crossbow discuss - Apr 2009 - Virtual devices disappear in dladm show-link

Using OpenSolaris 2008.11 build 110.

Created vnic''s and etherstub''s and they are displayed using
dladm show-links but when the server is rebooted a ''dladm
show-link'' is empty although the vnic and etherstub devices are still
there and function properly.

That is a real inconvenience, so why do the no longer show after a reboot or
shutdown/start?

Is there a trick to make sure they don''t hide because at some point in
time I will have to restart.
-- 
This message posted from opensolaris.org

On Fri, 2009-04-17 at 12:13 -0700, Gary Bainbridge
wrote:
> Using OpenSolaris 2008.11 build 110.
> 
> Created vnic''s and etherstub''s and they are displayed
using dladm show-links
> but when the server is rebooted a ''dladm show-link'' is
empty although the
> vnic and etherstub devices are still there and function properly.
That''s because the network/physical:nwam service doesn''t do
"dladm
up-vnic" like network/physical:default does.  It''s a bug in NWAM
that
will be fixed in "NWAM Phase 1".  Note that your VNICs are still in
the
persistent configuration, the problem is simply that the configuration
isn''t activated during boot when using NWAM.
> Is there a trick to make sure they don''t hide because at some
point in time
> I will have to restart.
You can either turn off NWAM and use network/physical:default to
configure the network, or you can manually bring up your VNICs using
"dladm up-vnic".

-Seb

> > Created vnic''s and etherstub''s and they are
displayed using dladm show-links
 > > but when the server is rebooted a ''dladm show-link''
is empty although the
 > > vnic and etherstub devices are still there and function properly.
 > 
 > That''s because the network/physical:nwam service doesn''t
do "dladm
 > up-vnic" like network/physical:default does.  It''s a bug in
NWAM that
 > will be fixed in "NWAM Phase 1".

I''m a bit surprised by this -- I thought that VNICs were not supported
under NWAM phase 1?

-- 
meem

On Fri, 2009-04-17 at 12:36 -0700, Peter Memishian
wrote:
> > > Created vnic''s and etherstub''s and they are
displayed using dladm show-links
>  > > but when the server is rebooted a ''dladm
show-link'' is empty although the
>  > > vnic and etherstub devices are still there and function
properly.
>  > 
>  > That''s because the network/physical:nwam service
doesn''t do "dladm
>  > up-vnic" like network/physical:default does.  It''s a
bug in NWAM that
>  > will be fixed in "NWAM Phase 1".
> 
> I''m a bit surprised by this -- I thought that VNICs were not
supported
> under NWAM phase 1?
I believe that they will be to some degree (this was one of the topics
discussed during the NWAM Phase 1 PSARC commitment review).  Perhaps
someone from the NWAM team can chime in and explain how they''ll be
handling them (or we can scoot over to nwam-discuss at opensolaris.org).

-Seb

On Fri, Apr 17, 2009 at 03:52:12PM -0400, Sebastien Roy
wrote:
> 
> On Fri, 2009-04-17 at 12:36 -0700, Peter Memishian wrote:
> > > > Created vnic''s and etherstub''s and they
are displayed using dladm show-links
> >  > > but when the server is rebooted a ''dladm
show-link'' is empty although the
> >  > > vnic and etherstub devices are still there and function
properly.
> >  > 
> >  > That''s because the network/physical:nwam service
doesn''t do "dladm
> >  > up-vnic" like network/physical:default does.  It''s
a bug in NWAM that
> >  > will be fixed in "NWAM Phase 1".
> > 
> > I''m a bit surprised by this -- I thought that VNICs were not
supported
> > under NWAM phase 1?
> 
> I believe that they will be to some degree (this was one of the topics
> discussed during the NWAM Phase 1 PSARC commitment review).  Perhaps
> someone from the NWAM team can chime in and explain how they''ll be
> handling them (or we can scoot over to nwam-discuss at opensolaris.org).
Seb is correct.  For NWAM Phase 1, we will include the "dladm up-vnic"
call in network/physical:nwam''s start method; and once vnics are
created,
they will be viewed by nwam as ordinary physical links, and can thus be
manipulated in nwam profiles.  What you will *not* be able to do via the
nwam tools is actually create vnics; for that, you will need to use the
existing dladm interface.

-renee

> Seb is correct.  For NWAM Phase 1, we will include the "dladm
up-vnic"
 > call in network/physical:nwam''s start method; and once vnics are
created,
 > they will be viewed by nwam as ordinary physical links, and can thus be
 > manipulated in nwam profiles.  What you will *not* be able to do via the
 > nwam tools is actually create vnics; for that, you will need to use the
 > existing dladm interface.

Is this also true for link aggregations and VLANs?

--
meem

Thank you.
-- 
This message posted from opensolaris.org

On Fri, Apr 17, 2009 at 01:20:46PM -0700, Peter Memishian
wrote:
> 
>  > Seb is correct.  For NWAM Phase 1, we will include the "dladm
up-vnic"
>  > call in network/physical:nwam''s start method; and once vnics
are created,
>  > they will be viewed by nwam as ordinary physical links, and can thus
be
>  > manipulated in nwam profiles.  What you will *not* be able to do via
the
>  > nwam tools is actually create vnics; for that, you will need to use
the
>  > existing dladm interface.
> 
> Is this also true for link aggregations and VLANs?
No.  Once created, aggregations and vlans have wider implications for
management of their underlying links.  NWAM will not support those until
it is able to fully manage creation/administration/destruction of them.
The expectation is that that support will be able to come fairly quickly
after phase 1 integrates.

-renee

> No.  Once created, aggregations and vlans have wider implications for
 > management of their underlying links.  NWAM will not support those until
 > it is able to fully manage creation/administration/destruction of them.
 > The expectation is that that support will be able to come fairly quickly
 > after phase 1 integrates.

Given that a VLAN is a specialized form of VNIC (e.g., I can specify a
VLAN ID to create-vnic), what implications does it have that VNICs do not?

-- 
meem

On Fri, Apr 17, 2009 at 01:33:15PM -0700, Peter Memishian
wrote:
> 
>  > No.  Once created, aggregations and vlans have wider implications for
>  > management of their underlying links.  NWAM will not support those
until
>  > it is able to fully manage creation/administration/destruction of
them.
>  > The expectation is that that support will be able to come fairly
quickly
>  > after phase 1 integrates.
> 
> Given that a VLAN is a specialized form of VNIC (e.g., I can specify a
> VLAN ID to create-vnic), what implications does it have that VNICs do not?
That''s a vaild point.  I suppose when you get down to it the answer
is historical: we''ve always lumped vlans and aggrs together, and
because of the extra work involved with aggrs, we just said support
for those would come later.  As you know, we''ve had an exceptionally
hard time drawing boundaries around what nwam will and won''t do, and
when each feature will be added; that was one clear line that we could
draw, so we''ve tried to stick to it.  But after vnics came along, 
psarc asked us (the nwam team) to figure out a way to at least co-exist
with them in phase 1; so we did that.

We''re on the verge of dev complete for nwam phase 1 (today or monday);
I''m loathe to add additional features at this point.  But I suppose if
it''s just a matter of adding an up-vlan call in net-nwam, it''s
hard to
argue against that.  I''ll investigate.

-renee

> That''s a vaild point.  I suppose when you get down to it the
answer
 > is historical: we''ve always lumped vlans and aggrs together, and
 > because of the extra work involved with aggrs, we just said support
 > for those would come later.  As you know, we''ve had an
exceptionally
 > hard time drawing boundaries around what nwam will and won''t do,
and
 > when each feature will be added; that was one clear line that we could
 > draw, so we''ve tried to stick to it.  But after vnics came along,
 > psarc asked us (the nwam team) to figure out a way to at least co-exist
 > with them in phase 1; so we did that.
 > 
 > We''re on the verge of dev complete for nwam phase 1 (today or
monday);
 > I''m loathe to add additional features at this point.  But I
suppose if
 > it''s just a matter of adding an up-vlan call in net-nwam,
it''s hard to
 > argue against that.  I''ll investigate.

Sounds reasonable.  To be clear, my goal was just to understand the
problem space, not add more work.

-- 
meem