#django

Koterpillar: what did he skip?

sorry i didn't mean to skip anything, what did I miss?

oops, my bad, it's me who missed the dpaste. Jon_R: sorry!

Jon_R: nothing, sorry

kk :-)

Jon_R: you have --python-path in your systemd unit file right now? if you didn't do that yet, please run systemctl stop gunicorn; systemctl daemon-reload; systemctl start gunicorn

also, show us the updated gunicorn.service file

ah that's a good point i've been reloading not restarting, 2 secs

gah

it's --pythonpath not --python-path, 2 secs

oh, right -- sorry

woohoo!

it works!

good :)

I'll stick with uwsgi, thanks :)

my systemd config now just says "/srv/wsgi/django/env/bin/gunicorn --env DJANGO_SETTINGS_MODULE=production.settings --pid /run/gunicorn/pid production.wsgi" , no pythonpath or anything

but it works :-)

which is what i had before, but now somehow it's different

pid files? really?

why?

FunkyBob: does uwsgi make a big difference here? you don't have to specify the pythonpath there?

because that's what the gunicorn documentation says to use

moldy> in uwsgi you just say virtualenv=/foo/env and job done

oh. i actually like the gunicorn approach better, tbh.

i'm trying to behave myself and do what I'm told as much as possible here :-)

well, you can install uwsgi in your venv

FunkyBob: yes, that's what i do with gunicorn, too

however, uwsgi provides a LOT more than gunicorn

but how does uwsgi detect the path to your django project root if you don't configure it?

moldy: that's nothing to do with venv

FunkyBob: yes, that is why i wonder

that's a pythonpath issue, and you configure that

haha now all i need to do is work out how to get nginx to behave

alright, so this is the same :)

sure, uwsgi can do more, but if you don't need any of that, it doesn't matter. i also think that uwsgi does a little too much, maybe. matter of taste :)

I really need to put fragment links in this doc

moldy: yeah.. Graham Dumpleton has the same view :)

it's a bit like AWS... once you start using one thing, it tends to borg the rest of you

i don't really enjoy it to take over maintainership of projects that are uwsi-only because they use their "cron" thing, for example

caching, spooling, and so much more :)

uhh :p

uwsgi cron thing?

yeah, it can run periodic tasks

iirc i was told you can run periodic tasks with uwsgi

moldy: ws API, mules, offload workers...

hah cool. I didn't even know that and I've used uwsgi forever

so you don't need celery (which is nice), but now your app is bound to uwsi (not so nice)

Heh. We've actually got a project where uwsgi spawns the celery worker with attach-daemon.

oh, yeah, so don't need supervisord either :P

i just use systemd :)

eww

by now it's a good practice, imho

:P

why should each app deploy its own process supervisor

and you get half-decent logging for free.. journalctl -u

oh, I agree... _no_ app needs to daemonise

and the template thing can be used to get zero-downtime deployment relatively easily

mattmcc: hah

and if you move resource limit enforcement, duration control ,et c etc into the launcher's config, you're moving closer to mainfame OS territory :)

what i do is: have gunicorn.project_name@.service, where the tag is a timestamp. deploy the new code to a directory with a timestamp in the name, start the new service, rewrite the nginx config, wait a little, reload nginx

I did something similar for opensm ... the @ syntax is neat

the way I configure nginx means I don't need to reconfigur it

oh, how do you do that?

see my link ^^ :)

will give it a read :)

though i probably need an nginx config change in some situations, because i also do fully-automatic letsencrypt :)

it's not quite what I use now... [many of my servers don't even use nginx now]

ah, yes...

so i first deploy the app with ssl, then get the cert, then enable ssl

*first deploy *without* ssl

Sep 28 01:17:52 dexter gunicorn[31296]: message repeated 19 times: [ browser reported failure]

well that's a handy error message

look at the line before, Jon_R

the line before is Sep 28 01:17:45 dexter gunicorn[31296]: browser reported failure

the line before that says "received unexpected data"

i really think i should polish up my stuff, document it, support more distros, and then we could have a nice community-managed standard deployment script

+1

https://github.com/plone/ansible-playbook the plone people did a nice job there

ooh.. http://uwsgi-docs.readthedocs.io/en/latest/SNI....

"uWSGI 1.9 (codenamed “ssl as p0rn”) added support for SNI (Server Name Identification) throughout the whole SSL subsystem. The HTTPS router, the SPDY router and the SSL router can all use it transparently."... why are they using pr0n as a codename, why are they doing SPDY...?

well, they do https2 now

in 1.9 it was still called spdy :)

yeah but uwsgi is wsgi why is it worried about http/http2/etc at all?

... no, it's FAR FAR more than that

besides, WSGI is just the back end... it needs a front end protocol to talk, right?

it does http, https, http2, uwsgi, fastcgi... and probably more

just like gunicorn talks http

oh i thought it was a wsgi server you stuck a web server in front of

yeah that was a thing that worried me about gunicorn ;-)

well, sure, but it needs a protocol for other servers to pass it requests, right?

yes, wsgi?

wsgi is not a protocol

it's an API specification

an interface

the whole point of tools like gunicorn, uwsgi, waitress, etc...

is to publish WSGI apps

provide some protocol for requests to come in, and then hand them to the app

ok so I now have no idea what "uwsgi_pass" does in nginx ;-)

FunkyBob: where is the uwsgi_pass thing coming from?

moldy: ?

Jon_R: it tells nginx to pass requests to another server using the uwsgi protocol

uwsgi is a protocol?

uWSGI designed its own protocol, unrelated to WSGI or HTTP

ah

here's me trying to avoid NIH syndrome and finding myself constantly the victim of other peoples' addiction to it ;-)

FunkyBob: nginx has support for that built-in?

browser <-- http --> nginx <-- uwsgi protocol --> uwsgi <-- python (using wsgi) --> your app

moldy: yep, just like it does for http and fastcgi

FunkyBob: ah, i see. thanks.

moldy: https://nginx.org/en/docs/http/ngx_http_uwsgi_m...

FunkyBob: and what does _ mean on that line? uwsgi_pass unix://srv/apps/_/server.sock;

or in my case: browser <-- http --> uwsgi <-- python (using wsgi) --> my app

moldy: it's a dir name?

# mkdir _

FunkyBob: ah, so on update, you just remove that directory and create a new one?

?

no?

FunkyBob: i'm trying to understand how you do zero-downtime updates without changing your nginx config

deploy nwe code... touch uwsgi.ini in the app's dir

FunkyBob: ah, ok. i do things quite differently. on update i deploy a whole new virtualenv with its own gunicorn.

can do that

if you add another /*/ to the emperor search path you can have multiple copies of your code

and the one that runs is the one with a uwsgi.ini file

hence i timestamp the whole directory, change the nginx config to point to the new one, reload nginx

the emperor discovers what to run by finding ini files