11:01:52 #startmeeting scientific-sig 11:01:53 Meeting started Wed Jul 17 11:01:52 2019 UTC and is due to finish in 60 minutes. The chair is martial_. Information about MeetBot at http://wiki.debian.org/MeetBot. 11:01:54 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 11:01:56 The meeting name has been set to 'scientific_sig' 11:02:03 hi o/ 11:02:44 so this week, we have a couple items to go through followed by AOB 11:03:01 #topic Complete the OpenStack user survey and register scientific OpenStack deployments 11:03:17 hi Martial 11:03:37 if you have a running OpenStack at your site, we request that you fill the survey 11:03:40 we inadvertently did the survey twice (left hand meet right hand) and wondered if/how the results are deduped 11:03:49 #link https://www.openstack.org/user-survey/survey-2019/landing 11:03:59 evening 11:04:00 janders: hi :) 11:04:05 #chair b1airo 11:04:06 Current chairs: b1airo martial_ 11:04:33 dh3: I do not remember if the survey is related to our OpenStack ID 11:04:42 b1airo: hi 11:05:12 i think it is OpenStack ID related, or at least was... 11:05:22 just checked and it shows my name at the top, so yes it is 11:05:44 OpenStack User Survey 2019 "Step 1 of 5 for Martial Michel" 11:06:07 Therefore it updates your content 11:06:37 (ot at least there is a way to do de-duplication) 11:07:06 but survey away please :) 11:07:30 #topic CFP for OpenInfra days Rome 11:07:49 Perfect time zone for this one :) 11:08:05 #link https://openinfraday.it/call-for-paper/ 11:08:46 "OpenInfra Day Italy 2019 will provide a 360-degrees view of Open Infrastructure – from computing, networking, and storage, through hands-on workshops and sessions focussing on real-world stories, more and more focusing on container orchestration and microservices architecture." 11:09:24 two times, two locations: Milan october, 2nd 2019 – Rome october, 3rd 2019 11:10:18 #topic Shangai Summit and PTG 11:10:26 Adding this one to the list 11:10:57 The Shanghai summit CFP is over, it was extended 11:11:24 I hope everybody had a chance to submit their proposals 11:11:37 o/ 11:12:02 good luck, sadly I spent my "foreign conference" budget already this year 11:12:25 the Scientific SIG will be present with our usual two back to back sessions 11:12:49 b1airo: coming to the summit? 11:13:26 belmoreira: hopefully other opportunities will arise 11:13:42 The Scientific SIG has also requested a presence at the PTG 11:14:19 From 1/2 day to 1 day (filled the survey last night) 11:14:23 it was a great session in Denver. I'm not sure if I'll be in Shanghai at this stage, but would be very happy to do that again 11:15:30 janders: if you are able to join us, please do so; because our role is much more outreach we tend to go to follow up sessions to understand how to best use novel technologies for our scientific workloads 11:15:52 don't think i'll make it i'm afraid, already two trips to the States on the horizon before December 11:16:08 #topic SC19 11:16:41 following up on our conferences, we need to submit the BoF shortly 11:17:03 Kubecon is at the same time as SC19 11:17:25 which will you choose martial_ ? 11:17:28 I will be at SC19 11:17:29 :-) 11:17:55 My colleague Rion (who has been on a few of our panels) will be at Kubecon 11:17:58 Kubecon not Docker-y enough for you ;-) 11:18:27 We are going to propose a BoF on cloud federation 11:18:41 (also) 11:19:05 as such my colleague Chris is the person that will be with us (if welcome) 11:19:18 we started this conversation on the Scientific SIG Slack 11:19:37 #topic Cloud Federation 11:19:45 ok, interesting, do you have a list of people interested in the federation bof yet? 11:20:20 am going to make this a regular topic for the next couple months :) 11:21:08 The work I have been participating on “Cloud Federation” is officially in public comment phase for publication as a NIST SP500 11:21:33 #link https://www.nist.gov/programs-projects/nist-cloud-computing-program-nccp 11:22:01 it is open until September 11:22:16 (open for review that is) 11:22:36 does that provide a window for Summit feedback session on it? 11:23:15 unfortunately no, which is why I want to talk about it at our meetings 11:23:29 we will be talking about it at PEARC19 11:23:40 (in a couple weeks) 11:24:14 ah yeah, i would have liked to head to PEARC, maybe next year 11:24:45 so for people interested, this is the change to get your comments in on the efforts of federation 11:25:25 note that this is a foundation document (reference architecture) 11:25:42 perhaps you can give us an overview of NIST SP500 in a future session here martial_ ? 11:25:42 and that there is still a lot of work to be done :) 11:26:14 I can, but I will invite my co-authors to join 11:26:26 sounds good 11:26:28 so likely the session that works better in the pm :) 11:26:35 (pm for me that is :) ) 11:26:57 better for me - almost midnight here now 11:27:25 #topic AOB 11:27:38 ad lib time :) 11:28:04 i'm working on a little containerisation problem at the moment and wondering if people can help... 11:28:57 if anyone has experience of instances spontaneously/persistently ending up kvm-paused I'd love to hear from you (pm/email) - not made much progress with Google or RH support 11:29:28 is temporary disk space depletion a possibility? 11:29:58 I wondered that but the hypervisor has >200GB free so I don't *think* so 11:30:01 have a use-case for a client-server workflow that builds map cache layers hanging around a Postgres DB plus frontend webserver API, want to be able to fire the whole stack up on-demand within the HPC environment... 11:32:23 we've got a Docker container that has all the dependencies, and were thinking we could go from there to Singularity image 11:32:39 have you looked at Sarus? 11:34:14 looked like a very promising OCI compliant runtime 11:34:34 creating the image is easy enough, but naturally things like postgres and apache that have been installed into the image via apt would (IIUC) normally be started in Docker via `service start apache`, but those init.d service scripts try to do that via sudo... 11:35:24 yeah heard about Sarus at CUG, don't think it really adds much that would help in this use-case though 11:36:13 running more than one process inside a container always seems to end up being fragile, would multiple-containers be acceptable? 11:37:12 dh3: yep, i suspect the "proper" way to do this might be multiple Singularity service "instances" of the same container 11:38:03 not sure how they work from inside a Slurm job yet though... 11:38:34 my knee-jerk reaction is "wrapper script" but it's not terribly elegant 11:39:37 my main question is whether there are some standard practices for building containers that allow launching services without root escalation requirements 11:41:13 dh3: i think i've seen that pause issue at some point in the past, wondering if sw3 is around and has better memory than me 11:41:16 do you mean, not-have-root when building the container? We haven't found a way round that (and suggest users do it in OpenStack where they do have root) 11:41:18 you can create a local account on the container that matches the local user; only problem is that this needs to be done per user 11:41:55 so a `docker build` per user 11:42:38 so create the user last to be able to rely on the previous union layers 11:43:13 building isn't a problem, can do that on a local dev machine or via remote build service (and with Singularity 3.3 there's new fakeroot functionality that enables rootless builds, albeit with a couple of caveats) 11:44:03 fakeroot> good to know, ta 11:46:16 how does such a matching container-internal user help martial_ ? 11:50:55 you were talking about root-less containers, I have a couple ubuntu based where you can sudo without a password but the container is run with a local user that matches the directory it is in 11:51:48 not sure if it is the use case that you are envisioning 11:52:27 sounds like a potential avenue worth looking at 11:52:47 any Dockerfiles you can point me to? 11:53:54 not public but the core component you are looking for is 11:54:14 https://www.irccloud.com/pastebin/uLdfgCPN/ 11:54:22 I gotta run, I'll see you guys later. Till next time! 11:54:40 that users get access to the webcam too 11:55:36 do i smell a deep learning inference application? :-) 11:55:47 you can specifiy the uid and gid as needed by adapting it 11:56:31 and using `ARG` 11:56:49 OpenCV / Tensorflow application 11:57:01 great, thanks martial_ , will see how we go 11:57:41 #link https://github.com/datamachines/cuda_tensorflow_opencv 11:57:52 built `FROM` that 11:58:47 ok, time to wrap up and sleep 11:58:55 till next time! 11:59:12 thanks, bye! 11:59:22 #endmeeting