as a more intuitive way of applying states.
For a formal description of the possible parameters accepted in this function, check state.apply_ documentation.
If at the end of the transaction there is a pending activation (i.e there is a new snapshot in the system), a new reboot will be scheduled (default False)
CLI Example:
salt microos transactional_update.apply
salt microos transactional_update.apply stuff
salt microos transactional_update.apply stuff pillar='{"foo": "bar"}'
salt microos transactional_update.apply stuff activate_transaction=True
Reinstall the bootloader
Same as grub.cfg, but will also rewrite the bootloader itself.
Check for newer transactional-update versions.
Use the given snapshot or, if no number is given, the current default snapshot as a base for the next snapshot. Use "continue" to indicate the last snapshot done.
CLI Example:
salt microos transactional_update bootloader snapshot="continue"
Executes a Salt function inside a transaction.
The chroot does not need to have Salt installed, but Python is required.
Salt execution module function
If at the end of the transaction there is a pending activation (i.e there is a new snapshot in the system), a new reboot will be scheduled (default False)
CLI Example:
salt microos transactional_update.call test.ping
salt microos transactional_update.call ssh.set_auth_key user key=mykey
salt microos transactional_update.call pkg.install emacs activate_transaction=True
Run both cleanup-snapshots and cleanup-overlays.
Identical to calling both cleanup-snapshots and cleanup-overlays.
Check for newer transactional-update versions.
CLI Example:
salt microos transactional_update cleanup
Remove unused overlay layers.
Removes all unreferenced (and thus unused) /etc overlay directories in /var/lib/overlay.
Check for newer transactional-update versions.
CLI Example:
salt microos transactional_update cleanup_overlays
Mark unused snapshots for snapper removal.
If the current root filesystem is identical to the active root filesystem (means after a reboot, before transactional-update creates a new snapshot with updates), all old snapshots without a cleanup algorithm get a cleanup algorithm set. This is to make sure, that old snapshots will be deleted by snapper. See the section about cleanup algorithms in snapper(8).
Check for newer transactional-update versions.
CLI Example:
salt microos transactional_update cleanup_snapshots
Call 'zypper dup'
If new updates are available, a new snapshot is created and zypper dup --no-allow-vendor-change is used to update the snapshot. Afterwards, the snapshot is activated and will be used as the new root filesystem during next boot.
Check for newer transactional-update versions.
Use the given snapshot or, if no number is given, the current default snapshot as a base for the next snapshot. Use "continue" to indicate the last snapshot done.
CLI Example:
salt microos transactional_update dup snapshot="continue"
Regenerate grub.cfg
grub2-mkconfig(8) is called to create a new /boot/grub2/grub.cfg configuration file for the bootloader.
Check for newer transactional-update versions.
Use the given snapshot or, if no number is given, the current default snapshot as a base for the next snapshot. Use "continue" to indicate the last snapshot done.
CLI Example:
salt microos transactional_update grub_cfg snapshot="continue"
Retrieve the state data from the salt master for this minion and execute it inside a transaction.
For a formal description of the possible parameters accepted in this function, check state.highstate documentation.
If at the end of the transaction there is a pending activation (i.e there is a new snapshot in the system), a new reboot will be scheduled (Default: False).
Instead of failing immediately when another state run is in progress, queue the new state run to begin running once the other has finished.
This option starts a new thread for each queued state run, so use this option sparingly (Default: False).
CLI Example:
salt microos transactional_update.highstate
salt microos transactional_update.highstate pillar='{"foo": "bar"}'
salt microos transactional_update.highstate activate_transaction=True
Check if Salt is executing while in a transaction
CLI Example:
salt microos transactional_update in_transaction
Regenerate initrd
A new initrd is created in a snapshot.
Check for newer transactional-update versions.
Use the given snapshot or, if no number is given, the current default snapshot as a base for the next snapshot. Use "continue" to indicate the last snapshot done.
CLI Example:
salt microos transactional_update initrd snapshot="continue"
Regenerate kdump initrd
A new initrd for kdump is created in a snapshot.
Check for newer transactional-update versions.
Use the given snapshot or, if no number is given, the current default snapshot as a base for the next snapshot. Use "continue" to indicate the last snapshot done.
CLI Example:
salt microos transactional_update kdump snapshot="continue"
Updates systems registered via SCC / SMT
On systems which are registered against the SUSE Customer Center (SCC) or SMT, a migration to a new version of the installed products can be made with this option.
Check for newer transactional-update versions.
Use the given snapshot or, if no number is given, the current default snapshot as a base for the next snapshot. Use "continue" to indicate the last snapshot done.
CLI Example:
salt microos transactional_update migration snapshot="continue"
Call 'zypper patch'
If new updates are available, a new snapshot is created and zypper patch is used to update the snapshot. Afterwards, the snapshot is activated and will be used as the new root filesystem during next boot.
Check for newer transactional-update versions.
Use the given snapshot or, if no number is given, the current default snapshot as a base for the next snapshot. Use "continue" to indicate the last snapshot done.
CLI Example:
salt microos transactional_update patch snapshot="continue"
Check if there is a pending transaction
CLI Example:
salt microos transactional_update pending_transaction
Install individual packages
Installs additional software. See the install description in the "Package Management Commands" section of zypper's man page for all available arguments.
Package name to install
List of packages names to install
String or list of extra parameters for zypper
Check for newer transactional-update versions.
Use the given snapshot or, if no number is given, the current default snapshot as a base for the next snapshot. Use "continue" to indicate the last snapshot done.
CLI Example:
salt microos transactional_update pkg_install pkg=emacs snapshot="continue"
Remove individual packages
Removes installed software. See the remove description in the "Package Management Commands" section of zypper's man page for all available arguments.
Package name to install
List of packages names to install
String or list of extra parameters for zypper
Check for newer transactional-update versions.
Use the given snapshot or, if no number is given, the current default snapshot as a base for the next snapshot. Use "continue" to indicate the last snapshot done.
CLI Example:
salt microos transactional_update pkg_remove pkg=vim snapshot="continue"
Updates individual packages
Update selected software. See the update description in the "Update Management Commands" section of zypper's man page for all available arguments.
Package name to install
List of packages names to install
String or list of extra parameters for zypper
Check for newer transactional-update versions.
Use the given snapshot or, if no number is given, the current default snapshot as a base for the next snapshot. Use "continue" to indicate the last snapshot done.
CLI Example:
salt microos transactional_update pkg_update pkg=emacs snapshot="continue"
Reboot after update
Trigger a reboot after updating the system.
Several different reboot methods are supported, configurable via the REBOOT_METHOD configuration option in transactional-update.conf(5). By default rebootmgrd(8) will be used to reboot the system according to the configured policies if the service is running, otherwise systemctl reboot will be called.
Check for newer transactional-update versions.
CLI Example:
salt microos transactional_update reboot
Set the current, given or last working snapshot as default snapshot
Sets the default root file system. On a read-only system the root file system is set directly using btrfs. On read-write systems snapper(8) rollback is called.
If no snapshot number is given, the current root file system is set as the new default root file system. Otherwise number can either be a snapshot number (as displayed by snapper list) or the word last. last will try to reset to the latest working snapshot.
Use the given snapshot or, if no number is given, the current default snapshot as a base for the next snapshot. Use "last" to indicate the last working snapshot done.
CLI Example:
salt microos transactional_update rollback
Run a command in a new snapshot
Execute the command inside a new snapshot. By default this snapshot will remain, but if --drop-if-no-change is set, the new snapshot will be dropped if there is no change in the file system.
Command with parameters that will be executed (as string or array)
Check for newer transactional-update versions.
Use the given snapshot or, if no number is given, the current default snapshot as a base for the next snapshot. Use "continue" to indicate the last snapshot done.
CLI Example:
salt microos transactional_update run "mkdir /tmp/dir" snapshot="continue"
Execute a single state function with the named kwargs, returns False if insufficient data is sent to the command
By default, the values of the kwargs will be parsed as YAML. So, you can specify lists values, or lists of single entry key-value maps, as you would in a YAML salt file. Alternatively, JSON format of keyword values is also supported.
If at the end of the transaction there is a pending activation (i.e there is a new snapshot in the system), a new reboot will be scheduled (Default: False).
Instead of failing immediately when another state run is in progress, queue the new state run to begin running once the other has finished.
This option starts a new thread for each queued state run, so use this option sparingly (Default: False).
CLI Example:
salt microos transactional_update.single pkg.installed name=emacs
salt microos transactional_update.single pkg.installed name=emacs activate_transaction=True
Execute the states in one or more SLS files inside a transaction.
Specify a salt fileserver environment to be used when applying states
List of states to execute
Run states in test-only (dry-run) mode
Exclude specific states from execution. Accepts a list of sls
names, a comma-separated string of sls names, or a list of
dictionaries containing sls
or id
keys. Glob-patterns
may be used to match multiple states.
If at the end of the transaction there is a pending activation (i.e there is a new snapshot in the system), a new reboot will be scheduled (Default: False).
Instead of failing immediately when another state run is in progress, queue the new state run to begin running once the other has finished.
This option starts a new thread for each queued state run, so use this option sparingly (Default: False).
For a formal description of the possible parameters accepted in this function, check state.sls documentation.
CLI Example:
salt microos transactional_update.sls stuff pillar='{"foo": "bar"}'
salt microos transactional_update.sls stuff activate_transaction=True
Check if the system is a transactional system
CLI Example:
salt microos transactional_update transactional
Call 'zypper up'
If new updates are available, a new snapshot is created and zypper up is used to update the snapshot. Afterwards, the snapshot is activated and will be used as the new root filesystem during next boot.
Check for newer transactional-update versions.
Use the given snapshot or, if no number is given, the current default snapshot as a base for the next snapshot. Use "continue" to indicate the last snapshot done.
CLI Example:
salt microos transactional_update up snapshot="continue"