Scalability for CNCF Demo about cluster HOT 10 CLOSED

I'm confused. In yesterday's meeting you mentioned that the goal was to serve as a tech-demo, not testing bare metal cluster, as it's written in this proposal. While I agree with the former, I don't with the latter - there are multiple teams across a number of companies that are working on testing clusters at scale. Cc'ing Google, RedHat, CoreOS and Samsung, but there are more people and companies involved - basically sig-scale is the forum that oversees our scale testing efforts.

Also there shouldn't be a difference on testing bare metal clusters and public-cloud-based ones once the cluster is running. The results would be different, but the tooling for running tests should be the same (and I do feel very very very strongly about this).

cc @jeremyeder @wojtek-t @timothysc @xiang90 @countspongebob

from cluster.

@gmarek More info on the CNCF Demo is at https://github.com/cncf/demo#cncf-technologies-demonstration. In short, we've been doing all of the work to date with a 20 node cluster, (and with a similarly small cluster on AWS), and are now interested in trying the same applications on a larger 100 node cluster. But we're particularly focused on the scalability of Weave and Countly, as we’re presuming that K8s easily scales to 100 nodes.

Separately, @Zilman and I will be in Berlin on 10/5 and 10/6 for ContainerCon in case you might be able to chat with us in person.

Cc @monadic @bboreham

from cluster.

@dankohn - I see that there's some unfortunate choice of words (e.g. "Anybody who is interested in reliably running Kubernetes on big clusters."), which made me think that it's about testing K8s scalability (yes, we do scale to 100 nodes easily). What you're saying (i.e. scale-testing Weave and Countly) makes perfect sense - thanks. It's still not clear to me if this effort is supposed to be a demo or scale tests, but I'm much less concerned now.

I now see that there might be substantial differences between the ways in which we load the cluster to test K8s performance and to test other technologies. It would be perfect if we were able to create a single framework for both those applications, but I don't think it's a hard requirement any longer.

from cluster.

I'd love to speak with you, but I wasn't planning on attending ContainerCon (@fgrzadkowski is going though).

from cluster.

@gmarek I also changed the problematic answer from "Anybody who is interested in reliably running Kubernetes on big clusters" to "People interested in running Weave and Countly on larger clusters."

@fgrzadkowski could you please look me up? I'll be around the CNCF booth and my email is on my Github profile.

from cluster.

@bprestonlf is this request approved as well? We could probably prepare the nodes for WW46 assuming it meets the requirements.

Is this request's intent to add 80 nodes to 20 that you have been using so far or do you want another 100 nodes beside the 20 you already have? Will the 20 nodes mentioned be freed after 2 weeks of testing on 100?

from cluster.

Thanks for the questions. I'll update later today with more info.

from cluster.

Please hold for now. We'd like to do some additional work before getting extra nodes.

from cluster.

Hi @Zilman @dankohn any update on this one?

from cluster.

@Zilman has all he needs for now and will reopen when he wants more resources. Thanks!

from cluster.