golangapp VS terraform-aws-ssh-bastion-service

resource "null_resource" "instance" { depends_on = [ aws_instance.main ] connection { type = "ssh" host = aws_instance.main.public_ip user = "ubuntu" private_key = file("${path.module}/main_key.pem") } provisioner "remote-exec" { inline = [ "sleep 10", "sudo apt update -y", "sudo apt install git -y", "sudo apt install golang -y", "sudo git clone https://github.com/viveksiddhartha/golangapp.git /home/ubuntu/golangapp", "cd golangapp", "sudo go build ./", "./main", "sleep 10" ] } }

#!/bin/bash sudo apt update -y sudo apt install git -y sudo apt install golang -y sudo git clone https://github.com/viveksiddhartha/golangapp.git /home/ubuntu/golangapp cd ./golangapp sudo go build ./ ./main

I'm not sure from this exactly what you're trying to do. If you're working with cloudinit/userdata then I can point you to an example.

I actually mashed together cloudposse/terraform-aws-dynamic-subnets (which has an option to use NAT instances instead of gateways) and joshuamkite/terraform-aws-ssh-bastion-service. That allowed for even cheaper setups where your bastion jump-host (the host you proxy through to get to the private subnets) is also the NAT gateway. I quite liked this setup because the bastion sessions were sufficiently sandboxed in a ephemeral container, and we didn't need twice the number of hosts to have both NAT instances and bastions.

I have a public terraform module which i migrated away from the template provider last year. You can see how I handled the changes and used the new provider if that helps

You should try to avoid using remote exec with terraform wherever possible. It's not a great fit with the declarative approach that Terraform aims for. Here you would be better off using 'user data' to provision your instance. I have a public project that deploys and builds a Golang app on EC2. It's a bit more complicated than your example here, but it does demonstrate a method for doing what you're trying to do with the approach that I have suggested.

The thing that really moved me on from junior to senior in this area was finding something that was a common pain point for many people that I could focus on to address as a 'simple' solution that could be used as a 'black box' or 'tick box'. Once it picked up steam I had to deal with other people's use cases and suggestions that I had never considered before, both in person (when i spoke at meetups) and online (open source project).

It looks like you might be misunderstanding the purpose of the yamlencode function here. Terraform doesn't need to know anything about the syntax of the file you're using with templatefile, as long as it's text, you're good. Just start with a yaml file and use the templatefile syntax for your substitutions. Examples

I have a public Terraform module with releases incrementing from tf 11 thru 1.2.0 where you can see changes dealing with this sorry of thing

I'm well aware. My repo that I linked is in multi part mime. An older version used different mime types but I couldn't get it to handle all the conditional logic reliably

I haven't seen your code but it sounds like you're using it wrong. I can well imagine that if you apply multiple cloudinit_config files to a single host that they get placed in alphabetical order but I don't really get why you would want to do that. In many environments, e.g. AWS, you can only supply a single user data file to your host anyway. cloudinit_config supports mullti-part MIME encoding so that you can combine sections from multiple sources in a predictable way.

You can see terraform rendering IAM policies with interpoalted assume role policies in my module.