2:33 p.m.
RDO Manager does a lot more than packstack. Too much, for many use
cases, which is why we have people continuing to use packstack. Where do
people use packstack?
1. Demos
2. Development Work
3. CI
I'm sure there are others, but lets start there. I personally need
support for use cases 1 and 2. Can we get to the point where rdo-manager
can support these use cases?
There are, as I can see it, two potential approaches that could make
this work:
1. Treat the undercloud deployment as an all-in-one, and not deploy the
overcloud.
2. Have rdo-manager assume that the undercloud is already deployed, and
only deploy the overcloud.
Are there others?
I think that 2 is an interesting option? What if ... we took something
like the next gen oslab and put only an rdo-manager undercloud out on
it? Then...allowed multiple people to run rdo-manager --overcloud
against it, perhaps each with their own slice?
I'm already deploying OpenStack on OpenStack. If most of the business
logic/value add comes from the overcloud, can we split the deployment
along these lines?
4:25 p.m.
On 08/05/2015 01:33 PM, Adam Young wrote:
RDO Manager does a lot more than packstack. Too much, for many use
cases, which is why we have people continuing to use packstack. Where
do people use packstack?
1. Demos
2. Development Work
3. CI
I'm sure there are others, but lets start there. I personally need
support for use cases 1 and 2. Can we get to the point where
rdo-manager can support these use cases?
There are, as I can see it, two potential approaches that could make
this work:
1. Treat the undercloud deployment as an all-in-one, and not deploy
the overcloud.
2. Have rdo-manager assume that the undercloud is already deployed,
and only deploy the overcloud.
Are there others?
I think that 2 is an interesting option? What if ... we took
something like the next gen oslab and put only an rdo-manager
undercloud out on it? Then...allowed multiple people to run
rdo-manager --overcloud against it, perhaps each with their own slice?
I'm already deploying OpenStack on OpenStack. If most of the
business logic/value add comes from the overcloud, can we split the
deployment along these lines?
+1
Followup question - how do RDO Manager developers develop/test RDO
manager? Do you require access to a lab with several bare metal machines?
_______________________________________________
Rdo-list mailing list
Rdo-list(a)redhat.com
https://www.redhat.com/mailman/listinfo/rdo-list
To unsubscribe: rdo-list-unsubscribe(a)redhat.com
8:07 p.m.
Hi Adam,
Revising this thread with my own reasons...
On 08/05/2015 03:33 PM, Adam Young wrote:
RDO Manager does a lot more than packstack. Too much, for many use
cases, which is why we have people continuing to use packstack. Where do
people use packstack?
1. Demos
2. Development Work
3. CI
Installing OpenStack on low-end hardware (1 NIC, no IPMI) or in VMs with
nested virt. Personally, I have some NUCs for a personal OpenStack
cluster. I also have access to an internal OpenStack cloud and I want to
test ODL for Neutron on that.
Ultra simple deployment architecture (no HA, one controller, N compute
nodes, or all-in-one on a home server).
Personally, I'm OK manually powering on a host (when prompted) to PXE
boot an undercloud host. I don't know if it's even possible, but if we
could detect that hosts have only one NIC, using the same NIC for
management, data and provisioning, that would be awesome.
Thanks,
Dave.
--
Dave Neary - NFV/SDN Community Strategy
Open Source and Standards, Red Hat - http://community.redhat.com
Ph: +1-978-399-2182 / Cell: +1-978-799-3338
9:50 p.m.
On 08/14/2015 09:07 PM, Dave Neary wrote:
Hi Adam,
Revising this thread with my own reasons...
On 08/05/2015 03:33 PM, Adam Young wrote:
> RDO Manager does a lot more than packstack. Too much, for many use
> cases, which is why we have people continuing to use packstack. Where do
> people use packstack?
>
> 1. Demos
> 2. Development Work
> 3. CI
Installing OpenStack on low-end hardware (1 NIC, no IPMI)
I'm guessing IPMI is a requirement of the Ironic driver.
or in VMs with nested virt.
This would work with the overcloud approach.
Personally, I have some NUCs for a personal OpenStack cluster. I also
have access to an internal OpenStack cloud and I want to test ODL for
Neutron on that.
What it seem like we want is to get the capabilities of the Undercloud
installable on a single Fedora box, and then be able to install the
Overcloud on top of that. An "undercloud light" based on libvirt?
Ultra simple deployment architecture (no HA, one controller, N compute
nodes, or all-in-one on a home server).
Personally, I'm OK manually powering on a host (when prompted) to PXE
boot an undercloud host. I don't know if it's even possible, but if we
could detect that hosts have only one NIC, using the same NIC for
management, data and provisioning, that would be awesome.
Thanks,
Dave.
9:56 p.m.
On Sat, Aug 15, 2015 at 12:50 PM, Adam Young <ayoung(a)redhat.com> wrote:
On 08/14/2015 09:07 PM, Dave Neary wrote:
> Hi Adam,
>
> Revising this thread with my own reasons...
>
> On 08/05/2015 03:33 PM, Adam Young wrote:
>
>> RDO Manager does a lot more than packstack. Too much, for many use
>> cases, which is why we have people continuing to use packstack. Where do
>> people use packstack?
>>
>> 1. Demos
>> 2. Development Work
>> 3. CI
>>
>
> Installing OpenStack on low-end hardware (1 NIC, no IPMI)
>
I'm guessing IPMI is a requirement of the Ironic driver.
or in VMs with nested virt.
>
This would work with the overcloud approach.
Personally, I have some NUCs for a personal OpenStack cluster. I also have
> access to an internal OpenStack cloud and I want to test ODL for Neutron on
> that.
>
What it seem like we want is to get the capabilities of the Undercloud
installable on a single Fedora box, and then be able to install the
Overcloud on top of that. An "undercloud light" based on libvirt?
Id plainly like to see a breakdown of how this compares to say FUEL, or HP
Helion, or IBMs openstack deployments
Though Id also like to see someone get RDO to work with XENServer
deployments also, seems RDO is pretty much kvm based only
> Ultra simple deployment architecture (no HA, one controller, N compute
> nodes, or all-in-one on a home server).
>
> Personally, I'm OK manually powering on a host (when prompted) to PXE
> boot an undercloud host. I don't know if it's even possible, but if we
> could detect that hosts have only one NIC, using the same NIC for
> management, data and provisioning, that would be awesome.
>
> Thanks,
> Dave.
>
>
_______________________________________________
Rdo-list mailing list
Rdo-list(a)redhat.com
https://www.redhat.com/mailman/listinfo/rdo-list
To unsubscribe: rdo-list-unsubscribe(a)redhat.com
5:04 p.m.
Hi,
On 08/14/2015 10:56 PM, Outback Dingo wrote:
Id plainly like to see a breakdown of how this compares to say FUEL,
or
HP Helion, or IBMs openstack deployments
Though Id also like to see someone get RDO to work with XENServer
deployments also, seems RDO is pretty much kvm based only
While any compute nodes deployed by RDO would be KVM nodes, I don't see
any reason that someone couldn't add Hyper-V, ESX or Xen nodes to Nova's
compute pool with RDO - the bits are the same as upstream Nova, and of
course whether such a configuration is supported is not an issue.
In fact, I'm pretty sure that CERN deploys Hyper-V and KVM nodes using
RDO. I don't see any reason why someone couldn't do the same with Xen,
and document any issues they have on rdoproject.org
Thanks,
Dave.
--
Dave Neary - NFV/SDN Community Strategy
Open Source and Standards, Red Hat - http://community.redhat.com
Ph: +1-978-399-2182 / Cell: +1-978-799-3338
10:06 a.m.
On Sun, Aug 16, 2015 at 06:04:57PM -0400, Dave Neary wrote:
Hi,
On 08/14/2015 10:56 PM, Outback Dingo wrote:
>Id plainly like to see a breakdown of how this compares to say FUEL, or
>HP Helion, or IBMs openstack deployments
>Though Id also like to see someone get RDO to work with XENServer
>deployments also, seems RDO is pretty much kvm based only
While any compute nodes deployed by RDO would be KVM nodes, I don't see any
reason that someone couldn't add Hyper-V, ESX or Xen nodes to Nova's compute
pool with RDO - the bits are the same as upstream Nova, and of course
whether such a configuration is supported is not an issue.
In fact, I'm pretty sure that CERN deploys Hyper-V and KVM nodes using RDO.
I don't see any reason why someone couldn't do the same with Xen, and
document any issues they have on rdoproject.org
Thanks,
Dave.
For what it's worth, Ironic supports all kinds of interesting power
drivers, way more than just IPMI. DRAC, ILO, AMT, ssh (for virt), WOL (a
super-simple wake-on-lan driver that only knows how to power machines
on), etc. cc-ing Lucas for more info.
--Hugh
--
Dave Neary - NFV/SDN Community Strategy
Open Source and Standards, Red Hat - http://community.redhat.com
Ph: +1-978-399-2182 / Cell: +1-978-799-3338
_______________________________________________
Rdo-list mailing list
Rdo-list(a)redhat.com
https://www.redhat.com/mailman/listinfo/rdo-list
To unsubscribe: rdo-list-unsubscribe(a)redhat.com
--
== Hugh Brock, hbrock(a)redhat.com ==
== Senior Engineering Manager, Cloud Engineering ==
== RDO Manager: Install, configure, and scale OpenStack ==
== http://rdoproject.org ==
"I know that you believe you understand what you think I said, but I’m
not sure you realize that what you heard is not what I meant."
--Robert McCloskey
11:20 a.m.
On 08/17/2015 11:06 AM, Hugh O. Brock wrote:
On Sun, Aug 16, 2015 at 06:04:57PM -0400, Dave Neary wrote:
> Hi,
>
> On 08/14/2015 10:56 PM, Outback Dingo wrote:
>> Id plainly like to see a breakdown of how this compares to say FUEL, or
>> HP Helion, or IBMs openstack deployments
>> Though Id also like to see someone get RDO to work with XENServer
>> deployments also, seems RDO is pretty much kvm based only
> While any compute nodes deployed by RDO would be KVM nodes, I don't see any
> reason that someone couldn't add Hyper-V, ESX or Xen nodes to Nova's compute
> pool with RDO - the bits are the same as upstream Nova, and of course
> whether such a configuration is supported is not an issue.
>
> In fact, I'm pretty sure that CERN deploys Hyper-V and KVM nodes using RDO.
> I don't see any reason why someone couldn't do the same with Xen, and
> document any issues they have on rdoproject.org
>
> Thanks,
> Dave.
For what it's worth, Ironic supports all kinds of interesting power
drivers, way more than just IPMI. DRAC, ILO, AMT,
I'm Guessing this is really what we want; an undercloud deployment
using ssh?
ssh (for virt),
Say I have a RHEL7 laptop. Could I install an undercloud deployment on
it, using the ironic driver as the target for my development, and then
install an overcloud on top of that?
WOL (a
super-simple wake-on-lan driver that only knows how to power machines
on), etc. cc-ing Lucas for more info.
--Hugh
> --
> Dave Neary - NFV/SDN Community Strategy
> Open Source and Standards, Red Hat - http://community.redhat.com
> Ph: +1-978-399-2182 / Cell: +1-978-799-3338
>
> _______________________________________________
> Rdo-list mailing list
> Rdo-list(a)redhat.com
> https://www.redhat.com/mailman/listinfo/rdo-list
>
> To unsubscribe: rdo-list-unsubscribe(a)redhat.com
12:27 p.m.
-----Original Message-----
From: rdo-list-bounces(a)redhat.com [mailto:rdo-list-bounces@redhat.com]
On Behalf Of Adam Young
Sent: 17 August 2015 18:20
To: Hugh O. Brock <hbrock(a)redhat.com>; Dave Neary
<dneary(a)redhat.com>
Cc: rdo-list(a)redhat.com
Subject: Re: [Rdo-list] Packstack and rdo-manager, looking forward.
On 08/17/2015 11:06 AM, Hugh O. Brock wrote:
> On Sun, Aug 16, 2015 at 06:04:57PM -0400, Dave Neary wrote:
>> Hi,
>>
>> On 08/14/2015 10:56 PM, Outback Dingo wrote:
>>> Id plainly like to see a breakdown of how this compares to say FUEL,
>>> or HP Helion, or IBMs openstack deployments Though Id also like to
>>> see someone get RDO to work with XENServer deployments also,
seems
>>> RDO is pretty much kvm based only
>> While any compute nodes deployed by RDO would be KVM nodes, I
don't
>> see any reason that someone couldn't add Hyper-V, ESX or Xen nodes to
>> Nova's compute pool with RDO - the bits are the same as upstream
>> Nova, and of course whether such a configuration is supported is not an
issue.
>>
>> In fact, I'm pretty sure that CERN deploys Hyper-V and KVM nodes using
RDO.
CERN uses the RDO packages as the basis for the deployment to the 4,000 or soc Linux
servers. We use the cloudbase Windows packages for Hyper-V (a few hundred).
These are then glued together using the OpenStack puppet modules on both Windows and
Linux.
We do use packstack for our R&D work. It provides a really simple way of bringing up a
small OpenStack set up for newcomers in a few virtual machines to experiment or for us to
evaluate if a backport of a recent feature is useful. This sort of case is shown by the
RDO test days of new releases.
I have concerns that the rdo-manager model does not seem to address this packstack use
case (unless the undercloud is also deployed using rdo-manager). Using devstack for these
deployments would make it a much larger transition to a Puppet/RDO packaged based
configuration for production.
Tim
>> I don't see any reason why someone couldn't do the
same with Xen, and
>> document any issues they have on rdoproject.org
>>
>> Thanks,
>> Dave.
...
3375
days inactive
3392
days old
8 comments
6 participants
participants (6)
-
Adam Young -
Dave Neary -
Hugh O. Brock -
Outback Dingo -
Rich Megginson -
Tim Bell