@davidrohr I assume the CUDA error from build/O2/fullCI is known and being resolved ? (and is unrelated to this PR)

@TimoWilken : The CUDA problem had been fixed by alisw/ali-bot@652ab06 yesterday. Do you have any idea why it fails now in the O2 CI? Is it possible that this change was not picked up by all builders? Or was something changed wrt. the GPU CI container?

@davidrohr I hadn't restarted the builders to avoid delays, so they were still running in the old container (which has CUDA under /opt). I'll restart them this evening.

But this I don't understand. Yesterday I saw the problem that CUDA was not found, so I changed the location in the ali-bot script. But this is only possible if the builders were using the new container. So this means currently some are using the old and some are using the new one?

@davidrohr Hm, that's odd, one of the three workers has CUDA under /opt, the other two have it under /usr/local/bin. Maybe a docker pull failed somewhere. I'll restart the old one now.