Best practices for close-to-zero downtime deployments?

onecrayon

I'm working to build out a Github Actions-based continuous delivery flow for deploying a Python web application to Opalstack when merging into the main branch, and am wondering if there are any best practices that folks can recommend to achieve as close to a zero-downtime deployment as is possible with a non-containerized webhost (previously this was getting auto-deployed to Render).

It looks like I will probably want to write a shell script to live alongside my start and stop scripts, and then execute it remotely via SSH from Github Actions, and there are effectively four things that script will need to do:

Pull the latest code from the main branch
Install/update Python dependencies based on a lock file
Apply database migrations
Restart the webserver

I can do all of that via shell script without a problem, but am wondering if anyone has crafted a flow like this and has recommendations for what worked and what didn't. Like, can I just do this all within the application folder on disk? Or is it better to clone a separate copy, stop the server, then swap the folder names and start it again? I'm not a devops person, so any advice (particularly as applies to Opalstack) is welcome!

sean

onecrayon one approach would be to use a releases directory to hold recent releases of your code, with a current symlink pointing to the most recent release, eg:

home
└── shelluser
    └── apps
        └── name_of_app
            ├── current -> /home/shelluser/apps/name_of_app/releases/20220120114500/
            └── releases
                ├── 20220080072500
                ├── 20220090083000
                ├── 20220100093500
                ├── 20220110104000
                └── 20220120114500

Your script would pull the code into a dated directory under releases and then update the current symlink to point to the new release and run the code from there. This lets you quickly roll back to a previous release if you need to.

This is the approach used by Capistrano, and I've seen it used in a few home-grown setups here and there on our service (and at WebFaction in the past).

Hope this helps 🙂

k4ml

This is the script that I use to deploy one of our site at opalstack:-

#!/bin/bash
set -euo pipefail

project_org=myorg
project_name=my-site

(
    cd versions
    git clone "git@github.com:${project_org}/${project_name}.git" $project_name
)                              
version=$(cd "versions/${project_name}" && git rev-parse --short HEAD)
app_dir="versions/$version"    
if [ -d $app_dir ]; then
    echo "$version already exists"
    exit 1
fi
(
    cd versions && mv $project_name $version
)
(
    cd $app_dir
    ln -s ~/etc/${project_name}/env .env
    POETRY_VIRTUALENVS_IN_PROJECT=true poetry install
    #npm install && npm run prod
    .venv/bin/$project_name manage collectstatic
)
echo "Current version:"
if [ -d $project_name ]; then
    ls -l $project_name
    rm $project_name
fi
ln -s $app_dir $project_name
echo "Deploy done, please run ./stop && ./start."