maas-devel team mailing list archive

[Gavin Panella <gavin.panella@xxxxxxxxxxxxx>]

On 8 November 2013 13:32, Andreas Hasenack <andreas@xxxxxxxxxxxxx> wrote:
>
> On Fri, Nov 8, 2013 at 11:15 AM, Gavin Panella <gavin.panella@xxxxxxxxxxxxx>
> wrote:
>>
>> Some solutions I can think of:
>>
>> 1. MAAS could arrange that reverse look-ups return the configured
>>    hostname of nodes instead of the generated one (using
>>    generated_hostname).
>
>
> Would work.
>
>>
>>
>> 2. Whatever writes pg_hba.conf could use the name from a reverse look-up
>>    instead of the hostname directly.
>
> Sounds cumbersome. The charm just gets private-address, and that is
> returning a name that it has to convert to an IP via a dns lookup?

Yeah, it is a bit, for no clear benefit over #2.

>
>>
>> 3. Whatever writes pg_hba.conf could use the IP address instead of the
>>    hostname directly.
>
> There is no IP address in the relation data. "relation-get private-address"
> returns a fqdn for some reason. I didn't check what unit-get returns on the
> postgresql server.

This wouldn't be too tricky; gethostip (in syslinux) would make this a
fairly trivial thing to do - or socket.gethostbyname() in Python. You
would need to add a trailing /32 to denote a host. See
http://www.postgresql.org/docs/9.1/static/auth-pg-hba-conf.html

>
>>
>> 4. Whatever writes pg_hba.conf could add extra rules for both of the
>>    above.
>>
>> I'm not sure of the repurcussions of a change like #1. We chose to use
>> the generated name for a reason, though I can't remember exactly what it
>> was. Not having to regenerate the zone file everytime the hostname is
>> changed might have been part of it.
>
> But you do add a CNAME everytime a machine comes up.

Yeah, which means we've not been averse to writing zone files, and so
that probably wasn't the reason we made it this way.

I suggest you do #3 for now while we try to figure this one out. To be
honest, I doubt we'll change MAAS's policy here soon; we've got too much
else to do already, and a switch to a different policy will have its own
problems most likely. Sorry :-/

>
>>
>> #2, #3 and #4 require changes to the charm, which I'll assume is easier
>> to do than making a change to core MAAS. I guess it's confusing to charm
>> authors though: a hostname is a hostname, right? It's an unexpected
>> wrinkle. We probably ought to fix #1, or at least justify better why
>> it's like that. In the meantime one of the other options ought to
>> unblock you.
>>
>> If you agree, can you file a bug about #1?
>
> What about the /etc/hosts entry pointing the machine name to 127.0.1.1, do
> you know what adds that? Is it juju? Without that entry, it also works.

I think this is just the usual setting-the-hostname stuff: MAAS will
pass in the configured hostname (instead of the generated one) to the OS
installer, which sets the hostname accordingly, and adds an /etc/hosts
line so that traffic destined for itself (by name) gets routed via the
lo interface. Someone else - roaksoax or smoser most likely - will have
to confirm or correct that explanation.

MAAS will also return this name to Juju, and so that's the name that
Juju passes around to hooks and suchlike.