[13:06:32] <jayme>	 akosiaris: jhathaway: As advertised yesterday, I've hacked my way around ci/releases issues for now and the change is ready for review https://gerrit.wikimedia.org/r/c/operations/puppet/+/909687
[14:08:19] <jhathaway>	 thanks jayme 
[15:52:52] <elukey>	 very nice summary of how nvidia allows to share gpus - https://towardsdatascience.com/how-to-increase-gpu-utilization-in-kubernetes-with-nvidia-mps-e680d20c3181
[15:53:32] <elukey>	 the MPS functionality seems nice but the implementation of the device plugin is a little heavy (extra server running, sharing memory space with the container, etc..)
[16:12:16] <elukey>	 jayme: just ran the first test successfully on a GPU on DSE, thanks for all the support <3
[16:12:32] <jayme>	 elukey: sweet! Congrats
[16:12:59] <elukey>	 image size (uncompressed) is around 2/3 GB, not great but also not a horror like before
[16:13:39] <jayme>	 still huge, would have expected less - but hey :)
[16:22:15] <elukey>	 yeah I know but simply installing tensorflow from pip brings in a ton of things etc..
[17:33:55] <akosiaris>	 https://groups.google.com/a/kubernetes.io/g/dev/c/g8rwL-qnQhk/m/GHe0i6hLAQAJ
[17:34:18] <akosiaris>	 Apparently kubedevs are planning to kill cephfs support
[17:34:44] <akosiaris>	 We shouldn't rely on it (if we ever were planning to)
[17:47:38] <bd808>	 akosiaris: since I see you here I'm reminded that legoktm and I would very much like your blessing or rejection of T320848 (pip install for pygments instead of deb packaging)