Arvados

New image: packer-build-compute-image: #305 - ami-08602fed623133256

Test run: tordo-xvhdp-asvj8h83ke63sh0

Brett Smith wrote in #note-2:

New image: packer-build-compute-image: #305 - ami-08602fed623133256

This test is bad because it pinned the packages. We want to see the "Install <software> package pins" tasks skipped. Added one configuration line to Jenkins to do that.

Now trying packer-build-compute-image: #306 but it seems to be stuck waiting for an executor. I'm just gonna leave it for now, if there's no useful progress by morning I'll have to dig in.

Brett Smith wrote in #note-3:

Now trying packer-build-compute-image: #306

Package list attached. No major changes except maybe CUDA. WGS workflow: tordo-xvhdp-6nh5z156inxcdkr

Brett Smith wrote in #note-5:

Brett Smith wrote in #note-4:

Package list attached. No major changes except maybe CUDA. WGS workflow: tordo-xvhdp-6nh5z156inxcdkr

No really this time. The workflow failed but it seems like a workflow bug or possibly #22466. Changes:

cuda 12.5 → 12.6
nvidia-container 1.16 → 1.17

Do we have a way to actually test these? Do we want to?

That workflow doesn't use CUDA.

These are failing with "Cancelled after exceeding MaxDispatchAttempts" which makes me think it we're getting intermittently broken compute nodes.

Peter Amstutz wrote in #note-6:

Brett Smith wrote in #note-5:

cuda 12.5 → 12.6
nvidia-container 1.16 → 1.17

Do we have a way to actually test these? Do we want to?

That workflow doesn't use CUDA.

Yes I know, hence my questions.

These are failing with "Cancelled after exceeding MaxDispatchAttempts" which makes me think it we're getting intermittently broken compute nodes.

Given that the image build itself took over an hour to even get started I suspect there was some cloud weather last night. Let's just retry for starters: tordo-xvhdp-dtcx4b0fvc0gni0