#11 Improving deploys

November 07, 2022
See the code for this post on the improving-deploys branch.

The current deployment setup needs some work. First, we currently have to copy all of our code manually via scp. Second, we have to manually restart the server for our deployment to take effect.

Wouldn't it be great to be able to do all this with a single command? Not only that - can we actually monitor a running deployment from our local machine? That's what we're going to be looking at today.

Remove the app directory in our production server to start fresh:

rm -rf app

We are going to initialize a Git repository directly on the server. During a deployment, we will pull our code from GitHub. This will allow us to immediately roll back to the previous version if we experience issues in production.

To start, ensure Git is installed. If it isn't, follow these steps. Then clone the repository.

git clone git@github.com:jurajmajerik/server.git

The following section might be controversial. To achieve a single-command deploy, we want to avoid typing sudo every time we start or stop a command that requires it. Here, I'll show the way to disable the password prompt. Opinions vary on whether this is a good idea from the security point of view. I'm not an expert, but we obviously want to avoid typing the password during our deploys. So at this point, I'm just gonna go ahead and turn off the prompt. I will update this section if I learn more about this topic and/or find a safer way to do this.

Open the sudoers file with sudo visudo and add the following line at the end. juraj is the name of my user account for which I want to disable the password prompt.


Then reboot the server for changes to take effect.

sudo reboot

One more thing before we start with the deploy script. So far, we've been running the server code with go run <module_name>. This command first compiles the Go program into binary on the fly, and then runs it. I am now going to introduce a build step that will compile and save the binary during deployment. At the end of this step, we will have a binary that we can run directly. This has two main advantages - restarting the server will be faster (should we need to) and process management (e.g. killing the process) is more straightforward. Read more about this topic.

During the deployment, we will compile our code with the go build command. Before that, we need to initialize our app module. In the home directory of our production server, run:

go mod init app

This creates the go.mod file that tells Go which version of the language the module is using.

Let's now transition back to our local machine where we're going to write the bash scripts that will handle the deployment. We need to create two files in the app directory.

prod_deploy.sh will run on the production server and handle the actual deployment. Here's the full script:

cd $HOME/app
msg () {
echo -e "$1\n--------------------\n"
msg "Stopping app"
sudo pkill app
msg "Pulling from GitHub"
git pull
msg "Building Go binary"
go build
msg "Starting server"
nohup sudo ./app &>/dev/null &
msg "Deploy finished in $(($duration % 60)) seconds."
msg "Press Enter to exit"

The script first stops the running server, pulls new code from the GitHub repository, compiles server code into a binary, starts the process and shows a summary. We're also tracking the deploy time using the special $SECONDS variable in bash. In the end, we output the success message and prompt a user for input with a read command (more on that in a second).

Normally, a program terminates after the user logs out. To avoid this, we can start the process with the nohup command. nohup ensures the program will keep running even after we log out. It does so by preventing the signal messages from reaching the program. By default, nohup writes its output to nohup.out. To prevent cluttering our server directory, we redirect the output to /dev/null. We also add the & at the end to run the process in the background.

Finally, we add the deploy.sh script that will run on our local machine. It will kick off the prod_deploy.sh script and monitor its progress from our local terminal.

sshcmd="ssh -t juraj@app.jurajmajerik.com"
$sshcmd screen -S "deployment" /home/juraj/app/prod_deploy.sh

The script logs into our server with ssh. It then kicks off a new screen and gives it the name deployment. We are using the screen command to monitor the deployment on our production server. screen starts a new terminal session and persists it on our production server. We can pull up an existing screen at any point. If our connection breaks down and we get logged out, we can simply log back in and grab the "deployment" screen. This allows us to check on our deployment if our connection gets interrupted.

Once the deployment finishes, the program would normally terminate. At that point, we would lose any useful info about the finished deploy. This is why we've added the read line at the end of the script. This command simply waits for any input from the user. The screen will terminate once we press Enter.

Once we're done, we simply need to run chmod +x <filename> on both deploy scripts to make them executable. Then we can push our changes to GitHub. Before our first deploy, we need to log into our server and manually pull the changes - the server doesn't have the deploy script yet! Afterwards, all we ever have to do is to run deploy.sh on our local machine:


And we can monitor the progress of our deployment live!

Stopping app
Pulling from GitHub
Already up to date.
Building Go binary
/home/juraj/app/prod_deploy.sh: line 17: go: command not found
Starting server
Deploy finished in 2 seconds.
Press Enter to exit

n.b. I don't know the reason for the /home/juraj/app/prod_deploy.sh: line 17: go: command not found message. I'll update the above section once I find out.

See the code for this post on the improving-deploys branch.