We recently put a Concourse CI instance behind an authenticated ALB in AWS, to make sure there are two distinct logins happening without having to resort to using a bastion host.
The way this works is that whenever you try to access the URL assigned to Concourse, there is a redirection that sends you to the identity provider. Once you log in there (using, hopefully, MFA), a cookie named AWSELBAuthSessionCookie-0
is set in your browser and then you will reach the tool again.
The code to do this is fairly straightforward. You set up a new app in your IdP
, from which you get a client_id
and a client_secret
. Then set up a check in the ALB using terraform
resource "aws_lb_listener" "web-alb" {
load_balancer_arn = "${aws_lb.web-alb.arn}"
port = "443"
protocol = "HTTPS"
ssl_policy = "ELBSecurityPolicy-TLS-1-2-2017-01"
certificate_arn = "${aws_acm_certificate_validation.web.certificate_arn}"
default_action {
type = "authenticate-oidc"
authenticate_oidc {
authorization_endpoint = "${var.idp_url}/oidc/auth"
client_id = "${var.client_id_oidc}"
client_secret = "${var.client_secret_oidc}"
issuer = "${var.idp_url}/oidc"
token_endpoint = "${var.idp_url}/oidc/token"
user_info_endpoint = "${var.idp_url}/oidc/me"
}
}
default_action {
type = "forward"
target_group_arn = "${aws_lb_target_group.web-alb.arn}"
}
}
However, this has an unintended consequence. fly
also accesses Concourse CI through that URL, and it cannot deal with the redirection. You are not supposed to trigger many things locally, as that should be covered by the pipelines themselves. You still need to upload a pipeline for the first time when you create a new one, and there are commands like intercept
that you also want to use, so it is nice to have this option open in this setup.
fly
itself does not help much, so we had to find a workaround. What we did was building a local proxy that forwards the request from fly
to the ALB and injects the cookie in the request to make it work. The proxy is simply nginx
with a parametrized configuration, running inside a Docker
container. One interesting thing to note is that we have to set up some headers to that intercept
works as well, as it relies on websockets.
events {
worker_connections 16;
}
http {
server {
listen ${CONCOURSE_PROXY_PORT};
server_name localhost;
location / {
proxy_pass ${CONCOURSE_URL};
proxy_set_header Cookie AWSELBAuthSessionCookie-0=${COOKIE};
# Fix `websocket: bad handshake` when using `fly intercept`
proxy_set_header Upgrade $http_upgrade;
proxy_set_header Connection "upgrade";
}
}
}
This configuration is converted to a valid one in the CMD
of the container, like this:
#! /usr/bin/env bash
set -e
# shellcheck disable=SC2016
envsubst '$COOKIE $CONCOURSE_URL $CONCOURSE_PROXY_PORT' < /tmp/nginx.conf.template > /etc/nginx/nginx.conf
nginx -g "daemon off;"
We run the container using docker-compose
and provide the COOKIE
in an .env
file.
Connecting to Concourse
Once the proxy is running, we just need to log in with fly
. We do that from our go
script directly:
goal_login-to-concourse() {
fly --target aftersales login --concourse-url "http://localhost:$CONCOURSE_PROXY_PORT"
}
Here is where we see one of the pain points of this solution. When you try to log in, you see something like this:
navigate to the following URL in your browser:
http://localhost:3232/sky/login?redirect_uri=http://127.0.0.1:50449/auth/callback
or enter token manually:
that URL won’t work. You need to replace the localhost
part with the actual URL of the Concourse server. We could not really figure out a way to make this a bit more user friendly, so we just need to deal with it.
Getting the cookie is also a bit cumbersome, as you need to get it manually from your browser and put it in the .env
file. I experimented a bit getting this cookie programatically, but I could not get a decrypted cookie from Chrome, so I just left at this for now.
And that’s basically it. It is a bit of a hack, to be honest, but it does the job without having to compromise on the two logins setup that we were after.
Top comments (0)