Before, including apt::unattended_upgrades on a host without
the unattended-upgrades package would fail on the first run,
because the module tries to install the package before apt is
finally configured.
This commit does:
- introduce the option $refresh_apt for apt::apt_conf
(Defaults to true). Can be used to not trigger Exec['refresh_apt']
- install the unattended-upgrades package after a final
Exec['refresh_apt']. To not run into a loop, it calls
Apt_conf['50unattended-upgrades'] with the option
refresh_apt => false, which is also not needed for the configuration
looks for a template based on the $lsbdistcodename, so you get this error:
Could not find template 'apt/Debian/preferences_n/a.erb'
The slash in 'n/a' appears to be a path (confusing!)
So I've set an error message when 'n/a' is encountered, encouraging the user to set the $lsbdistcodename.
I was not confident in the idea of pointing any occurrence of 'n/a' to the sid
templates, as I could imagine a case where this might occur on a non-sid system
where the lsbdistcodename isn't properly detected
What happened here was the $codename = $::lsbdistcodename was removed from
init.pp and replaced with just $::lsbdistcodename whereever $codename was
used. Then in the sources.list.erb and preferences files things were changed
like this:
<pre>+### Debian current: <%= codename = scope.lookupvar('::lsbdistcodename') %>
...
-deb <%= debian_url %> <%= codename %> <%= repos %>
...
+deb <%= debian_url=scope.lookupvar('apt::debian_url') %> <%= codename %> <%= repos=scope.lookupvar('apt::repos') %>
</pre>
This meant that the codename was always set to lsbdistcodename, and you because
in newer puppet versions you cannot assign a value to a top-level facter
variable, it is not possible to change this.
Because we cannot change $lsbdistcodename, we have to fix this by allowing the
user to pass a different, non-top-level scoped variable to the class as a
parameter, which defaults to $::lsbdistcodename, so that upgrades can be
triggered.
This is documented in the README in an upgrade notice
This ensures that resources that set a dependecy on the
"File['apt_config']" file, as mentioned in the README, can always
resolve.
Signed-off-by: Gabriel Filion <lelutin@gmail.com>
Since Debian Squeeze, apt supports a .d directory for preferences
(pinning). Simplify management by just placing files inside this .d
directory.
This removes the need for concat and concat::fragment resources, so
removes the dependency on the concat module. It also removes the need
for a moduledir that was needed for the concat module magic, so we also
get rid of the dependency on the common module.
Signed-off-by: Gabriel Filion <lelutin@gmail.com>
This define doesn't add much of a shortcut and imposes a dependency on
the "common" module. We'll be better off with the small repetition but
less tightly coupled modules.
Sadly, the "common" module is still required, but that is going to be
the focus of the commit that follows this one.
Signed-off-by: Gabriel Filion <lelutin@gmail.com>
This is implemented by adding a "package" parameter to apt::preferences_snippet,
so that define names can be kept unique while the package names are
not necessarily.
Closes: Redmine#3468.
sources_list doesn't currently force puppet to run 'apt-get update'
after creating/modifying/removing files in sources.list.d.
Signed-off-by: Gabriel Filion <lelutin@gmail.com>
The .d directories are only managed by the main 'apt' class. However,
both 'sources_list' and 'apt_conf' defines depend on those directories.
So in practice, the defines have an implicit need for those directories
to be somehow managed.
Let's turn this into an explicit relation, and include the directories
in the defines.
This makes it possible to use both defines without having to include the
main 'apt' class. (maybe when using puppet apply?)
Signed-off-by: Gabriel Filion <lelutin@gmail.com>
as we use $lsbdistcodename as source we cannot name the sources
which should be used to update "stable". -> Fix it by deploying
a per lsbdistcodename configfile. This can also be used as a pre-
work for the #2681 shared modules bug.
