Enabling Continuous Deployment of a Two-Tier AWS Architecture with GitHub & Terraform Cloud

Continuous Integration / Continuous Development (CI/CD) is the frequent delivery of applications by steamlining & automating development and operations processes. This method is beautifully illustrated as the key solution towards the end of the novel The Phoenix Project, which I summarized in a previous Medium article.

In this exercise we will create a highly available two-tier AWS architecture, and we will deploy this using Terraform Cloud as our CI/CD tool.

For this exercise, I needed:


A few key details are:

  • NAT Gateway: A NAT Gateway was added so that our Web Servers in private subnets could access internet for updates.
  • Web Servers: We are putting them in the private subnets so that they cannot be openly accessed by the internet. I therefore added a Bastion Host in an Auto Scaling Group (ASG), and we can SSH into our Web Servers from there. The capacity of the ASG that the Bastion Host is one, so that if it goes down it will be reproduced within the ASG.
  • Application Load Balancer: This is internet facing, so that traffic can be directed to our Web Servers. Also the web servers are in an ASG for high availability.
Image 1: Architecture

STEP 1: Fork Repository

Note that for this exercise, I updated my key_name in the root main.tf to my AWS EC2 Key Pair. Also please keep in mind that key pairs are unique to region, so make sure you are in the correct region when running this exercise!

STEP 2: Terraform Cloud

I then created a new workspace (Image 2) using Version Control in Terraform Cloud (Image 3).

Image 2: New workspace
Image 3: Version Control Workflow selected

I then connected to my version control provider which for this exercise is Github (Image 4).

Image 4: Connecting Terraform Cloud to my Github

I was prompted to confirm authorization to connect Terraform Cloud to my Github, which I did. Then I chose our previously forked repo (Image 5).

Image 5: Selecting the forked repo

I then named the Workspace (Image 6), and then created the workspace successfully bringing me to my new workspace (Image 7).

Image 6: Naming my Workspace “Week_20_Project_Terraform_Cloud”
Image 7: Arriving at the new Workspace

I then clicked the purple “Configure variables” box, and then added variables in Environment Variables. The variables I added were:

  • AWS Access Key, which I made sensitive (Image 8)
Image 8: Access Key
  • AWS Secret Access Key, which I made sensitive (Image 9)
Image 9: Secret Access Key
  • AWS Default Region = us-east-1 (Image 10)
Image 10: Default Region
  • Confirm Destroy (This is needed to later destroy our architecture. I set it to 1) (Image 11)
Image 11: Confirm Destroy

I then added the Access IP variable to the Terraform Variables as seen in Image 12. This was done rather than having a .tfvars file in the repo. I’m setting using for the sake of the example. Side Note: Terraform Variables & Environment Variables aren’t synonymous.

Image 12: Access IP

I then manually started the pipeline by clicking “Queue plan manually” (Image 13), and I gave the reason that this was a test.

Image 13: Queuing new Terraform plan

I received errors concerning my provider & I was admittedly stuck on this problem for a while (Image 14).

Image 14: Errors concerning provider

I explored the situation and considered a couple possible root causes. I checked that the Terraform version was up-to-date, scanned my forked files for typos, etc.

However, I had a suspicion that because “credentials” were so fundamental to the error, I would update go to AWS IAM and update new access and secret access keys. I then created a new workspace, because when you amend variables in Terraform Cloud it states that you cannot reuse them later on. I added new variables, and then re-ran the Plan.

Image 15 shows that after resolving the credentials error, I was back on track and the plan had finished.

Image 15: Confirming the error was solved

I was then ready to Apply, so I clicked “Confirm & Apply” which prompted me to make a comment before running, to which I typed “Testing.”

…But again I received an error. This time, I had reached my capacity for EIPs and VPCs in AWS (AWS offers you five & I had five in us-east-1).

After cleaning up my AWS environments and removing a few VPCs, releasing some EIPs, and clearing all dependencies, I was able to get a successful apply. You can see my resources in Images 16–18.

Image 16: Apply finished Part 1
Image 17: Apply finished Part 2
Image 18: Apply finished Part 3

Note that there were two changes (my launch templates were updated). Also there was 1 deletion (my load balancer needed to be deleted because I had one by the same name, and in turn a new one was made).

That concludes STEP 2: Terraform Cloud

STEP 3: Testing

Testing Application Load Balancer (ALB)

Image 19: Output

I then pasted the value into my web browser. I can confirm that the ALB is a successful load balancer because you can see when I refresh the page that the “Hello World” alternates web servers and the IPs aren’t the same. (Image 20 & 21).

Image 20: ip-10–0–3–156.ec2.internal
Image 21: ip-10–0–1–238.ec2.internal which is not identical to Image 20

Testing Bastion Host

To SSH into my Bastion Host, I first went to my EC2 Dashboard and identified the Bastion Host as it has the appropriately titled security group (Image 22).

Image 22: Identifying the Bastion Host

I then clicked on the Instance ID & selected the SSH Client Tab (Image 23).

Image 23: SSH Client Tab

I then changed directory in Terminal to where my .pem file for MyPrivateKey was, which was in /downloads (Image 24).

Image 24: Changing Directory to /downloads

I then used the command ls -ltr to see my files in /downloads, where -l longlists, -t sorts entries by modification date with newest first, and -r reverses the sorting order (Image 25).

Image 25: Identifying MyPrivateKey.pem

For security and to ensure my key isn’t publicly viewable, I then changed the permissions for the file using chmod 400 MyPrivateKey.pem (Image 26) & confirmed that it was now read-only by using ls -ltr again (Image 27).

Image 26: Changing permission to read-only
Image 27: Confirming it was read-only

At this point the file was more secure, so I then copied the command from the SSH Client tab in AWS, removed the quotations around the file name, and ran it in my Terminal. I was asked if I wanted to continue to which I typed “yes.”

I then received the EC2 Amazon Linux 2 AMI image, and I used the command pwd to ensure I as in /home/ec2-user which I was. All of this is in Image 28.

Image 28: SSHing into Bastion Host confirmed

If your connection is blocked or timing out, review these possible problems to troubleshoot. It is likely a firewall, the SSH service isn’t running on the instance, or Port 22 isn’t in the listening state. Also it could be that you incorrectly identified the bastion host and are accidentally trying to SSH into one of the other instances.

At this point I had confirmed my architecture.



Now we have the opportunity to continually integrate and develop our environment by altering the code in Github (not best practice because it can get messy), or committing new code to the repo which will automatically run in Terraform Cloud.


Image 29: Destroying
Image 30: Destroyed

Thanks for reading.

Credits: Level Up In Tech pointed me in the direction of this lab. Please get in contact with me if interested in learning from my coaches & community.

Cloud, DevOps, Blockchain.