| 27 May 2019 |
 martin | just that it receives the info via ssh really | 05:24:25 |
 @kinow:matrix.org | In reply to @revilo666:matrix.org
What's different about UI Server vs Notebook Server in this (authentication) context? I think the Notebook has some code, somewhere, that does not depend on jupyterhub, and is still able to authenticate against the hub. | 05:24:29 |
| @kinow:matrix.org | But both behave similarly, if you sniff the network, the packages should be similar.. .maybe some headers missing here and there | 05:24:56 |
 Hilary Oliver |
just that it receives the info via ssh really
But JupyterHub has been used to spawn Notebooks via ssh before, right?
| 05:25:07 |
| martin | remotespawner used to work | 05:27:30 |
| Hilary Oliver | In principle you should be able to use any kind of spawner (any conceivable way of starting a process on a remote user account) so long as the right hub info can be passed to the spawned process at start up. | 05:28:44 |
| martin | I'm not sure if it ever brought up an interface, I could never get that to work | 05:29:10 |
| martin | I can get the cylc-uispawner to start any process, yes | 05:31:06 |
| Hilary Oliver | Yes the original remotespawner has become unmaintened, so I expect we would have to fix it or use another one (such as the NERSC one I found). But fixing it shouldn't involve having to mess with the low level authentication details (I would have thought) - it should just be a matter of spawning a remote process via ssh, and passing the necessary Hub info to the spawned process (of course it's easy for me to say that when I haven't actually done it 😁) | 05:32:37 |
| Hilary Oliver | (and the spawned process has to know how to communicate with the hub, of course) | 05:33:05 |
| Hilary Oliver | (and that is where our UI Server needs to emulate the J Notebook Server) | 05:33:32 |
| martin | I've already produced that functionality, it's just the difficulty of debugging the return auth when i can't log in the handlers! | 05:33:52 |
| martin | so if that is fixed the rest should be simple | 05:34:09 |
| martin | that's why I foeked ui-server, to create that similar auth handling | 05:35:30 |
| martin | forked! | 05:35:52 |
| Hilary Oliver | OK, that does make sense - if we don't want to import J-hub code at the remote end - but I thought that Bruno P. Kinoshita 's UI Server already did that - by importing hub code. So we don't need to fork and do all that work that is causing you trouble, so long as we don't mind importing hub code in the UI Server. (Is that right Bruno?) | 05:37:06 |
| Hilary Oliver | (I like the word "foeked" BTW, nice!) | 05:37:54 |
| martin | i heard it in hoboeken 🙂 | 05:38:29 |
| @kinow:matrix.org | In reply to @revilo666:matrix.org
(I like the word "foeked" BTW, nice!) Yup, that's correct in theory. I'm trying the NERSC sshspawner now (I think it does almost the same as Martin's spawner). If that works, I think we can consider it done for now and revisit the issue about removing the jupyterhub dependency. | 05:39:18 |
| @kinow:matrix.org | hohoh, replied wrong message | 05:39:36 |
| Hilary Oliver | 😁 | 05:39:50 |
| martin | hey waht's wrong with my spawner? | 05:40:25 |
| @kinow:matrix.org | haha, nothing wrong, but I would have to work on the authentication with the hub | 05:40:42 |
| @kinow:matrix.org | That's quite scary :-| so I am taking the shortcut to validate SSH spawning 😁 | 05:41:04 |
| martin | I don't think nersc does the auth stuf does it? | 05:41:23 |
| @kinow:matrix.org | cylc_singleuser should do the auth 🤞 via SSH tunnel back and forth | 05:41:45 |
| martin | k | 05:41:55 |
| @kinow:matrix.org | In theory ... hehe | 05:41:57 |
| Hilary Oliver | So martin - definitely don't want to lose what you've done on the no-hub-import-authentication (see your Actions in the meeting notes) - but let's just prove the base level functionality works and debate the pros and cons of removing dependence on hub code later on as a possible improvement. | 05:41:57 |
| @kinow:matrix.org | Yup, keep those repositories as the issue for revisiting the jupyterhub dependency may be based on your previous work martin . | 05:42:45 |
