#django

yes

anarchean: there are advantages to both options

I think in your case I would most likely create groups with appropriate permissions, and put the users in groups

I would need an extra permission "read", tho, wouldn't I? Because there are some information that only "manager" users can access

hello, how to specify spcial group of users as a forign key for a model?

I mean i am creating a model class, only users who are teachers should be able to modify it

@geekodour08 I think what you want is to use django permissions to limit user access and create a group "teachers" that have CRUD access to the model

But I may not be the best reference since I'm looking for tips for the django auth system here too

geekodour08: take a look https://django-guardian.readthedocs.io/en/stabl...

i try ot understand some of the magic of django models, but here is one part idont quite understand. how can a function call with no args still result in args being passsed? django/db/models/base.py(318)__new__() -- new_class._prepare() to django/db/models/base.py(329)_prepare() -- def _prepare(cls): .. can somebody explain how this works?

kakulukia: that's Python objects/classes; I don't have a good guide handy

kakulukia: https://docs.python.org/3/tutorial/classes.html might do it

doismellburning: but ive never seen this before - right in the debugger its calling _prepare without args and stepping into the function call the cls arg is filled .. where does it come from?

kakulukia: you've never written `def foo(self):`?

kakulukia: the classes tutorial should explain it, I think

but it's effectively exposed syntactic sugar; `my_obj.foo(z)` -> `call myobj.foo with myobj and z`

I assume that running `manage.py migrate` simultaneously on the same DB is a bad idea. My coworker seems to think it's fine. Which of us is insane? :p

NfNitLoop: you may run into issues, especially if your db doesn't support DDL in transactions

Is there any sort of documentation about whether migrate has any sort of locking protecting it? I've been skimming https://docs.djangoproject.com/en/2.0/topics/mi... but didn't find anything explicit.

knbk: We're running Postgres, thankfully, but even then I really wouldn't want to rely on that because there could be application-level race conditions.

NfNitLoop: I wouldn't like to do it

NfNitLoop: yeah, I wouldn't risk it either. There's no docs on locking because there is no locking

(ex: 'migrate' reads the schema version simultaneously, N of them kick off ALTER TABLEs, one of them locks, the other one continues later and gets errors, etc.)

doismellburning: i did use the self arg before, but this prepare method is called when the class iscalled into existence rather than being instanciated and thus it really gets called without self/cls but still in the django code cls is there

NfNitLoop: we use Careful Deployment Optimism to ensure it doesn't happen

doismellburning: What's that?

NfNitLoop: Elastic Beanstalk config, and hoping that it works one at a time ;)

I suspect at greater scale we'll need something better but it works well enough for now

The issue is that we're doing continuous delivery to a kubernetes cluster. We'd hacked in a `manage.py migrate` before starting up the Django server, which worked OK when we only had 1 web head. Now we have many.

Kindof surprised at the lack of handling of this in the kubernetes/helm space. w/ all their talk of rolling deployments, it's like they think DB schemas never change. :p

they use mongodb

(afk: driving home) haha. Web scale! :p

NfNitLoop: ah, nasty

NfNitLoop: we essentially migrate_and_run, but that's because EB offers a "one at a time, srs"

I suspect that isn't a _strong_ guarantee but it works well enough

I think you're going to need to implement now what I think we'll need to eventually

So, if you could like, open source that, it'd be neat cheers ;P

(you have a ZK cluster with k8s right? lock there, job done)

(with the right ZK magics)

can i access the object a form creates inside the form_valid of a Generic Class (e.g. CreateThingView) ??

?*

doismellburning - we do a similar thing, but with EC2s. We stand up a full fleet of web servers, and then switch them all into the ASG's and ELB's at the same time, with the primary one running any pending migrations

We basically deploy "front to back", so any of the background tasks that are running may fail but will be picked up again in 12 hours, and the UX doesn't break

Doing that with k8s is actually incredibly hard.

I mean, guaranteeing any kind of ordering of how things happen.

You can bring up your pods and run the migration, then switch the service, but that means monitoring the state of the pod you ran. And that brings me to how painful it is to have a one time job execute in k8s :P

yeah, this is why I don't really trust our EB method

hrm, genuine question, what is the advantace of websocket over a rest api?

websockets are bidirectional