Zoom Logo

IRIS-HEP Retreat Parallel 1 Wednesday - Shared screen with speaker view
Brian Bockelman
10:33
Live minutes FWIW: https://docs.google.com/document/d/1c-dFS5-7IEUOUm4Yikb3v-T7jfuaw9f_KRUT7wvyprU/edit#heading=h.2itk9lr36liz
Brian Bockelman
20:45
Live minutes: https://docs.google.com/document/d/1c-dFS5-7IEUOUm4Yikb3v-T7jfuaw9f_KRUT7wvyprU
Brian Bockelman
26:34
Hi all - I think some folks are accidentally unmuted and am going to mute all real quick. Feel free to unmute yourself for questions!
Carlos Maltzahn
27:08
Thanks :-)
Frank Wuerthwein
32:58
Are the slides online somewhere?
Brian Bockelman
36:14
not yet - will post them after Carl's done.
Gordon T. Watts
40:24
Could we also get that notebook file posted.
Gordon T. Watts
40:29
That was impressive.
Kyle S Cranmer
41:22
Very nice!
Kyle S Cranmer
42:21
What do you envision the notebook to look like eventually. Currently there is quite a bit of code related to HTCondor, Dask, etc.
Benjamin Galewsky
44:27
Put the links in the live notes, please
Kyle S Cranmer
48:18
My question: What do you envision the notebook to look like eventually. Currently there is quite a bit of code related to HTCondor, Dask, etc.
Benjamin Galewsky
49:04
How hard is it to interact with the DASK cluster via python code on my local laptop instead of through Jupyter/
Gordon T. Watts
54:04
Side note: US ATLAS is also working on a k8 Tier 3 which has some similar elements. It might be nice to organize a blue print meeting that looks at the efforts and what is availible. @MarkN.
Rob G
56:53
Helm charts for any of this, for deployment elsewhere?
Frank Wuerthwein
57:03
about SSO: we should not tie entirely and only into CERN for authz. It makes working with undergraduates impossible. I’d like to be able to give non-sms members access to my facility because they are undergraduates that come and go.
Gordon T. Watts
57:18
+1 to Rob's and Frank's comment.
Gordon T. Watts
59:01
Also - needs to be able to run this sort of thing from a python script - not just a notebook environment.
Brian Bockelman
01:06:20
@RobG: The Helm chart is in the live notes. Apologies to those who are helm experts for the ugly code.
Benjamin Galewsky
01:15:39
Yes! @msn
Gordon T. Watts
01:16:18
Blueprint!
Gordon T. Watts
01:18:19
We normally support libraries - people build a fitting library. Are we just talking about moving that model to supporting a service rather than a library?
Alexander Held
01:20:22
I think the service would take a workspace + an instruction of what to do with it and return you the result (there’s a slide or two in Matthew’s pyhf talk in the next parallel).
Kyle S Cranmer
01:20:51
I think the libraries is a SW side, but when we talk about analysis systems, it includes the infrastructure / service elements to run the code
Kyle S Cranmer
01:21:36
There is a demo of a fitting service already that Lukas demo’d at CERN. It’s a function-as-a-service model and you stream pyhf JSON to it.
Benjamin Galewsky
01:22:02
How much does it make sense to talk about specific “services” when we have the kubernetes and DASK backplane? Does something specific need to be running, or is it just config that can be launched into the framework?
Gordon T. Watts
01:22:10
Yes, and there is also FuncX which takes the idea and makes it more general.
Benjamin Galewsky
01:22:18
Yes
Gordon T. Watts
01:22:25
And I wonder if people couldn't just "code-to-that" as it were.
Gordon T. Watts
01:22:53
So you still do what you are interested in - the software that does the fitting or whatever - but then you slot it into an infrastructure that makes it easy to live in the analysis system's ecosystem.
Gordon T. Watts
01:22:59
(e.g. play with other tools)
Kyle S Cranmer
01:23:08
For me the service does something specific with a specific API
Gordon T. Watts
01:23:29
A thin layer around the software that does the hard work?
Mark Neubauer
01:23:35
Yes (well-defined, useful analysis services to talk to)
Gordon T. Watts
01:23:36
(at least, that is where I was trying to head)
Gordon T. Watts
01:23:58
(and not that infrastructure isn't hard work)
Gordon T. Watts
01:24:20
What RobG is talking about now.
Kyle S Cranmer
01:25:24
Here is the fitting service demo:
Kyle S Cranmer
01:25:25
https://github.com/lukasheinrich/lhoodbinder2/blob/master/PlotContour.ipynb
Gordon T. Watts
01:28:27
I just saw a bunch of people wanted access to the minutes. I've granted it - if you refresh your page you should be able to contribute. Appologies for having missed the messages!
lincoln
01:30:22
my mic may be busted but this is definitely something we've been trying to simplify with SLATE. We would like to reduce the amount of stuff that admins have to worry about in the world of K8S
lincoln
01:30:38
part of that is having very opinionated deployments of K8S
Rob G
01:33:36
reminder we have wlcg-k8s working group
G. J. Peter Elmer
01:37:13
I agree 100% with what FKW is saying, but need to drop out for a Princeton administrative discussion.
Rob G
01:39:02
n.b. we do have the SSL convo coming next, https://docs.google.com/document/d/1BZAR1p_54xn2d8yekzjYmTPjsjj5qhfTCUISZ9T_mz8/edit# for those who want to continue
Ivo Jimenez
01:40:41
Related to handling dynamic environments in kubernetes that are CI/CD’d, there are tools such as Orca that can help automate this an plug it to Gitlab https://github.com/nuvo/orca