Samrat Man Singh

Email: mail@samrat.me

Hi! I’m Samrat, a programmer from Nepal.

Flask + Nginx + Gunicorn(on a Vagrant box)

2012-05-27


I had some difficulty in grasping how exactly to set up a server when I tried to do so recently, so I decided to write a tutorial that will guide you through the process. Hopefully, this post will help you avoid at least some of the confusion that I encountered.

We’ll be using Nginx + Gunicorn to host a simple Flask app. Many of you may not have access to a server but don’t worry, we’ll use Vagrant, which makes use of a VirtualBox VM to emulate a server.

###The Flask App

Because this is a post about deployment more than development, we’ll make the web app super-simple. If you’re not familiar with Flask, please check it out, its awesome and really easy to learn. You’ll also probably want to develop the app inside virtualenv- it makes things a lot neater. Make a folder in your local machine(we’re not working with the virtual-machine yet) for your app, I’ll call it codebase. Create two folders called static and templates, and a Python file called app.py. codebase should now look like this:

.
├── app.py
├── static
└── templates

Now, open app.py with a text editor and add the following:

from flask import Flask

app = Flask(__name__)

@app.route('/')
def index():
    return "Hello world!"

if __name__ == '__main__':
    app.run()

At this point, if you run app.py with python app.py, you should be able to open http://localhost:5000/ and see a “Hello World!” printed. Now, freeze your requirements with

pip freeze > requirements.txt

Great, now we’ll start working on the actual server.

Vagrant

As I said before, Vagrant allows you to work with server-like environments on your local machine. It’s absolutely great. To get Vagrant up and running:

# first make sure Virtualbox is installed, then,
gem install vagrant
vagrant box add base http://files.vagrantup.com/precise32.box
vagrant init
vagrant up

If nothing went wrong, you should now see a file called Vagrantfile inside codebase- that’s Vagrant’s configuration file. Open the file, we’ll need to make a few changes to the file.

First, uncomment the line:

config.vm.network :hostonly, "192.168.33.10"
If nothing went wrong, you should now see a file called `Vagrantfile` inside `codebase`- that's Vagrant's configuration file. Open the file, we'll need to make a few changes to the file.

and change “192.168.33.10” to “33.33.33.33”. This will enable the host-machine(that is your computer) to access the webserver running on the VM.

That way we should be able to access a web app running in the VM’s localhost, on our machine.

Because, we did a vagrant up the Vagrant box should already be running. Now, run

vagrant reload

so that the changes we made to the Vagrantfile take place.

After the VM restarts, run

vagrant ssh

This allows you to run commands into the VM. Once inside the VM, we’ll need to get some things installed. Run

apt-get install nginx
pip install virtualenv

Now let’s create a folder inside the VM where we’ll keep the application

cd /home/vagrant
mkdir www
cd www
virtualenv --no-site-packages .
mkdir codebase

And let’s grab the application from our local machine

cp /vagrant/* /home/vagrant/www/codebase/

Note that while I used cp, its always a better idea to use git or some other version-control system. For more on that, I recommend that you read this post.

Then, activate the virtualenv we created.

cd /home/vagrant/www
source bin/activate

Install gunicorn with pip

pip install gunicorn

Also install the other Python dependencies your app has with

pip install -r requirements.txt

That will grab and install your app’s required dependencies like Flask.

Now, if you run

gunicorn /home/vagrant/www/codebase/app.py:app -b 127.0.0.1:8000

you’ll have your app running but if you try opening it from your browser you’ll find that you can’t actually see the “Hello World” message that we were expecting. That’s where nginx comes in.

Nginx

First of all, you need to start nginx with

/etc/init.d/nginx start

Then

rm /etc/nginx/sites-enabled/default
touch /etc/nginx/sites-available/codebase
ln -s /etc/nginx/sites-available/codebase  /etc/nginx/sites-enabled/codebase

To /etc/nginx/sites-enabled/codebase add

server {
        location / {
                proxy_pass http://127.0.0.1:8000;
        }
}

And restart nginx with

/etc/init.d/nginx restart

Now, from inside codebase run

gunicorn app:app -b localhost:8000

If everything went right, if you visit http://33.33.33.33/ you should now see the “Hello World!” message. Congratulations! You’ve successfully set up your own server.

Update- I’ve written a follow-up to this post which covers Puppet, a really handy tool that’s comes packaged with Vagrant- you can find the post here.


Videodropper- Behind the scenes

2011-10-21


My previous post about my new web app- Videodropper, which lets you send Youtube videos to your Dropbox account got quite huge on Hacker News yesterday(At least, much more than I’d anticipated). So, I decided to write another post about Videodropper on how it works.

Videodropper is powered by Python, Flask, Redis and Celery. It is hosted on Epio(as you might have guessed from the ep.io subdomain). All the downloading is handled by youtube-dl and of course, the uploading is done using the Dropbox API.

Basics

What happens when you press the “Send to Dropbox” button is that Videodropper gets the Youtube video URL(playlists won’t work, more on that later), and it queues it up on a Celeryd process; when the download starts, another process also starts that monitors the size of the video file so that it doesn’t cross the upload limit. So, that’s two processes for one file download. Currently, Videodropper runs only four processes which means that it can process only two videos at a time. Up until now, these modest resources have served quite well, as the download and upload speed is pretty high. However, the Celery instance, which is limited to 128MB of memory runs out of memory when Videodropper starts getting a lot of requests.

When Celery starts a task which in this case is a “download, then upload”, youtube-dl starts downloading the Youtube video which, as I mentioned above happens quite fast. Then, if the size hasn’t exceeded Dropbox’s upload limit(which is 150MB now, but it should soon be upgraded to 300MB), the upload process begins. That’s basically the core functionality of the app. Oh, and if you choose to optimize the video for iPhone, youtube-dl simply downloads the .mp4 file(Youtube format=18) of the video, no transcoding takes place.

Size monitoring

One obvious optimization for Videodropper is the size-monitoring as you’ve seen that it eats up a whole process. One solution would have been to find out the file size before downloading, but I couldn’t find any way to do that(if you happen to know of any please, please tell me about it.)

Playlist support

I’ve got some requests for Videodropper to start supporting playlists which is quite a reasonable request, given that Playlists is a huge part of Youtube. However, at present supporting playlists is just not an option. As I’ve mentioned above, Videodropper runs on a very modest server configuration and downloading playlists would surely cause Videodropper to run out of memory. The limits on the server resources are mostly because Videodropper is a free app, so I am not able to invest much money on it, so if you’re interested in supporting Videodropper you could donate some money on Flattr to help upgrade Videodropper’s server capacity.

In a nutshell, Flask serves the website, Celery queues up the download tasks(and also a size monitoring task) and Redis is used mostly as a backend to Celery, but also for storing the “Recent Downloads” of the user. That’s it. If you have any query or suggestion feel free to shoot them at the comments below.


Videodropper

2011-10-20


If you’re constantly troubled by having to wait for Youtube to buffer videos, or want to keep some videos with you so you can watch them again and again without wasting bandwidth, then I’ve just recently released a web service that lets you download videos from Youtube. But there are plenty of websites that already do that, why do I need another one?. What Videodropper(yeah, that’s what its called) actually does is it sends videos to your Dropbox folder.


Why send it to Dropbox?

How do I use Videodropper?

Using Videodropper is really easy. After linking Videodropper with your Dropbox account, you just copy the URL of the video you want to send, paste it into Videodropper and then send it. A bookmarklet is also being worked on in order to further simplify this process.

Note that Videodropper is still in Beta, so some of your videos may not get through properly, in case you notice any thing strange kindly send an email to samratmansingh@gmail.com or reach me on Twitter @samratmansingh


Instamator is now open source

2011-09-14


Instamator, which I released just a few days ago is now open source. You can find the source code on Github. Instamator is powered by Flask and makes use of Redis.


Instamator

2011-09-08


I have been wanting to create a serious web app for some time now, and I think I just did. Well, to be more accurate its a pretty simple web service. Let me introduce to you Instamator, it lets you easily grab and use the photos you’ve liked on Instagram, the popular photo-sharing app(for iOS devices only- but you can check Inkstagram for a web interface).

How does it do that? It produces feeds(JSON, RSS and text) about the photos you liked on Instagram which you can use to easily download all your Instagram “likes” or integrate on your website or blog.

Instamator was inpired by Flickmator, a similar web service built by Thomas Pelletier. It is powered by Flask, Redis and hosted on ep.io.