[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Guix "ops"
From: |
Thompson, David |
Subject: |
Re: Guix "ops" |
Date: |
Thu, 28 May 2015 12:13:44 -0400 |
On Wed, May 27, 2015 at 3:41 PM, Ludovic Courtès <address@hidden> wrote:
> David Thompson <address@hidden> skribis:
>
>> Ludovic Courtès <address@hidden> writes:
>>
>>> Perhaps one addition eventually would be to allow IPs to be
>>> automatically allocated and have host name lookup DTRT in each VM.
>>
>> Do you have any idea how we could do that for local VMs? There's no
>> daemon managing the provision of these resources, so I don't know what
>> strategy can be used to automatically allocate static IPs.
>
> QEMU allows you to specify the guest’s IP, I think, and apparently it
> can create VLANs and connect several unprivileged QEMU instances
> together via -net socket (info "(qemu-doc) sec_invocation").
>
> Things like libvirt probably provide a higher-level interface to that.
> (I don’t know if it justifies the extra dependency.)
If libvirt's API was really useful, perhaps it could be an optional
dependency for users that want to deploy QEMU VMs?
>> The automagic hostname lookup part is particularly interesting to me. A
>> more complete deployment configuration would have the web server
>> dependent on the db server. I originally intended to handle this by
>> delaying the creation of the web server until after the db server was
>> made, and invoking a procedure that accepted the db server's state as
>> input and output the correct configuration for the web server.
>
> I’ve seen that Docker can do that. ;-) IIRC it populates /etc/hosts in
> each container. That’s something we could do. Another possibility
> would be to rely on mDNS.
If Docker can do it, so can we! :)
>> Thinking out loud here: Maybe 'guix deploy' can kick off the
>> provisioning for all machines first, and afterwards the OS configs can
>> be altered to include the correct /etc/hosts file.
>
> The transform procedure could force the right /etc/hosts in each OS, I
> suppose?
Yes. Perhaps the extensibility I had in mind could be better achieved
by allowing additional, user specified transformations in the machine
declaration.
Thanks for your input!
- Dave
- Re: Guix "ops", (continued)