[Starlingx-discuss] multi-node AIO / PXE booting

[Outback Dingo]

On Tue, May 10, 2022 at 6:48 PM Pratik M. <pvmpublic at gmail.com> wrote:
>
>
> On Mon, May 9, 2022 at 4:01 PM Outback Dingo <outbackdingo at gmail.com> wrote:
>
> > can i deploy a duplex aio on 2-3 nodes? and worker/storage on 4 nodes
>
> Duplex AIO = Controller (and storage and workers) on 2 nodes. There is no 3-controller option in StarlingX. So typically you would get 2 controllers and 6 workers. Pl. see:
> https://docs.starlingx.io/deploy/index-deploy-da06a98b83b1.html
>

so never more then 2 controllers.... i can pxe boot and manually add
all hosts from a central pxe server, since we cant use stx 169.254,
due to bonds and vlans.
that leaves me with what do i do for storage can a compute node be a
storage node, as in all 6 nodes being compute/storage (ceph), what
about controller-0 storage... meaning none except for boot drives.

ie...
control-1 2x128Gb boot ssd, 1 x 1TB, document states i need 2 500gb + drives
control-2 2x128Gb boot ssd, 1 x 1TB, document states i need 2 500gb + drives
compute-1 1x1TB boot drive, and 8x3.8TB drives
compute-2 1x1TB boot drive, and 8x3.8TB drives
compute-3 1x1TB boot drive, and 8x3.8TB drives
compute-4 1x1TB boot drive, and 8x3.8TB drives
compute-5 1x1TB boot drive, and 8x3.8TB drives
compute-6 1x1TB boot drive, and 8x3.8TB drives

not sure this scenerio is even viable without
controller-0/controller-1 having 0 ceph storage, as it would then all
be on the nodes





>
> > so questions:
> > 1) can i boot all nodes from pxe server?
>
> In the standard/typical deployment, only controller-0 is PXE booted from an "external" PXE server. Other hosts PXE boot off the controller-0 on the cluster-internal mgmt/cluster network. The second link you refer to, is about PXE booting the controller-0.
>
> But, the first link you refer to, allows PXE booting ALL nodes from an external network. I haven't tried this but it seems supported. For e.g.
> https://docs.starlingx.io/developer_resources/stx_ipv6_deployment.html
>
> > 2) after booting installing the controller-0 node i can then pxe boot
> > install controller-1 node
> > 3) in this scenerio how does controller-0 discover controller-1 so i
> > can set the host  "system host-update 2 personality=controller"
>
> In the typical deployment, once controller-0 is up, it runs a DHCP and TFTP server on the internal mgmt/cluster network. The controller-1 and other worker nodes are supposed to be configured to network boot off the NIC connected to this internal network. When they boot, controller-0 will "discover" them and add them to the host inventory.
>
> https://docs.starlingx.io/planning/kubernetes/starlingx-boot-sequence-considerations.html
> https://docs.starlingx.io/planning/kubernetes/internal-management-network-planning.html#internal-management-network-planning
>
> > 4) does the same apply to workers
>
> Yes
>
> >
> > 5) if you have 6 fully configured server each with 24TB and 512Gb
> > memory, how would you deploy to get a full 6 nodes with storage and
> > compute?
>
> You want more storage than controllers can provide? Pl see if this helps:
> https://docs.starlingx.io/planning/kubernetes/storage-planning-storage-resources.html
>
> Thanks