<
d1b2>
<david.rysk> In my ci-work branch I’ve replaced relative with absolute paths for some of the gitmodules
<
d1b2>
<azonenberg> Yeah we decided a while back relative paths made more sense
<
d1b2>
<azonenberg> most folks clone / push via SSH since thats more compatible with a CLI workflow than HTTPS
<
d1b2>
<azonenberg> but you cant put ssh urls in the gitmodules without making anonymous clone impossible
<
d1b2>
<azonenberg> relative paths means if you fork you have to fork the other submodules but that seemed a cost worth paying
<
d1b2>
<david.rysk> I was only using the absolute paths for my CI testing
<
d1b2>
<david.rysk> I wasn't going to upstream those changes
<
d1b2>
<azonenberg> ah ok
Degi has quit [Ping timeout: 246 seconds]
Degi has joined #scopehal
<
d1b2>
<johnsel> Ping
<
d1b2>
<azonenberg> On it
<
d1b2>
<johnsel> Also thanks, found a Debian guide with unspecified name of driver and just went with that
<
d1b2>
<azonenberg> can you shut down your bouncer and the orchestrator? or am i good to hard stop them?
<
d1b2>
<johnsel> Please stop yourself via the stop button which will induce a clean shutdown by the xoa tools on the vm
<
d1b2>
<johnsel> I am in bed sleeping, actually woke up just for this
<
d1b2>
<johnsel> I actually think we should be able to do both, I’ll look at that tomorrow
azonenberg has quit [Ping timeout: 240 seconds]
azonenberg has joined #scopehal
<
d1b2>
<johnsel> @azonenberg is it possible you have the GPU assigned to one of your VMs now?
<
d1b2>
<johnsel> (the linux box has it's gpu back but Windows won't start, nor will another new vm from the template)
<
azonenberg>
Looking
<
azonenberg>
not seeing any other vms with a gpu
<
azonenberg>
does windows start if linux is off?
<
d1b2>
<johnsel> It may not be visible from xoa since we did some cli stuff
<
d1b2>
<johnsel> /assigned
<
azonenberg>
yes want me to remove it and see if it boots without?
<
d1b2>
<johnsel> sure
<
azonenberg>
the 296... VM no longer shows a GPU
<
d1b2>
<azonenberg> still wont start and it's showing the pci id of the GPU
<
d1b2>
<azonenberg> conjecture: the template is not "pass through a GPU"
<
d1b2>
<johnsel> interesting
<
d1b2>
<azonenberg> it's "pass through this specific GPU"
<
d1b2>
<azonenberg> you might have to make the windows template while the linux builder is active with a gpu attached
<
d1b2>
<azonenberg> to ensure you dont try to allocate that one gpu to both isntances
<
d1b2>
<johnsel> I don't think it would work that way
<
azonenberg>
Me neither
<
d1b2>
<johnsel> try re-assigning passthrough to the vm
<
azonenberg>
just grasping at straws lol
<
d1b2>
<johnsel> the first part isn't that bad of an idea
<
d1b2>
<johnsel> but it should just fix itself on re-association now
<
azonenberg>
It's saying no hosts available now
<
d1b2>
<johnsel> right
<
azonenberg>
let me explicitly bind it to the one server and figure out if that sheds any light
<
d1b2>
<johnsel> it did complain just now
<
d1b2>
<johnsel> let me deploy one fresh
<
d1b2>
<johnsel> it's no problem starting fresh again since there's not much on this one anyway
<
d1b2>
<johnsel> no hosts available
<
d1b2>
<johnsel> blegh
<
d1b2>
<johnsel> let me try one without a GPU in the template and we can try assigning one then
<
azonenberg>
That one seems to start OK
<
d1b2>
<johnsel> after assigning a gpu?
<
azonenberg>
no i wanted a baseline sans gpu t ostart
<
d1b2>
<johnsel> oh I wanted not to start it since it kicks in setup afterinstall
<
d1b2>
<johnsel> let me make a new one
<
d1b2>
<johnsel> I don't want to end up dealing with filesystem corruption once we have it working lol
<
d1b2>
<johnsel> and the shutdown without management agent is just a hard kill while the windows setup is definitely still active
<
azonenberg>
"no hosts available" as soon as i put the gpu on it
<
d1b2>
<johnsel> lame
<
azonenberg>
So yeah lets just go back to lavapipe on the windows vm
<
azonenberg>
if we have gpu on linux working, at least, we can catch some hardware specific issues there?
<
d1b2>
<johnsel> we could still try to remove them from xoa and assign them by id
<
azonenberg>
yeah but this has cost us so much time debugging already
<
azonenberg>
i just want to get working CI and working vulkan
<
azonenberg>
if we can do this without a dGPU, great
<
d1b2>
<johnsel> I mean it's a 15 minute job to try this final thing. We're throwing away a huge amount work that I can do all over again as well
<
d1b2>
<johnsel> or rather if we don't move to docker either on linux then it doesn't really matter
<
azonenberg>
yeah if we have one gpu shared by all the linux builder vms that would be nice
<
azonenberg>
all the linux builder instances i mean
<
azonenberg>
under one outer vm
<
_whitenotifier-c>
[scopehal-apps] azonenberg ba26b4c - Session: fixed bug where Clear() did not delete markers
<
_whitenotifier-3>
[scopehal] azonenberg 18c5cc3 - Unit: don't print unnecessary space after hexadecimal numbers, since there's no unit suffix
<
_whitenotifier-3>
[scopehal-apps] azonenberg b277de2 - Updated submodules
<
_whitenotifier-3>
[scopehal-apps] azonenberg 87915b0 - Fixed Marker::operator<() not being marked const (broke build on macos)