Ansible Tutorial: Setup HAProxy and HTTPS from Let’s Encrypt with Ansible

In the third article of this series, I set up Docker, MySQL and WordPress with Ansible on my server. In this fourth and final article, I will show you how to set up HAProxy – again with Ansible – as well as a free HTTPS certificate from Let's Encrypt / CertBot to make the website accessible via HTTPS.

The article code can be found in my GitLab repository at https://gitlab.com/SvenWoltmann/happycoders-tutorial-server-setup.

In addition to the article, there is again a (German-only) video tutorial: Setup of HAProxy and a HTTPS certificate from Let's Encrypt with Ansible

The article series is divided into the following four parts:

• Setup of a dedicated server with Ansible
• Setup of user accounts, SSH and firewall with Ansible
• Setup of Docker, MySQL and WordPress with Ansible
• Setup of HAProxy and a HTTPS certificate from Let's Encrypt with Ansible

Installing HAProxy

In the first step, I will install HAProxy without specific configuration because I want to use the configuration file haproxy.cfg created in the process as a starting point for my further configuration. However, I don't want to install version 1.7 available under Debian Stretch but the latest version 1.8 because it supports the modern HTTP/2 protocol. I get this version from the stretch-backports repository.

I create a role, happy1_haproxy with the task definition file roles/happy1_haproxy/tasks/main.yml, and I add the entry happy1_haproxy to the array roles in my playbook happy1.yml so that the new role is executed the next time the playbook is executed. In the main.yml, I enter the following two tasks:

---
- name: Add "stretch-backports" to sources (required for HAProxy 1.8)
  apt_repository:
    repo: deb http://httpredir.debian.org/debian stretch-backports main
    state: present
    update_cache: yes

- name: Install HAProxy
  apt:
    name: haproxy
    state: present
    default_release: stretch-backports
    cache_valid_time: 3600
Code language: YAML (yaml)

With the Ansible module apt_repository, I add the backports repository to APT. In the apt module, I then use the default_release parameter to specify that the backports repository should be used to install HAProxy.

I'm running the playbook:

ansible-playbook --ask-vault-pass happy1.yml
Code language: plaintext (plaintext)

That docker-compose up -d reports the status changed can be ignored. This is because the command is executed through the Ansible module shell, which does not interpret the output of the command (which is: "wordpress_db_1 is up-to-date, wordpress_wordpress_1 is up-to-date") as "no change".

Configuring HAProxy

I now copy the HAProxy default configuration file /etc/haproxy/haproxy.cfg from the server to the local template directory for my role happy1_haproxy as roles/happy1_haproxy/templates/haproxy.cfg.j2. I copy the file to the template directory and not the file directory because I first want to protect the website with a password – but want to make it easy to disable this with a variable later. Now I extend the configuration with the following three sections:

Section frontend – here I define that HAProxy is bound to port 80 and should forward all requests to the backend happycoders_wordpress:

frontend happycoders_80
    bind *:80
    default_backend happycoders_wordpress
Code language: Nginx (nginx)

Section backend – here the previously referenced backend happycoders_wordpress is defined, which should forward the requests to the local port 8001, on which the Docker WordPress container is running.

backend happycoders_wordpress
    mode http
    server server1 localhost:8001 check
    http-response del-header x-powered-by
{% if happycoders_testing|default(false) %}
    acl authorized http_auth(happycoders_test_users)
    acl wp-cron path_beg -i /wp-cron.php
    http-request auth realm HappyCoders.eu unless authorized or wp-cron
{% endif %}
Code language: Nginx (nginx)

The server name server1 is arbitrary, and the check option ensures that regular health checks are performed. With http-response del-header, I remove the version information returned by PHP as a response header for security reasons. The remaining lines activate basic authentication if the variable happycoders_testing is set to true. In detail:

• The ACL authorized checks whether the user is already authorized.
• With the ACL wp-cron, I make sure that WordPress can call its own scheduler. It is important that this ACL is defined via the path /wp-cron.php, not via the source IP address because – when the scheduler is called – that call's source IP is not the server's address, but the address of the Docker container.
• With the directive http-request auth, the authorization is finally requested, unless the user is already authorized or the request goes to the WordPress scheduler.

The last section userlist lists the user(s) who should have access via basic authentication. My password is encrypted with mkpasswd:

I enter the user list as follows:

{% if happycoders_testing|default(false) %}
userlist happycoders_test_users
    user sven password X/UgQ53BouggY
{% endif %}
Code language: Nginx (nginx)

The finished file roles/happy1_haproxy/templates/haproxy.cfg.j2 now looks like this (I combined the two if blocks into one):

global
    log /dev/log    local0
    log /dev/log    local1 notice
    chroot /var/lib/haproxy
    stats socket /run/haproxy/admin.sock mode 660 level admin expose-fd listeners
    stats timeout 30s
    user haproxy
    group haproxy
    daemon

    # Default SSL material locations
    ca-base /etc/ssl/certs
    crt-base /etc/ssl/private

    # Default ciphers to use on SSL-enabled listening sockets.
    # For more information, see ciphers(1SSL). This list is from:
    #  https://hynek.me/articles/hardening-your-web-servers-ssl-ciphers/
    # An alternative list with additional directives can be obtained from
    #  https://mozilla.github.io/server-side-tls/ssl-config-generator/?server=haproxy
    ssl-default-bind-ciphers ECDH+AESGCM:DH+AESGCM:ECDH+AES256:DH+AES256:ECDH+AES128:DH+AES:RSA+AESGCM:RSA+AES:!aNULL:!MD5:!DSS
    ssl-default-bind-options no-sslv3

defaults
    log     global
    mode    http
    option  httplog
    option  dontlognull
    timeout connect 5000
    timeout client  50000
    timeout server  50000
    errorfile 400 /etc/haproxy/errors/400.http
    errorfile 403 /etc/haproxy/errors/403.http
    errorfile 408 /etc/haproxy/errors/408.http
    errorfile 500 /etc/haproxy/errors/500.http
    errorfile 502 /etc/haproxy/errors/502.http
    errorfile 503 /etc/haproxy/errors/503.http
    errorfile 504 /etc/haproxy/errors/504.http

frontend happycoders_80
    bind *:80
    default_backend happycoders_wordpress

backend happycoders_wordpress
    mode http
    server server1 localhost:8001 check
    http-response del-header x-powered-by
{% if happycoders_testing|default(false) %}
    acl authorized http_auth(happycoders_test_users)
    acl wp-cron path_beg -i /wp-cron.php
    http-request auth realm HappyCoders.eu unless authorized or wp-cron

userlist happycoders_test_users
    user sven password X/UgQ53BouggY
{% endif %}
Code language: Nginx (nginx)

To enable basic authentication, I add the following line to my host configuration file host_vars/happy1.happycoders.eu:

happycoders_testing: true
Code language: YAML (yaml)

To copy the configuration file to the server, I add the following task to the file roles/happy1_haproxy/tasks/main.yml:

- name: Configure HAProxy
  template:
    src: haproxy.cfg.j2
    dest: /etc/haproxy/haproxy.cfg
    owner: root
    group: root
    mode: 0644
  notify:
  - Restart HAProxy
Code language: YAML (yaml)

I already explained the template module and the notify function in the second part of this article series. For this, I need the corresponding handler in the file roles/happy1_haproxy/handlers/main.yml:

---
- name: Restart HAProxy
  service:
    name: haproxy
    state: restarted
Code language: YAML (yaml)

I'm running the playbook again:

ansible-playbook --ask-vault-pass happy1.yml
Code language: plaintext (plaintext)

The website is now directly accessible via port 80 via the URL http://happy1.happycoders.eu. Port 8001 can now be closed in the firewall by removing the corresponding entry of the array firewall_allowed_tcp_ports from the file host_vars/happy1.happycoders.eu. I run the playbook again and then restart Docker (sudo service docker restart), because – as described in the previous article – the Docker rules were removed by changing the firewall rules.

Installing Certbot

Now to the exciting part of the article: Next, I want the page to be accessible via HTTPS. I have to bind HAProxy to HTTPS port 443 and forward it to port 8001, and I have to install an SSL/TLS certificate. I get an SSL/TLS certificate for free at Let's Encrypt. The certificate request runs automatically via Certbot. Certbot can also install the certificates automatically in Apache and nginx. For the installation in HAProxy, a separate plugin is required, which is offered by Greenhost: HAProxy plugin for Let's Encrypt's Certbot

First of all, Certbot must be installed. I will follow the instructions at https://certbot.eff.org/lets-encrypt/debianstretch-haproxy to do this. As always, I'll transfer them into an Ansible role. I don't create a separate role for Certbot, but add the installation steps to the happy1_haproxy role because the configuration of Certbot is very closely linked to the configuration of HAProxy.

However, this time I would like to make things a bit more orderly right at the start, and I move the first three commands for installing and configuring HAProxy from roles/happy1_haproxy/tasks/main.yml to roles/happy1_haproxy/tasks/install_haproxy.yml and write the following into main.yml instead:

---
- import_tasks: install_haproxy.yml
- import_tasks: install_certbot.yml
Code language: YAML (yaml)

I create the file roles/happy1_haproxy/tasks/install_certbot.yml with the following two tasks to install Certbot:

---
- name: Add "stretch-backports" to sources (required for Certbot)
  apt_repository:
    repo: deb http://httpredir.debian.org/debian stretch-backports main
    state: present
    update_cache: yes

- name: Install Certbot
  apt:
    name: python-certbot
    state: present
    default_release: stretch-backports
    cache_valid_time: 3600
Code language: YAML (yaml)

According to instructions for the HAProxy plugin for Certbot, the following command should now be executed:

openssl dhparam -out /opt/certbot/dhparams.pem 2048
Code language: plaintext (plaintext)

For this, there is the Ansible module openssl_dhparam, which I enter into the install_certbot.yml as follows. But first I have to create the target directory /opt/certbot:

- name: Create Certbot directory
  file:
    path: /opt/certbot
    state: directory
    owner: root
    group: root
    mode: 0755

- name: Generate 2048 bit dhparams.pem file
  openssl_dhparam:
    owner: root
    group: root
    mode: 0644
    path: /opt/certbot/dhparams.pem
    size: 2048
Code language: YAML (yaml)

Furthermore, as the article recommends, I would like to start Certbot as an unprivileged user and create a user certbot. Since the Ansible module user also creates the home directory, I can remove the Create Certbot directory command right away and create the user before I generate the dhparams.pem file. Finally, I create the three requested directories:

- name: Create Certbot group
  group:
    name: certbot
    state: present

- name: Create Certbot user
  user:
    name: certbot
    group: certbot
    groups: haproxy
    shell: /bin/bash
    home: /opt/certbot
    state: present

- name: Generate 2048 bit dhparams.pem file
  openssl_dhparam:
    owner: root
    group: root
    mode: 0644
    path: /opt/certbot/dhparams.pem
    size: 2048

- name: Create Certbot log directory
  file:
    path: /opt/certbot/logs
    state: directory
    owner: certbot
    group: certbot
    mode: 0755

- name: Create Certbot config directory
  file:
    path: /opt/certbot/config
    state: directory
    owner: certbot
    group: certbot
    mode: 0755

- name: Create Certbot .config/letsencrypt directory
  file:
    path: /opt/certbot/.config/letsencrypt
    state: directory
    owner: certbot
    group: certbot
    mode: 0755
Code language: YAML (yaml)

Next, I have to create the configuration file /opt/certbot/.config/letsencrypt/cli.ini on the server. I save it locally in my role under roles/happy1_haproxy/files/cli.ini with the following content:

work-dir=/opt/certbot/
logs-dir=/opt/certbot/logs/
config-dir=/opt/certbot/config/
Code language: plaintext (plaintext)

To copy the file to the server, I add the following to the tasks definition file:

- name: Copy Certbot configuration
  copy:
    src: cli.ini
    dest: /opt/certbot/.config/letsencrypt/cli.ini
    owner: root
    group: root
    mode: 0644
Code language: YAML (yaml)

If a certificate is renewed, HAProxy must be restarted afterwards. Allowing the certbot user to do this is done with an entry in the /etc/sudoers file and can be done very easily with the Ansible module lineinfile, where I specify that an existing line starting with %certbot will be overwritten. So I add the following task:

- name: Allow the certbot user to restart HAProxy
  lineinfile:
    dest: /etc/sudoers
    state: present
    regexp: '^%certbot'
    line: '%certbot ALL=NOPASSWD: /bin/systemctl restart haproxy'
Code language: YAML (yaml)

The next step is to install the certbox-haproxy module. For this, I use the Ansible modules git, easy_install and pip and enter the following into the tasks definition file:

- name: Clone certbot-haproxy repository
  git:
    repo: https://code.greenhost.net/open/certbot-haproxy.git
    dest: /opt/certbot/certbot-haproxy

- name: Install python-setuptools
  apt:
    name: python-setuptools
    state: present

- name: Install pip
  easy_install:
    name: pip

- name: Install certbot-haproxy
  pip:
    name: /opt/certbot/certbot-haproxy/
Code language: YAML (yaml)

The first task clones the git repository of the certbot-haproxy module. The second task installs the python-setuptools, which contain the program easy_install, which in turn installs pip, a package management program for Python packages in the third task. The fourth task finally calls pip to compile and install the certbot-haproxy module.

Now you have to make changes to the haproxy.cfg.j2. First, the file dhparams.pem created by Ansible must be entered. For this purpose, a line is appended in the global section, so that the section ends as follows:

    ssl-default-bind-ciphers ECDH+AESGCM:DH+AESGCM:ECDH+AES256:DH+AES256:ECDH+AES128:DH+AES:RSA+AESGCM:RSA+AES:!aNULL:!MD5:!DSS
    ssl-default-bind-options no-sslv3
    ssl-dh-param-file /opt/certbot/dhparams.pem
Code language: plaintext (plaintext)

Also, the Let's Encrypt server's requests (whose path begins with /.well-known/acme-challenge) must be redirected to the certbot-haproxy module, which is bound to port 8000. Therefore, I extend the frontend section as follows:

frontend happycoders_80
    bind *:80

    # Forward Certbot verification requests to the certbot-haproxy plugin
    acl is_certbot path_beg -i /.well-known/acme-challenge
    use_backend certbot if is_certbot

    default_backend happycoders_wordpress
Code language: Nginx (nginx)

And at the end of haproxy.cfg.j2, I enter the backend certbot:

backend certbot
    log global
    mode http
    server certbot 127.0.0.1:8000
Code language: Nginx (nginx)

Since the password query is defined in the happycoders_wordpress backend, the Let's Encrypt server's challenge requests via the path /.well-known/acme-challenge are not affected. For those who have lost the overview: here is again the complete content of the file roles/happy1_haproxy/templates/haproxy.cfg.j2:

global
    log /dev/log    local0
    log /dev/log    local1 notice
    chroot /var/lib/haproxy
    stats socket /run/haproxy/admin.sock mode 660 level admin expose-fd listeners
    stats timeout 30s
    user haproxy
    group haproxy
    daemon

    # Default SSL material locations
    ca-base /etc/ssl/certs
    crt-base /etc/ssl/private

    # Default ciphers to use on SSL-enabled listening sockets.
    # For more information, see ciphers(1SSL). This list is from:
    #  https://hynek.me/articles/hardening-your-web-servers-ssl-ciphers/
    # An alternative list with additional directives can be obtained from
    #  https://mozilla.github.io/server-side-tls/ssl-config-generator/?server=haproxy
    ssl-default-bind-ciphers ECDH+AESGCM:DH+AESGCM:ECDH+AES256:DH+AES256:ECDH+AES128:DH+AES:RSA+AESGCM:RSA+AES:!aNULL:!MD5:!DSS
    ssl-default-bind-options no-sslv3
    ssl-dh-param-file /opt/certbot/dhparams.pem

defaults
    log     global
    mode    http
    option  httplog
    option  dontlognull
    timeout connect 5000
    timeout client  50000
    timeout server  50000
    errorfile 400 /etc/haproxy/errors/400.http
    errorfile 403 /etc/haproxy/errors/403.http
    errorfile 408 /etc/haproxy/errors/408.http
    errorfile 500 /etc/haproxy/errors/500.http
    errorfile 502 /etc/haproxy/errors/502.http
    errorfile 503 /etc/haproxy/errors/503.http
    errorfile 504 /etc/haproxy/errors/504.http

frontend happycoders_80
    bind *:80

    # Forward Certbot verification requests to the certbot-haproxy plugin
    acl is_certbot path_beg -i /.well-known/acme-challenge
    use_backend certbot if is_certbot

    default_backend happycoders_wordpress

backend happycoders_wordpress
    mode http
    server server1 localhost:8001 check
    http-response del-header x-powered-by
{% if happycoders_testing|default(false) %}
    acl authorized http_auth(happycoders_test_users)
    acl wp-cron path_beg -i /wp-cron.php
    http-request auth realm HappyCoders.eu unless authorized or wp-cron

userlist happycoders_test_users
    user sven password X/UgQ53BouggY
{% endif %}

backend certbot
    log global
    mode http
    server certbot 127.0.0.1:8000
Code language: Nginx (nginx)

This completes the Certbot installation tasks, and it's time to run the Ansible playbook:

ansible-playbook --ask-vault-pass happy1.yml
Code language: plaintext (plaintext)

Creating the Certbot script

Now that Certbot and the certbot-haproxy module are installed, it is time to run Certbot. I have several HappyCoders domains that I would like to redirect to happycoders.eu in a later step. Therefore, I need an SSL/TLS certificate for all my domains. Certbot can create certificates for all domains with a single call. First, I create the following array in the host configuration file host_vars/happy1.happycoders.eu:

happycoders_domains:
  - happycoders.eu
  - happycoders.at
  - happycoders.biz
  - happycoders.ch
  - happycoders.club
  - happycoders.de
  - happycoders.info
  - happycoders.it
  - happycoders.me
  - happycoders.net
  - happycoders.uk
Code language: YAML (yaml)

I deliberately wrote happycoders.eu at the beginning because the first domain in the certificate is registered as the main domain and all others as alternative domains. The DNS entries for all these domains must point to my server for the Let's Encrypt server's challenge requests to arrive. To call Certbot, I create the following script in the file roles/happy1_haproxy/templates/run-certbot.j2:

runuser -l certbot -c 'certbot run 
    --authenticator certbot-haproxy:haproxy-authenticator 
    --installer certbot-haproxy:haproxy-installer 
    --non-interactive 
    --domains {% for domain in happycoders_domains %}{% if loop.index > 1 %},{% endif %}www.{{ domain }},{{ domain }}{% endfor %} 
    --expand 
    --agree-tos 
    --email [email protected]'
Code language: plaintext (plaintext)

With runuser -l certbot -c '<command>', I call the command specified in apostrophes by the certbot user. The parameters of the certbot command mean in detail:

• --authenticator specifies the authenticator plugin – in this case, the previously installed certbot-haproxy module. The authenticator plugin checks whether the domains I want to create certificates for are under my control by making a file available over HTTP, which is then checked by Let's Encrypt.
• --installer defines the installer plugin – here, I also use the certbot-haproxy module.
• --non-interactive specifies that Certbot runs without user interaction.
• --domains lists the domains for which I want to create certificates. I list each domain first with and then without www.
• --expand specifies that the certificate may be extended if the list of domains is extended. If the list is extended without specifying this parameter, the command aborts with an error message.
• --agree-tos means that I agree to the terms of service.
• --email specifies the email address at which I will be informed about expiring certificates, for example.

To copy the script to the server, I create the task file roles/happy1_haproxy/tasks/setup_certbot_script.yml with the following content:

---
- name: Create Certbot scripts directory
  file:
    path: /opt/certbot/scripts/
    state: directory
    owner: root
    group: root
    mode: 0755

- name: Copy Certbot script
  template:
    src: run-certbot.j2
    dest: /opt/certbot/scripts/run-certbot
    owner: root
    group: root
    mode: 0755

- name: Link cronjob to renew Certbot certificates
  file:
    src: /opt/certbot/scripts/run-certbot
    dest: /etc/cron.daily/run-certbot
    owner: root
    group: root
    state: link
Code language: YAML (yaml)

With the already known Ansible modules, I create a directory, copy the script to it and link it to /etc/cron.daily/ so that it runs daily. The reason for this is that the script not only creates the certificates for the first time, but also automatically renews them as soon as they are about to expire. I now import the file setup_certbot_script.yml into the main task definition file roles/happy1_haproxy/tasks/main.yml, which now looks like this:

---
- import_tasks: install_haproxy.yml
- import_tasks: install_certbot.yml
- import_tasks: setup_certbot_script.yml
Code language: YAML (yaml)

I'm running the playbook to install the Certbot script on the server:

ansible-playbook --ask-vault-pass happy1.yml
Code language: plaintext (plaintext)

I check on the server whether the script was created correctly:

Running the Certbot Script

The script looks good, so I'll execute it right away. I have to do this as root user because only root knows the runuser command.

The certificates were generated successfully – but my site is not yet accessible via HTTPS because I haven't created a frontend section in the HAProxy configuration yet.

Before I do, I want to satisfy my curiosity and find out what happens when I run the script again. This time, I get a succinct, satisfying answer:

Cert not yet due for renewal
Keeping the existing certificate

Activating HTTPS in HAProxy

Now I will create the HTTPS frontend. Since this requires the certificate file created by Certbot, HAProxy would not run on a fresh server before Certbot was executed. Without HAProxy, Certbot could not be called – a hen-egg-problem. The solution is to let Ansible check if the certificate exists and create the HTTPS frontend only if this is the case. The check is done by the following task in the file roles/happy1_haproxy/tasks/install_haproxy.yml, which is inserted before the task Configure HAProxy:

- name: Check if the Certbot certificates exist
  stat:
    path: /opt/certbot/haproxy_fullchains/__fallback.pem
  register: certbot_cert_check_result
Code language: YAML (yaml)

I explained the Ansible module stat in the first article of the series. And the following is inserted into haproxy.cfg.j2, before backend happycoders_wordpress:

{% if certbot_cert_check_result.stat.exists %}
frontend happycoders_443
    bind *:443 ssl crt /opt/certbot/haproxy_fullchains/__fallback.pem crt /opt/certbot/haproxy_fullchains alpn h2,http/1.1

    # Important, so that WordPress doesn't send a redirect to HTTPS ("https" must be in lowercase letters!)
    http-request set-header X-Forwarded-Proto https

    default_backend happycoders_wordpress
{% else %}
    # SSL not yet activated because certificates do not exist yet.
    # To create the certificates, run:
    # sudo /opt/certbot/scripts/run-certbot
    # Then execute the playbook again.
{% endif %}
Code language: Nginx (nginx)

If the certificate file exists, the HTTPS frontend will be included, otherwise a comment will be entered, with which I will virtually tell myself that the Certbot script still has to be executed. Important is the line http-request set-header X-Forwarded-Proto https – this inserts a header into the HTTP request which informs WordPress that the page was called via HTTPS. Otherwise, WordPress would always reply with a redirect to HTTPS if the page URL configured in WordPress starts with https://.

I run the Ansible playbook again to upload the updated HAProxy configuration to the server. Since nothing significant has changed in the output, I will not take a screenshot here. After executing the playbook, the page is accessible via HTTPS:

Via the Chrome console, I verify that the HTTP/2 protocol is used:

Redirecting to https://www.happycoders.eu

Now I want to redirect HTTP to HTTPS – and all alternative HappyCoders domains to www.happycoders.eu. For the redirection from HTTP to HTTPS, one line of code is enough – I enter the following in haproxy.cfg.j2 in the frontend happycoders_80 directly before default_backend:

    # Redirect HTTP to HTTPS for all other cases
    redirect scheme https code 301 if !is_certbot
Code language: Nginx (nginx)

With if !is_certbot, I make sure that the Let's Encrypt server's challenge requests can still be made via HTTP. The redirection to www.happycoders.eu can also be done via one line of code:

    # Redirect to https://www.happycoders.eu if another host name was specified
    redirect prefix https://www.happycoders.eu code 301 if !{ hdr(host) -i www.happycoders.eu }
Code language: Nginx (nginx)

However, it becomes more complicated if you also want to return the HTTP Strict Transport Security header to the client. Why this doesn't work if you enter the corresponding directive in the frontend and what to do instead, is explained in the HAProxy blog article HAProxy and HTTP Strict Transport Security (HSTS) Header in HTTP Redirects. Summarized: the frontend only inserts response headers if the response comes from a backend – therefore a dummy backend is required.

I enter the following in the frontend happycoders_443 – also directly before default_backend:

    http-response set-header Strict-Transport-Security "max-age=31536000; includeSubDomains; preload"

    # Redirect to https://www.happycoders.eu if another host name was specified.
    # Need to do this via a dummy backend + frontend as otherwise the HSTS response header would not be sent.
    # (see https://www.haproxy.com/de/blog/haproxy-and-http-strict-transport-security-hsts-header-in-http-redirects/)
    use_backend redirect_to_www_happycoders_eu if !{ hdr(host) -i www.happycoders.eu }
Code language: Nginx (nginx)

With the first directive, the HSTS header is inserted, so that the browser will redirect to the HTTPS page via "internal redirect" in the future if http:// is entered – without even calling the HTTP page first. The use_backend directive redirects all requests that do not go to the host www.happycoders.eu to the dummy backend redirect_to_www_happycoders_eu. I insert this together with the dummy frontend as follows – before the backend happycoders_wordpress:

# This dummy backend is needed so that the HSTS response header is also being sent with a redirect (see comment above).
backend redirect_to_www_happycoders_eu
    server dummy_redirect_server 127.0.0.1:8002

# This dummy frontend is needed so that the HSTS response header is also being sent with a redirect (see comment above).
frontend redirect_to_www_happycoders_eu
    bind 127.0.0.1:8002
    redirect prefix https://www.happycoders.eu code 301
Code language: Nginx (nginx)

The dummy backend thus redirects to the dummy frontend, which finally returns a "permanent redirect" to the host www.happycoders.eu. The resulting overhead can be safely ignored, as this redirect takes place only once a year per user.

The file roles/happy1_haproxy/templates/haproxy.cfg.j2 finally looks like this:

global
    log /dev/log    local0
    log /dev/log    local1 notice
    chroot /var/lib/haproxy
    stats socket /run/haproxy/admin.sock mode 660 level admin expose-fd listeners
    stats timeout 30s
    user haproxy
    group haproxy
    daemon

    # Default SSL material locations
    ca-base /etc/ssl/certs
    crt-base /etc/ssl/private

    # Default ciphers to use on SSL-enabled listening sockets.
    # For more information, see ciphers(1SSL). This list is from:
    #  https://hynek.me/articles/hardening-your-web-servers-ssl-ciphers/
    # An alternative list with additional directives can be obtained from
    #  https://mozilla.github.io/server-side-tls/ssl-config-generator/?server=haproxy
    ssl-default-bind-ciphers ECDH+AESGCM:DH+AESGCM:ECDH+AES256:DH+AES256:ECDH+AES128:DH+AES:RSA+AESGCM:RSA+AES:!aNULL:!MD5:!DSS
    ssl-default-bind-options no-sslv3
    ssl-dh-param-file /opt/certbot/dhparams.pem

defaults
    log     global
    mode    http
    option  httplog
    option  dontlognull
    timeout connect 5000
    timeout client  50000
    timeout server  50000
    errorfile 400 /etc/haproxy/errors/400.http
    errorfile 403 /etc/haproxy/errors/403.http
    errorfile 408 /etc/haproxy/errors/408.http
    errorfile 500 /etc/haproxy/errors/500.http
    errorfile 502 /etc/haproxy/errors/502.http
    errorfile 503 /etc/haproxy/errors/503.http
    errorfile 504 /etc/haproxy/errors/504.http

frontend happycoders_80
    bind *:80

    # Forward Certbot verification requests to the certbot-haproxy plugin
    acl is_certbot path_beg -i /.well-known/acme-challenge
    use_backend certbot if is_certbot

    # Redirect HTTP to HTTPS for all other cases
    redirect scheme https code 301 if !is_certbot

    default_backend happycoders_wordpress

{% if certbot_cert_check_result.stat.exists %}
frontend happycoders_443
    bind *:443 ssl crt /opt/certbot/haproxy_fullchains/__fallback.pem crt /opt/certbot/haproxy_fullchains alpn h2,http/1.1

    # Important, so that WordPress doesn't send a redirect to HTTPS ("https" must be in lowercase letters!)
    http-request set-header X-Forwarded-Proto https

    http-response set-header Strict-Transport-Security "max-age=31536000; includeSubDomains; preload"

    # Redirect to https://www.happycoders.eu if another host name was specified.
    # Need to do this via a dummy backend + frontend as otherwise the HSTS response header would not be sent.
    # (see https://www.haproxy.com/de/blog/haproxy-and-http-strict-transport-security-hsts-header-in-http-redirects/)
    use_backend redirect_to_www_happycoders_eu if !{ hdr(host) -i www.happycoders.eu }

    default_backend happycoders_wordpress
{% else %}
    # SSL not yet activated because certificates do not exist yet.
    # To create the certificates, run:
    # sudo /opt/certbot/scripts/run-certbot
    # Then execute the playbook again.
{% endif %}

# This dummy backend is needed so that the HSTS response header is also being sent with a redirect (see comment above).
backend redirect_to_www_happycoders_eu
    server dummy_redirect_server 127.0.0.1:8002

# This dummy frontend is needed so that the HSTS response header is also being sent with a redirect (see comment above).
frontend redirect_to_www_happycoders_eu
    bind 127.0.0.1:8002
    redirect prefix https://www.happycoders.eu code 301

backend happycoders_wordpress
    mode http
    server server1 localhost:8001 check
    http-response del-header x-powered-by
{% if happycoders_testing|default(false) %}
    acl authorized http_auth(happycoders_test_users)
    acl wp-cron path_beg -i /wp-cron.php
    http-request auth realm HappyCoders.eu unless authorized or wp-cron

userlist happycoders_test_users
    user sven password X/UgQ53BouggY
{% endif %}

backend certbot
    log global
    mode http
    server certbot 127.0.0.1:8000
Code language: Nginx (nginx)

I run the Ansible playbook one last time to update the HAProxy configuration on the server. I will not take a screenshot this time.

For testing purposes, I call http://happycoders.de, and I am happy to see that this call is finally redirected to https://www.happycoders.eu, i.e. that both redirections to HTTPS and www.happycoders.eu work as intended. Finally, I check whether the HSTS header is set and correctly evaluated by the browser.

curl -I https://happycoders.de
Code language: plaintext (plaintext)

You can see very well that the call to http://www.happycoders.eu was redirected by the browser via an "internal redirect" (status 307) to https://www.happycoders.eu and then by the server via a "temporary redirect" (status 302) to the install.php page.

Summary and outlook

In this last part of the series, I showed you how to make my WordPress blog accessible with HAProxy via HTTP and HTTPS, and how to request and install free SSL certificates with Let's Encrypt/Certbot. This is the end of this series of articles about setting up my server.

I hope you enjoyed this series of articles and the corresponding tutorial videos. I'd be very happy if you left comments for me. You can also write me any time you're stuck in the middle of something. Not everything went as smoothly as you might think when reading this tutorial. I also had to try a lot, repeat steps, jump back to previous sections and even start from the beginning a few times with the provisioning of the server. Thanks to Ansible, that wasn't difficult at all.

Thank you so much for staying with me and happy coding!

If you liked the article series, I would also be very happy if you share this article via one of the following share buttons.