Comment 5 for bug 1665459

Same network.

Nodes can talk to either MaaS device on this trunk over the untagged
default net. All MaaS hosts and PXE hosts are on the same net (per the
diagram.) The rack controller handles DHCP and PXE, then I have to assume,
also provides the iSCSi target to the PXE node, or we wouldn't be syncing
images to the rack controllers right?

My understanding was that only the rack controller needed to be able to
reach the region controller for management functions, but all nodes can hit
both in this scenario regardless.

Here is the PXE process now (from the screen shots I attached.)

1. New node starts and makes the PXE call. The DHCP call hits the egress
gateway router (192.168.199.1) which the sends the request to the DHCP
Forwarded Address of the MaaS rack controller on the same untagged network
(192.168.199.6) because it is the device providing DHCP for hosts that are
not already statically assigned.

2. MaaS rack controller accepts the request, assigns an IP address from
the dynamic pool, and starts the PXE process.

3. The PXE node completes the cloud-init step, reaches the "Reached Target
Network." step, and starts the image download process.

4. PXE nodes that do not already have a machine entry in MaaS for their
MAC address attempt to download the image from:

"http://169.254.169.254/2009-04-04/meta-data/instance-id"

region controller: 192.168.199.2
rack controller: 192.168.199.6
someone's link local address: 169.254.169.254

per the image this connection fails until PXE exits.

5. Manually adding a machine entry to the region controller seems to
change this behavior, causing the call to hit the default gateway instead
of whatever machine's link local metadata address it was searching for
before (second image provided) and likewise fails because the egress router
does not host this metadata service.

Does this make sense at all?

On Feb 16, 2017 4:50 PM, "Blake Rouse" <email address hidden> wrote:

So the issue here is that the deploying node want to talk to the region
controller over the same IP address that the rack controller talks to
the region controller. Based on your report your saying that the rack
controller has two interfaces? One to talk to the region and another for
the machines to boot from?

MAAS architecture is to have the deploying machines to be able to
directly talk to the region controller. There is a way to configure your
systems to perform in this manner, but I want to be sure that this is
the problem you are experiencing?

--
You received this bug notification because you are subscribed to the bug
report.
https://bugs.launchpad.net/bugs/1665459

Title:
  MaaS devel bad request routing on PXE

To manage notifications about this bug go to:
https://bugs.launchpad.net/maas/+bug/1665459/+subscriptions