Matrix of possible slave nodes role combinations

[Pavel Vaylov]

Matrix with possible slave nodes role combinations should be in the Fuel\MOS documentation.

Customers frequently ask which roles can be or can not be combined and why.

We should put a matrix with possible combinations to the docs.
Also we should describe why roles can or can not be combined.

[Meg McRoberts (dreidellhasa)]

We do have http://docs.mirantis.com/openstack/fuel/master/planning-guide.html#node-server-hardware-recommendations.
It's a very tiny little piece and needs to be improved, but how much do we do here?

Some general information is availablle in the Community docs:

http://docs.openstack.org/grizzly/openstack-compute/install/yum/content/compute-system-requirements.html
http://docs.openstack.org/trunk/install-guide/install/yum/content/object-storage-system-requirements.html

Are these recommendations appropriate for MOSt? If so, should we replicate this information or just point to it?

We also have some information about calculating ones hardware needs:
http://docs.mirantis.com/openstack/fuel/master/planning-guide.html#calculate-hardware-requirements
Is this information still good or does it need to be updated?

We began to address the question of storage requirements for Ceilometer metering in:
http://docs.mirantis.com/openstack/fuel/master/planning-guide.html#ceilometer-and-mongodb

At one time, we had decided that the Hardware Caclulator was the better way to handle this material
(https://www.mirantis.com/openstack-services/bom-calculator/) but it sounds like we are not maintaining
this to keep up with new hardware products and advances in our software

[Meg McRoberts (dreidellhasa)]

We have some of this covered in:

http://docs.mirantis.com/openstack/fuel/master/planning-guide.html#nodes-and-roles
http://docs.mirantis.com/openstack/fuel/master/reference-architecture.html#nodes-roles-arch

We can strengthen these sections. I'm not sure that a matrix is the right approach for a subject
that has so much nuance, however. For example, we say that the Controller and Compute nodes
must always be on separate servers, except when running vCenter, in which case the Compute role
must be on the same server as the Controller. MongoDB can actually run on the Controller node as
long as you do infrequent samplings of small amounts of metering data. If you sample a lot of stuff
and sample frequently, the I/O load may be enough to bring down the whole environment.

We also need to consider disk partitioning when combining multiple roles on one server. By default,
Fuel just partitions the disk to give each role an equal portion of the available disk space.

Let me think about this and, of course, if you have specific suggestions about the docs referenced above,
please let me know.

[Dmitry Borodaenko (angdraug)]

"> By default, Fuel just partitions the disk to give each role an equal portion of the available disk space."

Except the case for Ceph OSD: Fuel will try to allocate whole drives to this partition type.

[Meg McRoberts (dreidellhasa)]

What if I put Ceph OSD and Compute roles on one node?
I know, not a great idea but I think it's allowed. When I put
Cinder LVM and Compute roles on one node, Fuel split the
available disk space evenly.

I'll try it out later...

On Fri, Sep 26, 2014 at 9:39 AM, Dmitry Borodaenko <<email address hidden>
> wrote:

> "> By default, Fuel just partitions the disk to give each role an equal
> portion of the available disk space."
>
> Except the case for Ceph OSD: Fuel will try to allocate whole drives to
> this partition type.
>
> --
> You received this bug notification because you are a member of Fuel
> Documentation Team, which is a bug assignee.
> https://bugs.launchpad.net/bugs/1371626
>
> Title:
> Matrix of possible slave nodes role combinations
>
> Status in Fuel: OpenStack installer that works:
> Confirmed
>
> Bug description:
> Matrix with possible slave nodes role combinations should be in the
> Fuel\MOS documentation.
>
> Customers frequently ask which roles can be or can not be combined and
> why.
>
> We should put a matrix with possible combinations to the docs.
> Also we should describe why roles can or can not be combined.
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/fuel/+bug/1371626/+subscriptions
>

[Fabrizio Soppelsa (fsoppelsa)]

Meg, if you put Compute + Ceph on the same node, in presence of other disks, Fuel tries to partition those other disks as Ceph by default.

[Fabrizio Soppelsa (fsoppelsa)]

I propose to just finish summarizing node roles in the Terminology Reference (see https://bugs.launchpad.net/fuel/+bug/1383240), and switch this bug to Won't Fix.

[Dmitry Borodaenko (angdraug)]

No, I think it's the other way around: #1383240 is a newer bug, so I marked *that* as a duplicate of this bug. Lets finish the discussion of how we clarify roles relationships here and not spread it over multiple bugs.

[Fabrizio Soppelsa (fsoppelsa)]

I would write in the Terminology Reference, for each role, "Finally, this can be combined with Telemetry-Mongo role" or "Finally, this can be combined with Cinder LVM storage role"... I wouldn't add ulterior sparse lists around (User Guide?) to increase complexity.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/135156

[Dmitry Borodaenko (angdraug)]

Patch is still in review, postponing to 6.1.

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/135156
Committed: https://git.openstack.org/cgit/stackforge/fuel-docs/commit/?id=02d728fe72254e0ead115061d424e2b6c0a6564b
Submitter: Jenkins
Branch: master

commit 02d728fe72254e0ead115061d424e2b6c0a6564b
Author: Meg McRoberts <email address hidden>
Date: Mon Nov 17 19:30:45 2014 -0800

    Nodes and roles -- terminology articles

    Create/improve articles for Nodes, Roles, plus one for each
    Role that Fuel recognizes.

    We have requests for more detailed information about how Fuel
    manages nodes and we also have requests for more accessible
    information about which roles can and cannot be combined. This
    CR does not completely solve the second request but it's a start.

    Add ref-arch/logical-setup/0800-node-internals article to give
    detailed information about the information Nailgun stores for nodes.

    Added tags to ref-arch/logical-setup files so they can be xref'ed

    A couple unrelated terminology articles were breaking the build
    so I had to add vertical spaces; this is why they are part of this CR.

    Change-Id: I76551658e975e2d910686df26c371be142238cc1
    Closes-Bug: 1371626
    Closes-Bug: 1385141