Django email tracebacks are truncated (no stack trace)

[cbay]

I migrated my application from Apache/mod_fcgid to gunicorn 0.12.2 with gevent as worker class. I'm using Django 1.2.5.

My email tracebacks are now received truncated, for instance:

DoesNotExist: Transaction matching query does not exist.


<WSGIRequest
GET:<QueryDict: {}>,
POST:<QueryDict: {}>,

[...]

There's no stack trace, only the exception message. It used to work fine with Apache/mod_fcgid (I had a complete stack trace.)

I'll try to reproduce it with a minimal application. Everything else is working fine, by the way.

[davisp]

@jasondavies ran into this today as well. He also noticed that Exception location in the debug=True report page didn't list the exception location. I could not reproduce this behaviour even grabbing his exact versions (except for maybe Gevent which is whatever is in PyPi as of about 10 minutes ago.

[jasondavies]

Okay, I've narrowed down the issue to being something to do with these lines in my worker config:

from psyco_gevent import make_psycopg_green
make_psycopg_green()
worker.log.info("Made Psycopg Green")

Removing them makes the DEBUG=True traceback work. So this could well be an issue with psyco_gevent rather than gunicorn.

[jasondavies]

You can get psyco_gevent from here.

[cbay]

I've got the exact same lines in my config.

[jasondavies]

I've managed to reproduce in a vanilla Django 1.3 project. The files are here: http://www.jasondavies.com/gunicorn-214/

[benoitc]

Can't reproduce on 325c5a8 with the test site above:

Traceback appears in console (This isn't displayed on the page by default)
Tracback appear on the page if Debug=True

Can you retry with lst trunk?

[benoitc]

Can't reproduce at all with different configurations. Im' closing this ticket then. Feel free to reopen one if needed. Thanks a lot for the feedback and tests anyway.

[cbay]

The bug is still present. Jason's test project doesn't exhibit the bug because you have to connect to the database in the view, which will call psycopg_gevent's code. I've recreated a test project: http://share.alwaysdata.com/gunicorn_bug_214.tar.gz. Here is my full shell session, from creating the virtualenv to getting the bug (on my Ubuntu 11.04):

http://www.friendpaste.com/4tRW1ergfFWnUeKQXGSVRr

Note that you'll have to edit the settings.py to enter your PostgreSQL database settings. The database won't be touched at all (there are no models in the project).

[cbay]

It works fine with eventlet. So I guess I'll have to file a bug report at gevent.

[obeattie]

I'm encountering this issue now in production, which is extremely annoying. Does anyone know what happened with this (if anything?)

[denik]

Here's some info on the issue: https://groups.google.com/forum/?fromgroups=#!topic/gevent/hdCT8cddnrw

[fdhenard]

From the previous comments, it appears that this was a bug in gevent. I'm using gunicorn 18 and pypi is showing that as the latest version. I don't have gevent in my pip requirements file, so I assume that gevent is a dependency for gunicorn and gunicorn resolves it (although I don't see that in setup.py). If that's the case, in order to call this fixed, wouldn't it make sense to change gunicorn so that the dependency on gevent is updated to the version where this issue is fixed, then bump the gunicorn version and publish that to pypi?

If I'm wrong about this, should I put an entry in my pip requirements file on the latest stable version of gevent in order to fix it?

Thanks!

[tilgovi]

Gevent is not required for gunicorn. It is only required if you are using the gevent worker.