On Tue, Sep 22, 2009 at 10:31 AM, Iustin Pop <[email protected]> wrote:
> On Tue, Sep 22, 2009 at 10:26:10AM +0100, Guido Trotter wrote:
>> On Tue, Sep 22, 2009 at 10:25 AM, Guido Trotter <[email protected]> wrote:
>> > On Tue, Sep 22, 2009 at 10:24 AM, Iustin Pop <[email protected]> wrote:
>> >> On Tue, Sep 22, 2009 at 10:09:43AM +0100, Guido Trotter wrote:
>> >>> On Tue, Sep 22, 2009 at 9:41 AM, Iustin Pop <[email protected]> wrote:
>> >>> >
>> >>> > I was going to add another flag to the cluster settings, and to me it
>> >>> > seems that it's not really good to have N flags directly int the 
>> >>> > cluster
>> >>> > object. For example, we have today “modify_etc_hosts”, and I want to 
>> >>> > add
>> >>> > now “enable_bridging”, we'll also need “modify_root_ssh”, etc.
>> >>> >
>> >>> > I think we should either name them all flag_$blah (not so nice for
>> >>> > cmdline) or move them to a sub-object cluster.flags (this is not good 
>> >>> > as
>> >>> > then gnt-cluster modify is harder to work, or maybe not).
>> >>> >
>> >>> > Is it a non-issue actually? Or is it, but I should just ignore it and
>> >>> > add my flag for now?
>> >>>
>> >>> Why making the interface to them more complicated by hiding them in a
>> >>> "flags" object?
>> >>> In the end what problem is there if they stay in the main cluster one?
>> >>> They're settings for a cluster.
>> >>
>> >> Yes, true, which is why I said maybe it's a non-issue.
>> >>
>> >> It just seems to me that if we have 8-10 boolean flags it is better if
>> >> we separate those, as just from the name it's not clear what they mean.
>> >> I'm happy to leave things as they are if you think it's fine.
>> >>
>> >>> Also, for the bridge, should we have an explicit enable/disable
>> >>> bridging, or just not populate the default bridge, if --no-bridging is
>> >>> passed (same as we do for --no-lvm-...)?
>> >>
>> >> We need explicit disable, in which case we won't run the cluster verify
>> >> checks for the dependencies of bridging.
>> >
>> > Which we could not run anyway, if there is not default bridge set (and
>> > no instance needs a bridge).
>>
>> == all the nics are in "routed" mode... Isn't this safer?
>
> We're back at square one, circa May 2009 :)
>

Wow! Like playing stairs and snakes! :)

> I still argue that explicit (yes, I need bridging) is better than
> implicit.  Otherwise, right after a cluster creation, with no instances,
> your proposal would make the cluster verify say everything's OK

Not if a default bridge is set up

> whereas
> it's not. And I like more if gnt-instance add rejects my misguided
> attempts to create a bridged network when the cluster config doesn't
> allow it.
>

But why not allowing it? We already support more than one bridge, so
if there is no default, and no instance has a bridge, no need to
check, if there is no defualt and some instance has some bridge, need
to check all bridges specified by instances, if there is a default,
that should exist everywhere! And at instance creation time, if you
specify a bridge, we check anyway!
(of course if no default one is there, default nic mode is routed) :)

> Now if this is done via empty default bridge or a boolean flag it's
> irrelevant.
>

Well, it depends on the semantics we want to give :)

Thanks,

Guido

Reply via email to