On 31 May 2016 at 16:45, Neil Williams neil.williams@linaro.org wrote:
On 31 May 2016 at 16:04, Christian Ziethén christian.ziethen@linaro.org wrote:
---------- Forwarded message ---------- From: Steve McIntyre steve.mcintyre@linaro.org Date: 31 May 2016 at 16:46 Subject: Re: pipeline vland help To: Christian Ziethén christian.ziethen@linaro.org Cc: neil.williams@linaro.org
On Tue, May 31, 2016 at 03:29:45PM +0200, Christian Ziethén wrote:
Hi,
Hi Christian,
You'd be much better asking on the lava-users mailing list rather than just directly to me, I'll be honest.
Been struggling with setting up a vland between two arndale targets. I have managed to create a multinode job that uses the pipeline model: https://lng.validation.linaro.org/scheduler/job/9702/multinode_definition Have also managed to create valid yaml that seems to conform to the code in lava-server/lava_scheduler_app/schema.py https://lng.validation.linaro.org/scheduler/job/9743/multinode_definition This one does however not do anything after being submitted, I tried putting 100M in the tags for the vland, I also tried requiring that the arndale targets in the multinode protocol had the 100M tag, but that didn't work. According to the device dictionary for lng-arndale-01, it should have a 100M tag on iface1.
Hmmm, OK. I can see that it's sat waiting for devices to be assigned. Looking at the definition, you've got one group (role) defined with 2 entries. I believe that for VLANd jobs you need to have individual roles for each machine. Neil can confirm.
The job is fine - it has a count of 1 for each role, so it does have individual roles.
I've found the scheduler bug - I've had to cancel the job. There's a bug in the tags handling. For now (until I fix this bug), you'll need to specify all - and exactly all - of the interface tags in your job submission.
https://lng.validation.linaro.org/scheduler/job/9790/multinode_definition
Currently, that then fails to start, so I'll need to work on this support.
Also have this job (v1): https://lng.validation.linaro.org/scheduler/job/9118/multinode_definition Which runs my test using iface1 (I think) but it doesn't use vland.
Right, v1 doesn't do vland.
I am unsure how to debug this.
It's a scheduling bug, specific to VLANd support (and therefore V2). I'm looking into it using the beaglebone-black test job - a standard multinode bbb job runs, adding the vland requirements and it fails to schedule.
There's nothing that can be done at your end until I've worked out where it is getting blocked.
The 2016.6 release is due shortly, so I'll get a fix for this into that release. Sorry for the delay.
--
Neil Williams
neil.williams@linaro.org http://www.linux.codehelp.co.uk/