Specify lxd container bridge

[Dimiter Naydenov]

Hey Corey,

That specific error I haven't seen at that stage - allocating container
addresses. Can you please paste the machine-0.log as well? Are you able
to consistently reproduce this or it's intermittent?

Cheers,
Dimiter

On 09/17/2016 12:17 AM, Corey Bryant wrote:
> 
> 
> On Thu, Sep 1, 2016 at 4:25 AM, Dimiter Naydenov
> <dimiter.naydenov at canonical.com <mailto:dimiter.naydenov at canonical.com>>
> wrote:
> 
>     Hello!
> 
>     When using juju 2.0 on maas 1.9 or 2.0, you should get lxd containers
>     provisioned with as many interfaces as their host machine has, because
>     we're creating bridges on all configured host interfaces at initial boot
>     (e.g. eth0 becomes br-eth0, ens4.250 - br-ens4.250 and so on). Nothing
>     needs configuring to get this behaviour, but there's a caveat:
> 
>     In order for the above to work, there's a limitation currently being
>     addressed - all interfaces on the host machine in MAAS need to be linked
>     to a subnet and have an IP address configured - either as Static or
>     Auto, but not DHCP or Unconfigured. Otherwise the process of allocating
>     addresses for the container (represented as a MAAS Device, visible on
>     the host node's details page in MAAS UI under Containers and VMs) can
>     fail half way through and Juju will instead fall back to a the single
>     NIC LXD default profile, using lxdbr0 on a local subnet. You can tell
>     whether this happened, because there will be a WARNING in
>     /var/log/juju/machine-0.log on the bootstrap machine, like: `failed to
>     prepare container "0/lxd/0" network config: ...` describing the
>     underlying error encountered.
> 
>     Please note, the above limitation will be gone very soon - likely
>     beta18, not beta17 scheduled for release this week. In that upcoming
>     beta, unlinked or unconfigured host machine interfaces won't prevent the
>     multi-NIC container provisioning and address allocation - Juju will just
>     allocate addresses where it can, leaving the rest unconfigured, and not
>     falling back to using LXD default profile's lxdbr0.
> 
>     HTH,
>     Dimiter
> 
> 
> Hey Dimiter,
> 
> I'm hitting the same issue.  I have all the interfaces linked to subnets
> with auto but I still get the 'failed to prepare container "0/lxd/0"'
> error message saying 'connection is shut down'.  The containers are
> still using lxdbr0 (see http://paste.ubuntu.com/23188824/
> <http://paste.ubuntu.com/23188824/>).  The containers show up on the
> nodes page with juju*-lxd-*.maas names.  Do you have any other tips for
> getting past this?
> 
> Thanks,
> Corey 


-- 
Dimiter Naydenov <dimiter.naydenov at canonical.com>
Juju Core Sapphire team <http://juju.ubuntu.com>

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 473 bytes
Desc: OpenPGP digital signature
URL: <https://lists.ubuntu.com/archives/juju/attachments/20160917/7f0f6ef0/attachment.pgp>