[PKG-Openstack-devel] debian openstack and puppet

[Thomas Goirand]

On Tue Jun 24 2014 04:56:31 AM HKT, Benedikt Trefzer <benedikt.trefzer at cirrax.com> wrote:

> Hi list
> 
> I'm installing an openstack cluster, that is spanned over several nodes.
> All config files (and also package installation) is handled with puppet.
> The Debian packages do quite a lot of configuration work using
> preseeding etc. But actually these configuration do not work together
> well with the puppet manifests.

This is not at all the experience that reported the
company that sponsors my packaging work.

> Problems occuring when package updates arriving. Either certain
> configuration parameters are asked interactivly

This can only be your fault. Either do:
dpkg-reconfigure debconf

and select the non-interactive mode, so that debconf
is disabled forever (note: this can also be
preseeded or edited somewhere in /etc) or use
DEBIAN_FRONTEND=non-interactive each time you do the
upgrades.


> or the default values
> from installation are taken. Both results in overwriting the
> configuration made by puppet.

I very much doubt that. All config scripts are
reading the configuration files and do the db_set
calls accordingly, just like the Debian policy
dictates. If you find such behavior in one of the
packages, make sure you can reproduce it, and then
please report it as a bug!

> Even worse db synchronisation (something made by the postinst script) is
> not executed correctly, since the database connection string is at its
> default value at the time of execution....

Same here. Directives are read from the config
files. So this cannot happen. Well, unless you are
producing obsolete directives which have moved to a
new section / have been renamed in Icehouse, which
is possible, but then packages can't be blamed:
update your puppet scripts...

All this is by the way described in both the Debian
policy manual and OpenStack official (Debian)
install guide. There's a chapter I wrote there
especially conserning non-interactive mode.

> (also endpoint registrations
> does not work (which is actually good, since it does not corrupt the
> whole cluster, but waiting for the timeout and an error message is not
> the gents way of doing....)).

By default, no endpoint registration will happen,
unless you explicitly request for it using debconf.
The value is also forgotten willingly after the
endpoint registration. So no way this affects you
unless you are running debconf in ineractive mode.

> Actually anyone using a
> config system like puppet, chef, ansible etc. will face the same problem.

Well... NO!!! :) I'm convinced that you are having
a wrong impression here, or doing the wrong
assemptions.

Thomas (from my phone)