29 Aug 2019 |
@kinow:matrix.org | I check their channel every week. I feel like 90% of the questions there are about Kubernetes 😕 | 07:50:07 |
30 Sep 2019 |
| @zoejr:matrix.org joined the room. | 00:01:46 |
| @zoejr:matrix.org left the room. | 00:02:15 |
9 Oct 2019 |
| oliver sanders invited melhall. | 08:18:30 |
17 Oct 2019 |
| melhall joined the room. | 13:52:14 |
24 Oct 2019 |
| melhall set a profile picture. | 08:33:09 |
20 Dec 2019 |
David Sutherland | Redacted or Malformed Event | 00:20:09 |
David Sutherland | * . | 00:23:37 |
3 Jan 2020 |
David Matthews | Had a play getting Jupyterhub authenticating with Azure AD today. Had a few issues but I think it's working now. Hopefully we can use this for our site installation. | 20:02:14 |
David Matthews | Still a bit unsure what problems we might hit trying to allow access to other users UIS. | 20:04:09 |
David Matthews | Also not sure how we're going to scale this to support lots of users (can we distribute UISs over multiple servers securely?) | 20:05:52 |
David Matthews | Something for the workshop agenda? | 20:07:20 |
@kinow:matrix.org | We might be able to use something from jupyterhub docs on how to scale. | 23:37:18 |
5 Jan 2020 |
Hilary Oliver | In reply to @davidmatthews:matrix.org Still a bit unsure what problems we might hit trying to allow access to other users UIS. Yes we haven't really investigated this yet, beyond being reasonably confident that it is feasible. | 19:34:13 |
Hilary Oliver | In reply to @davidmatthews:matrix.org Also not sure how we're going to scale this to support lots of users (can we distribute UISs over multiple servers securely?) Do you mean lots of users running their own workflows, or lots of users looking at the same (e.g.) operational worflows? | 19:35:52 |
Hilary Oliver | The former is naturally distributed (own UIS runs on own user acccount) - should not be a problem? | 19:36:14 |
Hilary Oliver | The latter might be problem (one UIS serving e.g. 100 users)? But in the first instance we can easily limit the number of connections and at least prevent it causing performance issues. | 19:37:00 |
Hilary Oliver | But yes, something to talk about in the workshop. | 19:37:32 |
oliver sanders |
Do you mean lots of users running their own workflows, or lots of users looking at the same (e.g.) operational worflows?
| 20:19:23 |
oliver sanders | This kind of scaling would require an additional reverse proxy for load balancing between UIServers, potentially possible but horizontal scaling is not as yet a feature of the new architecture. | 20:21:13 |
oliver sanders | Other than potentially redundancy I wouldn't imagine we would ever need to scale beyond the capability of one tornado server though. | 20:22:25 |
Hilary Oliver | You mean the second kind of scaling, right? (there's two kinds in that sentence) | 20:22:31 |
oliver sanders | (horizontal scaling i.e. more servers doing the same thing) | 20:22:59 |
Hilary Oliver | Yes, which would only be needed for many users looking at the same workflows. | 20:23:34 |
oliver sanders | Yep, though I don't think we are ever likely to hit a sufficient level of usage for that to be a problem. | 20:24:27 |
oliver sanders | e.g. Cylc Review does just fine with one server running for many accounts. | 20:25:29 |
oliver sanders | (under cherrypy) | 20:25:44 |
Hilary Oliver | So I think we don | 20:37:14 |
Hilary Oliver | ... | 20:37:17 |
Hilary Oliver | ...'t need to worry abut this now. If it becomes a problem in the future, we can address it then. | 20:37:52 |