[IPython-User] ipython notebook server experience

Skipper Seabold jsseabold@gmail....
Thu Jul 12 07:48:06 CDT 2012


On Thu, Jul 12, 2012 at 4:26 AM, Matthias BUSSONNIER <
bussonniermatthias@gmail.com> wrote:

>
> Le 11 juil. 2012 à 17:56, Skipper Seabold a écrit :
>
> On Wed, Jul 11, 2012 at 11:36 AM, Matthias BUSSONNIER <
> bussonniermatthias@gmail.com> wrote:
>
>> I'm curious about a couple of things. Is the same notebook intended to
>> be run by multiple users connected to the same server?
>>
>>
>> We plan on having simultaneous user being able to edit the same notebook
>> at the same time,
>> probably with a per-cell lock.
>>
>> Would it make sense to have an option for copy-on-open that could be set
>> in global
>> settings with the default to the current behavior?
>>
>>
>>  I don't really see the reason why, or not enough people will have the
>> use of it.
>> In the end we might allow to create a new notebook from an URL,
>> otherwise,
>> you just have to open a notebook and create a copy from the file menu.
>>
>>
> Well my thinking is that with this set to copy on open/write, when each
> user opens the notebook, it is no longer the same notebook. It is now
> theirs. IIUC, with the cell locks, you would have users editing the same
> file, just not overwriting each others changes?
>
>
> Yes, that's the plan, to be able to collaborate, on the same *notebook*,
> with cell-lock and I think push
> to all connected client when the lock is released.
> I know of at least one french-linux-centerd-web-site[1] that allows users
> to write news in this collaborative
> editing manner, where each paragraph have a lock, and every lock release
> trigger a version snapshot.
>
> Its true that it's sometime difficult to make the distinction between
> *notebook* and *notebook server*.
> And as there is no multi-user capability, there is no notion of identity
> so no notion of *theirs*.
>
>
>  Relatedly and i don't know if this make sense, but is it possible to have
>> these copied
>> notebooks not show up in the directory ?
>>
>> Ie., I have a set of examples
>> that can serve as templates, but no one is interested in seeing the
>> copies of others or new files created by others in the list.
>>
>>
>> All this is related to multiuser capability, but in your case (and
>> generally with multi user), you wish to have
>> user isolated in their home. Which is not the case right now.
>>
>
> Right. I give the whole server a scratch environment with limited
> privileges.
>
>
>> Trying to get around with invisible files and so on is IMHO the wrong way
>> to solve
>> a bigger problem of multi-user on the same *server*. If you really want
>> copy on write
>> notebooks, it might be doable later when we'll have the possibly of
>> choosing a backend to store
>> the notebooks (PR in Progress), but I doubt this particular backend will
>> be implemented by us.
>>
> You might want to have a look at nbcloud[1], which run the IPython
>> notebook an amazon EC2.
>>
> With this each of your student will have its own notebook instance in its
>> own Vm in the cloud.
>> You could provide a small first notebook that download more notebooks in
>> the current directory if needed.
>>
>>
> Thanks I've seen this, though it's not exactly what I have in mind. I'd
> prefer to maintain my own server. Devil's advocate here: It seems like to
> me, and maybe I'm just totally misunderstanding, but a notebook server that
> only serves well for one person is a bit of a misnomer. Why wouldn't I just
> provide different user accounts where people could SSH into the box with
> their own account and run the qt-console?
>
>
> We will add a multi-user, and multi-home directory aware server. But it is
> still a lot of work, and we will have to be
> really careful with security. Depending on how we add multi users , we
> might provide the same mechanism as unix
> account with a skel, that is copied when a new user is created.
>
> Notebook and (Qt-)console are really different UI, and I thing notebook is
> better for beginners.
> Nothing prevent you to run n notebook-servers with different profiles.
> Every student will then just connect on a different port.
> It will give you a slightly better isolation by still keeping you under
> control.
>
> That might be one thing that can be done in next week sprint.
>
>
>
>>  I also saw fairly poor performance due to our server. I currently only
>> have two virtual processors allocated on this machine. I'm curious
>> what others have found to be satisfactory for notebook servers (or
>> similar endeavors) with say 10-30 concurrent users. I can request more
>> resources.
>>
>>
>> As the notebook run as a particular user and there is non privileges
>> separation, I doubt you'll find
>> 30 person sharing the same account.
>>
>
> With nbcloud? Currently, I do have 30 people logged in to a single box
> with the server being run by a single account.
>
>
> No, I speak in general, the problem is single account, so anyone that have
> access can by mistake erase everything.
> So IMHO, right now,  you won't find a real case of real work done on one
> box with only one *notebook server*, sysadmins would
> I think spawn at least 1 server per user.
>
>
Thanks, this was helpful for me to understand the notebooks server a bit
more.


>  Also, in my opinion, it would be great to have the first impressions
>> from your students.
>>
> Things that might look obvious from our point of view might not make sense
>> at all to new user.
>> So don't hesitate if you have more feedback.
>>
>
> As I mentioned most of them had very little Python background. As is often
> the case, I think they took to it faster than I did to be honest, since I
> have more set ideas and context.
>
>
> Well it was more in a UI point of view, are they confused with icons, do
> they feel like items are in the right menu.
> For example I was surprised by the fact that *pasting a cell* replace the
> current cell and does not insert it above or below.
> This has nothing to do with python programming, but if it really confuses
> new *notebook* user, python master or not,
> we might consider changing the behavior.
>
>
I didn't hear too much but will be sure to let you know as I
do.<http://mail.scipy.org/mailman/listinfo/ipython-user>One thing I
wanted that I don't *think* I saw was an undo. I pressed ctrl-m
d instead of ctrl-m b several times! I also kept wanting my vim keyboard
shortcuts, but that's another discussion I think...

Skipper
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.scipy.org/pipermail/ipython-user/attachments/20120712/c841082c/attachment-0001.html 


More information about the IPython-User mailing list