should be able to set default arch for unknown hardware

Bug #1273382 reported by Scott Moser on 2014-01-27
12
This bug affects 2 people
Affects Status Importance Assigned to Milestone
MAAS
Wishlist
Unassigned

Bug Description

if a machine tries to network boot from maas, and maas cannot determine its architecture, then maas sends back payload for i386.

I'd like to be able to set that hard coded default.

Ie, if all systems in a lab that will ever be network booted are of one arch, I should be able to just tell maas that.

Rather than falling back to 'i386/generic' (arch/subarch), I'd like to tell my the cluster controller that the default should be 'armhf/highbank'.

Maas's grand plan is that the hardware should identify itself via the pxe request (falling back to pxelinux/<arch> or some such). Thats wonderful if your bios netboot implementation does that, but if it does not, then you're SOL.

I'd rather just tell maas:
 Any hardware you don't know about, expect it is $ARCH/$SUBARCH.

Julian Edwards (julian-edwards) wrote :

Can you give a concrete reason for wanting this behaviour? I see a lot of "should" and "I'd like" but no actual use case :)

We believe that MAAS will be able to identify all hardware types properly by the time we have finished the next round of hardware enablement infrastructure so it would make this moot.

Changed in maas:
status: New → Incomplete
Gavin Panella (allenap) wrote :

Additionally, I'd like it to be considered a bug where MAAS cannot determine the hardware type before first boot. Where it really isn't possible to detect ahead of time, we should demand the manual selection of an arch (well, driver, which covers more than just arch) before we accept a machine into MAAS.

Scott Moser (smoser) on 2014-02-19
Changed in maas:
status: Incomplete → New
Scott Moser (smoser) wrote :

"We believe that MAAS will be able to identify all hardware types properly by the time we have finished the next round of hardware enablement infrastructure so it would make this moot."

I'm willing to make a considerable wager that you're wrong.

There will never be a case when maas is able to correctly enumerate all hardware systems prior to their existence. That is simply unrealistic, and would require that *all* hardware manufactures hold hardware release until Canonical says "it looks good".

I'm just requesting a simple setting, that already exists.
Right now, if maas gets a pxe boot request for a MAC address it doesn't know, it sends i386/generic.
I'm just saying you should let me configure it to send $ARCH/$SUBARCH.

Changed in maas:
status: New → Triaged
importance: Undecided → Wishlist

On Wednesday 19 Feb 2014 18:09:33 you wrote:
> "We believe that MAAS will be able to identify all hardware types
> properly by the time we have finished the next round of hardware
> enablement infrastructure so it would make this moot."
>
> I'm willing to make a considerable wager that you're wrong.
>
> There will never be a case when maas is able to correctly enumerate all
> hardware systems prior to their existence. That is simply unrealistic,
> and would require that *all* hardware manufactures hold hardware release
> until Canonical says "it looks good".

Obviously I mean "where you have a hardware driver" in place. Enumerating all
hardware is patently silly :)

> I'm just requesting a simple setting, that already exists.
> Right now, if maas gets a pxe boot request for a MAC address it doesn't
> know, it sends i386/generic. I'm just saying you should let me configure it
> to send $ARCH/$SUBARCH.

We'll come back to this when the driver work is done after our sprint in
March.

Scott Moser (smoser) wrote :

As far as I can tell, this is also a problem with subarch.
For enlisting or commissionging systems, I may very well need to use amd64/hwe-v but maas will boot amd64/hwe-t (latest lts with GA kernel).

Changed in maas:
milestone: none → 1.9.0
Jeff Lane (bladernr) wrote :

Please bump this higher than wishlist. We believe this is what is blocking the Power8 certs with PXE issues.

Because we are unable to default to HWE-U or later for the Power8 systems, PXE booting new P8 hardware is dodgy at best, sometimes it works, most times it seems to kernel panic when trying to enlist.

To work around this, we have replaced the default Power8 boot files with HWE-U boot files. We should be able to, at the least, be able to set the default kernel series for unknown hardware.

So while I don't think the initial problem is actually an issue here, the P8s do seem to be trying to boot, Scott's comment #5 does seem to apply to certification.

tags: added: blocks-hwcert-server
Jeff Lane (bladernr) wrote :

as Mike points out, it actually makes a lot of sense to also have MAAS just default to the latest HWE kernel for enlistment and commissioning.

MAAS should be, IMO, smart enough to see that Trusty has a HWE-U kernel and boot media and start using that for enlistment and commissioning to ensure that newly discovered systems are enlisted and commissioned properly (all their hardware is addressed by the right drivers and probed successfully).

Setting a default kernel as noted in previous comments is also a helpful thing to have, but really, MAAS should be defaulting to whatever is the most recent kernel it has for a given release.p

Gavin Panella (allenap) on 2015-10-09
Changed in maas:
importance: Wishlist → Critical
Andres Rodriguez (andreserl) wrote :

MAAS 1.9 will have the ability of selecting the HWE kernel to be used for enlistment/commissioning, but it won't have the ability to select the architecture. This should solve the problems being experienced, and as such, I'm marking this as a wishlist for1.9, although this won't be fixed.

Changed in maas:
importance: Critical → Wishlist
Jeff Lane (bladernr) on 2016-01-26
tags: added: hwcert-server
removed: blocks-hwcert-server
Changed in maas:
status: Triaged → Fix Committed
Changed in maas:
status: Fix Committed → Fix Released
To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers