Skip to main content

Flask, uWSGI, and NGINX - a saga

I have been creating a web page for my wife. It is a booking site for her business and is written in a combination of jquery, HTML, CSS, CouchDB, and python.

For the python side of the house I am using Flask. This micro-framework works well for me and allowed me more freedom than I saw reading about Django.
It has taken some time, but the app works well. It can retrieve bookings from CouchDB, display them in a calendar, and accept new bookings from a form on the same page. Jquery handles the calendar display, as well as the AJAX call to populate it. Flask handles the data collection from Couch, the data, and the writing of new data to Couch. For cleanliness, two repositories are used for bookings: one for confirmed bookings and one for requests that have not been reviewed. Another repository provides a client list, but is not accessible from the website.

Then came the fun part. To serve a website with Flask, the internal web server is insufficient. You need additional tools. In my case, NGINX and uWSGI. uWSGI provides an intermediary between Flask and NGINX. NGINX provides a reverse proxy, handling the requests to serve the website.

To make the three elements talk, you need to configure them. A good starter guide is found here. However, your mileage may vary. Mine did.
Configuration is not the same as in the article for my systems. I found the same thing worked in an AWS EC2 instance as worked on a local Vagrant Debian Stretch instance.
So, referencing the starter guide, here are my deltas.

myproject.ini

My file ended up being the same as the one in the document, but I have added some explanations to the item below:


[uwsgi]

module = wsgi #This is the name of the file to be run to start the application. In the example it is wsgi.py, less the suffix.

master = true

processes = 5

socket = myproject.sock #The name is irrelevant. This is just what the socket will be called. Because it has no path, it will be created in the same directory as this ini file.

chmod-socket = 660 #permissions for the above socket once created

vacuum = true

die-on-term = true


Upstart script

This is how I made a raspberry pi SD card unbootable. Upstart is not the standard system process manager for Debian (or Ubuntu, apparently). Casually installing it will delete Systemd, the default. I would not recommend doing this unless you know what you're doing, in which case you probably wouldn't be taking my advice anyway.
I wrote a systemd service rather than upstart config. There are translation tables between the two to help.
To make the equivalent systemd file, you'll need to sudo vi /lib/systemd/system/<name_of_choice>.service to make a service file and open it for editing.
If you don't like vi, substitute editor of choice.

The file I made looks like:

[Unit]
Description=Whatever you want to write
After=syslog.target  

[Service]
User=<user to run uwsgi>
Environment="PATH=/path/to/your/apps/virtualenv/bin"  #The quotes and PATH were needed here
WorkingDirectory=/path/to/your/ini  #Do not include the filename in this path
ExecStart=/path/to/your/apps/virtualenv/bin/uwsgi --ini /path/to/your/ini/<filename.ini>  #No relative paths here. Uses the virtualenv uwsgi to open the ini file.
Restart=always

[Install]
WantedBy=multi-user.target  #This is runlevel3. WantedBy ensure it starts at this runlevel.


Final item

This was the most annoying part and betrays my relative newness to Linux. Nginx could not do anything with the socket. Without the socket, it can't talk to uWSGI. Without talking, my app is not going to be served.
I resolve this with a poor approach, but it worked. I edited /etc/nginx/nginx.conf.
For good practice, when editing this kind of file, it's worth doing something like cp /etc/nginx/nginx.conf /etc/nginx/nginx.conf.backup.
The above command will make a copy of the unedited file, allowing you to recover from catastrophic mistakes.

All I did to this file was edit the very first line from:

user www-data

...to:

user <my user>

Then everything worked. That being said, the better approach would be to work out why www-data was prevented from handling the socket, but I chose not to die on that hill today. My app works inside a vagrant box and on AWS. 


This post was just to highlight the variations this user experienced. I regard the linked guide very highly and would recommend it for starting out. Aside from the above items, everything else worked fine.

Comments

Popular posts from this blog

Context in Node-Red

It's probably not news to anyone willing to read documentation, but I was able to simplify a number of Node-Red flows recently after a primer on context. In Node-Red, each function node has a self-contained context. Variables are local to that node and nothing is permanent. Unless you use a different context. It is possible to set flow-level and global-level variables that can be used to store values, provide them to multiple other nodes without links, and give the illusion of memory to Node-Red. Setting flow context uses simple syntax: flow.set( 'name of the variable' , 'value of the variable' ) This new flow-level variable can be called anywhere in the same Flow tab. This is great for recording values that don't change every time the flow runs (e.g. a maximum recorded value), or need to be used in isolation from the pathway that generates them. Calling the value is relatively simple as well: var varName = flow.get( ' name of the varia...

Return to the garden

After the mediocre performance of my vegetable garden last year (50% of the plants produced), winter is the perfect time to reflect on what went wrong. First, I started the project with a simple idea and absolutely zero experimentation. Second, the methods I chose did not work as I had hoped they would and my fall back was too simple. Third, minor tech troubles exacerbated the issues caused in the previous two steps. To address item one, I have started prepping my solution as of the end of November 2017 with an eye on March 2018. This is giving me time to test and refine as I go. On item two, I had to look at what worked and what didn't. The pump system worked well, but needs to be reconfigured to deliver water at the soil level; even a moderate drop of four inches resulted in erosion and root exposure over time. The planters were acceptable, but the height differential was tough to deal with. New planters will be needed. The right microcontroller was not available immedia...