Migrating a website to DigitalOcean
Posted: 2020 January 3
Last updated: 2020 January 6
My goal is to migrate my recipe website (currently called Lazy Baker, but subject to change if I ever want to get it its own domain) from Heroku to DigitalOcean. There are a couple reasons why:
- Heroku’s free tier doesn’t support SSL, so Lazy Baker is technically insecure (not that there’s anything worth stealing or messing with).
- Now that a few other people occasionally look at this website, it sometimes runs out of free tier hours in the last few days of the month, at which point I can no longer access my own recipes.
- I’m already paying for a DO droplet, so I might as well make use of it.
There are a couple of obstacles that have made me drag my feet on this transition up to this point.
- I need to figure out how to set up my droplet (what DO calls a VPS) to support multiple sites. (After a bit of frustrated Googling, I’ve discovered that this is called Server Blocks on an Nginx server.) (Instructions here)
- I don’t remember how to set up a Django site on DO, especially when worrying about environment variables and static content hosted on AWS S3. (I did it once for the blog website but didn’t write it down. This time I’m writing it down.)
- I need to also transfer the website’s database from Heroku to DO.
- I’m pretty sure my DO droplet needs a bunch of updates/security patches that I don’t want to face.
There are a couple other things I’d like to add to the server while I’m at it:
- Automatically pulling any changes that are pushed to master.
- Running an install/update script to integrate the changes without manual intervention (more like what Heroku does.)
First, I need to get in to the droplet. For some reason, I do still have it set up to allow SSHing into the system as root with the password instead of an authorized SSH key. But of course, I don’t remember the password. Luckily, DO has a tutorial for adding an SSH key (I did it manually, using Access > Console access from the DO website.) Now I can log in as a user (currently I have the user
tribune for the blog), but for some reason still not as root (probably because the password option is enabled?) Luckily, it turns out I know the password for the
tribune user, which has admin permissions, so I can still use
sudo. This is already off to a hacky start.
Time to install the 105 updates (oof) to my Ubuntu 16.04 server. Here’s hoping it doesn’t break anything. At least 16.04 is an LTS and still supported for awhile. (LPT: don’t run a server on a non-LTS release of Ubuntu.)
Since I currently have my single website running within a user account, the first step seems like setting up a new user for the new site. (I’m following this guide.)
- Create a new user:
sudo adduser recipes
- Add it to the sudo group:
usermod -aG sudo recipes
- Log out and SSH in as the new user:
(You’ll need to use the password because SSH stuff isn’t set up yet)
- Generate an SSH key for the new user:
- From your local machine, copy its public key to the authorized_keys for the new user on the server using one of these methods. (I ended up doing it manually.)
You should now be able to SSH in without your password. For some reason it’s still attempting to use the password for this new user (and failing at the SSH key), despite the fact that my local publickey does show up in the new server user’s
~/.ssh/authorized_keys file, and the permissions are correct. It’s not a general issue with SSH key logins, since I can log in with my SSH key to the user I created long ago. Restarting the server didn’t solve the problem, nor did editing
/etc/ssh/sshd_config to disallow login by password(
PasswordAuthentication no) – after a reboot, that just prevents me from logging in at all (
Permission denied (publickey).) I do want to figure this out (it’s best not to have password login enabled), but for now at least I can get in.
While I’m at it, for security, I modified
PermitRootLogin no, to disallow root login.
Update: It turns out that
ssh-copy-id copied the server’s public key to its own
authorized_keys, which wasn’t what I wanted. I did it manually and it works now.
I already have the firewall set up, so I can skip that step.
Now I can move on to setting up the Django stuff (following this tutorial).
- Install Python3:
sudo apt-get update sudo apt-get install python3-pip python3-dev libpq-dev postgresql postgresql-contrib nginx
(In my case, they were already installed.)
- Set up a database for the project:
- Log into a postgres session:
sudo -u postgres psql
- Create a databse for the project:
CREATE DATABASE recipes_db;
- Create a postgres user for the database:
CREATE USER recipes WITH PASSWORD 'password';
- Set up user to play well with Django:
ALTER ROLE recipes SET client_encoding TO 'utf8'; ALTER ROLE recipes SET default_transaction_isolation TO 'read committed'; ALTER ROLE recipes SET timezone TO 'UTC';
- Give the user permissions for the database:
GRANT ALL PRIVILEGES ON DATABASE recipes_db TO recipes;
- Quit postgres with
- Log into a postgres session:
- Set up Python virtual environment
- Update pip3:
sudo -H pip3 install --upgrade pip sudo -H pip3 install virtualenv
- Create a virtual environment in the home folder:
cd ~ virtualenv venv
- Activate the virtual environment and check that it’s correct:
If you run
which python, you should see
/home/recipes/recipes/venv/bin/python. If you run
python --versionyou should see that it’s Python 3.
- Update pip3:
Now we diverge from the tutorial because we’re not creating a new project from scratch; it already exists in a repository with its dependencies specified.
- Add server user’s SSH key to Github (so I can clone):
- On the server, copy the output of
- On Github, go to Settings > SSH and GPG keys > New SSH key, and paste the public key.
- On the server, copy the output of
- Clone the repository into the user home folder:
git clone email@example.com:jtebert/lazy-baker.git
- With the virtual environment activated, install the Python dependencies:
pip install -r lazy-baker/requirements.txt
- Within the repository, create a configuration file called
touch settings.ini) and add the following:
[settings] DJANGO_SETTINGS_MODULE=recipe_box.settings.production SECRET_KEY=XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXxx AWS_ACCESS_KEY_ID=XXXXXXXXXXXXXXXX AWS_SECRET_ACCESS_KEY=XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
(filling in the real information, of course)
- Install any system dependencies. (In my case they were all installed already.)
- Migrate the databse to get it set up the first time:
python manage.py migrate
- Create a superuser to be able to actually use/log into the website:
python manage.py createsuperuser
- Add a rule to make your port externally visible:
sudo ufw allow 8000
- Try running the server:
python manage.py runserver 0.0.0.0:8000
You should now be able to view it at
http://IP_ADDRESS:8000(Note: in the this particular case, the main page doesn’t exist yet so you’ll get an error. But you can go to the
/adminpage and see if it works.)
I ran into a couple of issues here. First, I perviously only partially implemented my usage of python decouple for using the settings file, so I needed to move some things to there (like the database name, and adding the database username and password) to get it to work. I also intended to switch to using a single
settings.py file, instead of the way that Wagtail handles things by default (a settings folder containing
production.py). I still had the secondary settings files around (unused), so I removed them, moved
base.py to the parent folder as
settings.py, and got rid of the settings folder. I then needed to change the contents of
manage.py to get the environment from
recipe_box.settings instead of
… But then I was still getting an error that
password authentication failed for user "'recipes'". Notice this nested quotes? Eventually I figured out that the problem was that I put everything in my
settings.ini file in single quotes, which you don’t do for that, apparently.
I then switched over to this tutorial on ngninx server blocks to set up a new separate server block for Lazy Baker. But it’s hard to test it as is. (This is going to turn into a pain in the ass to debug, if it doesn’t magically work on the first try.)
Then I hopped back to the previous tutorial and continued on with Gunicorn (which also needed
recipe_box.settings.dev changed to
wsgi.py). Following through on the nginx part, the file it has me create in
/etc/nginx/sites-available doesn’t completely look like the one I have for the blog, but then again, the way I did that didn’t have me setting up a gunicorn service, either.
In order to see if any of this works, though, I do actually need to move over the URL. Currently, my domain registrar has an
A record named
blog pointing at the IP address. Since Nginx is supposed to handle the domain routing, I think I should just be able to add another
A record named
lazybaker pointing at the same IP address and it will magically work. (That never happens.) But now I need to wait for the record to update to tell if anything worked. (Update: that part actually worked after very limited debug – I’d put the wrong path to the
.sock file in the nginx site configuration.)
Right now I can load the admin at
lazybaker.juliaebert.com, but the site has no content, so I still get a 500 error on the home page. Time to steal the database from Heroku.
I have the Heroku CLI installed on my local machine, so I’ll just use that and then copy the dump with
scp (because I’m lazy).
- Create a snapshop:
heroku pg:backups:capture -a lazybaker
- Download it as a
heroku pg:backups:download -a lazybaker
- Copy it to server:
scp latest.dump recipes@IP_ADDRESS:/home/recipes/
- Restore it to the server database:
pg_restore --verbose --clean --no-acl --no-owner -h localhost -U recipes -d recipes_db latest.dump
Note that this will overwrite anything that you have in the database!
- Restart the Django server:
sudo systemctl restart gunicorn
Now everything shows up in the admin when I log in, but I’m still getting a 500 error on the home page! Turning on Debug again (in production, sorry), it’s saying it can’t find the template. After local debugging, it turns out that the issue came from relative paths: when I moved the settings file out of a subdirectory, the
BASE_URL reference in that file was then one level too high, so it couldn’t find the templates. Honestly, I’m surprised that’s the only thing I saw failing from that.
The website now works (including, miraculously, the AWS stuff), but it’s not secure (HTTP only, no HTTPS). To do this, I need to add an SSL certificate (which are no conveniently free and easy with Let’s Encrypt). DigitalOcean even has easy integration… as long as you’re using their nameservers for managing your domain. Since I’m absolutely overloading my use of the juliaebert.com domain, this isn’t a good option; I wouldn’t be able to keep doing everything else with it.
But since I already did this for blog.juliaebert.com, I know it’s possible to do this more manually. And DigitalOcean has a tutorial for that.
The first step is installing Certbot, which is system-wide and therefore already done for the previous site. Similarly, steps 2 and 3 are already covered by what I’ve done so far. That means it’s just a matter of running one command to get it configured:
sudo certbot --nginx -d lazybaker.juliaebert.com
There’s one prompt: whether you want to redirect HTTP traffic to HTTPS. I picked the redirect here, because I don’t think there’s any reason to let the non-secure stuff hang around.
It also provides a convenient link to check your configuration for your URL (check mine out here – it got an A!)
You can also check the automatic renewal with a dry run, using a single command:
sudo certbot renew --dry-run
Everything just worked, which is pretty amazing! I guess I didn’t remember how I did it for the previous website because it’s just so straightforward and painless.
I ran into trouble uploading images (getting a 413 error), because it turns out the default file upload size limit for Nginx is 1 MB. It’s easy to change, though (as explained here). In
/etc/nginx/sites-available/lazybaker.juliaebert.com, I added the following line to the main
server section to increase the limit to 10 MB:
Then restart Nginx with
sudo systemctl restart nginx and you’re good to go.
Now, another security note: I don’t feel comfortable with my database password in plain text in the
settings.ini file. From finding Google results like this Medium post I felt less bad about it by restricting access to the file by limiting its permissions:
chmod 600 settings.ini
… And then, after doing all of that setup, I decided to get a new, real domain name for this site: meet Reckless Ham. Which means I now need to go back and modify all the URL/website name related stuff on my server.
With the domain name
recklessham.com purchased, I started by setting an
A record in the DNS settings to point the apex domain (
@) to my server’s IP address.
Now to change/fix all of my names/references. First, finding all the files named this way:
sudo find / -name "*lazy*"
Then finding all the files containing references to this name/these files:
grep -iRl "lazy" /var/www grep -iRl "lazy" /etc/nginx
(Not in any useful order)
- Migrate to the new URL (and change all associated issues/names)
- Redirect www to apex domain
- redirect lazy baker to reckless ham
- Deal with all my many lingering GitHub issues
- Fix favicon (change to reckless ham)
- Update references to the website elsewhere (broken links)