I use fedora 17, and when I setup nginx with uwsgi using unix domain socket, when I place the socket in a directory with proper permission it's ok, but when I place the socket in the /tmp it will cause nginx error:
connect() to unix:/tmp/MySite.sock failed (2: No such file or directory) while connecting to upstream
The file does exist and has read/write permission for nginx user. But what cause this error, its really drive me crazy, can somebody figure it out.
You can't place sockets intended for interprocess communication in /tmp.
For security reasons, recent versions of Fedora use namespaced temporary directories, meaning every service sees a completely different /tmp and can only see its own files in that directory.
To resolve the issue, place the socket in a different directory, such as /run (formerly known as /var/run).
/run is not writable directly, but you can instruct systemd to create a directory under it which is writable by both processes.
Commented
Mar 4, 2015 at 13:04
YES! THIS HELPED!
Was seeing check-startup-logs messages from uwsgi, but had no data in the startup logs, wasn't able to find out what was happening. Turned out, I needed to change the sock file to a non-/tmp location, and add:
catch-exceptions = true
to the wsgi's ini file. turned out, I'd specified logging filehandler location to a non-existent directory, so it couldn't log the errors in the first place.
Basically, the problem arises with ONE process setting up the sock in the /tmp path, but it would not be visible by ANOTHER process.
There is a setting in the service config, which would block files created by different process, even though using the same user.
Find your service config file (ex: /etc/systemd/system/myapp.service) and then on add this setting: PrivateTmp=No to both your services.
[Service]
...
PrivateTmp=No
more info about systemd units here:
https://www.freedesktop.org/software/systemd/man/systemd.unit.html
