12 KiB
Using your own webserver, instead of this playbook's Traefik reverse-proxy (optional, advanced)
Note: the playbook is in the process of moving to Traefik. The documentation below may be incomplete or misleading.
By default, this playbook installs its own nginx webserver (called matrix-nginx-proxy
, in a Docker container) which listens on ports 80 and 443.
If that's alright, you can skip this.
Soon, this default will change and the playbook will install its own Traefik reverse-proxy instead.
Traefik
Traefik will be the default reverse-proxy for the playbook in the near future.
There are 2 ways to use Traefik with this playbook, as described below.
Traefik managed by the playbook
To switch to Traefik now, use configuration like this:
matrix_playbook_reverse_proxy_type: playbook-managed-traefik
devture_traefik_config_certificatesResolvers_acme_email: YOUR_EMAIL_ADDRESS
This will install Traefik in the place of matrix-nginx-proxy
. Traefik will manage SSL certificates for all services seamlessly.
Note: during the transition period, matrix-nginx-proxy
will still be installed in local-only mode. Do not be alarmed to see matrix-nginx-proxy
running even when you've chosen Traefik as your reverse-proxy. In the future, we'll be able to run without nginx, but we're not there yet.
Traefik managed by you
matrix_playbook_reverse_proxy_type: other-traefik-container
matrix_playbook_reverse_proxyable_services_additional_network: your-traefik-network
devture_traefik_certs_dumper_ssl_dir_path: "/path/to/your/traefiks/acme.json/directory"
In this mode all roles will still have Traefik labels attached. You will, however, need to configure your Traefik instance and its entrypoints.
By default, the playbook congiures services use a web-secure
(443) and matrix-federation
(8448) entrypoints, as well as a default
certificate resolver.
You need to configure 3 entrypoints for your Traefik server: web
(TCP port 80
), web-secure
(TCP port 443
) and matrix-federation
(TCP port 8448
).
Below is some configuration for running Traefik yourself, although we recommend using Traefik managed by the playbook.
Note that this configuration on its own does not redirect traffic on port 80 (plain HTTP) to port 443 for HTTPS, which may cause some issues, since the built-in Nginx proxy usually does this. If you are not already doing this in Traefik, it can be added to Traefik in a file provider as follows:
[http]
[http.routers]
[http.routers.redirect-http]
entrypoints = ["web"] # The 'web' entrypoint must bind to port 80
rule = "HostRegexp(`{host:.+}`)" # Change if you don't want to redirect all hosts to HTTPS
service = "dummy" # Unused, but all routers need services (for now)
middlewares = ["https"]
[http.services]
[http.services.dummy.loadbalancer]
[[http.services.dummy.loadbalancer.servers]]
url = "localhost"
[http.middlewares]
[http.middlewares.https.redirectscheme]
scheme = "https"
permanent = true
You can use the following docker-compose.yml
as example to launch Traefik.
version: "3.3"
services:
traefik:
image: "docker.io/traefik:v2.9.6"
restart: always
container_name: "traefik"
networks:
- traefik
command:
- "--api.insecure=true"
- "--providers.docker=true"
- "--providers.docker.network=traefik"
- "--providers.docker.exposedbydefault=false"
- "--entrypoints.web-secure.address=:443"
- "--entrypoints.federation.address=:8448"
- "--certificatesresolvers.default.acme.tlschallenge=true"
- "--certificatesresolvers.default.acme.email=YOUR EMAIL"
- "--certificatesresolvers.default.acme.storage=/letsencrypt/acme.json"
ports:
- "443:443"
- "8448:8448"
volumes:
- "./letsencrypt:/letsencrypt"
- "/var/run/docker.sock:/var/run/docker.sock:ro"
networks:
traefik:
external: true
Another webserver
If you don't wish to use Traefik or matrix-nginx-proxy
, you can also use your own webserver.
Doing this is possible, but requires manual work.
There are 2 ways to go about it:
-
(recommended) Fronting the integrated reverse-proxy webserver with another reverse-proxy - using a playbook-managed reverse-proxy (either
matrix-nginx-proxy
or Traefik), disabling SSL termination for it, exposing this reverse-proxy on a few local ports (e.g.127.0.0.1:81
, etc.) and forwarding traffic from your own webserver to those few ports -
(difficult) Using no reverse-proxy on the Matrix side at all disabling all playbook-managed reverse-proxies (no
matrix-nginx-proxy
, no Traefik)
Fronting the integrated reverse-proxy webserver with another reverse-proxy
This method is about leaving the integrated reverse-proxy webserver be, but making it not get in the way (using up important ports, trying to retrieve SSL certificates, etc.).
If you wish to use another webserver, the integrated reverse-proxy webserver usually gets in the way because it attempts to fetch SSL certificates and binds to ports 80, 443 and 8448 (if Matrix Federation is enabled).
You can disable such behavior and make the integrated reverse-proxy webserver only serve traffic locally (or over a local network).
This is the recommended way for using another reverse-proxy, because the integrated one would act as a black box and wire all Matrix services correctly. You would only need to reverse-proxy a few individual domains and ports over to it.
For Traefik fronted by another reverse-proxy, you would need some configuration like this:
matrix_playbook_reverse_proxy_type: playbook-managed-traefik
# Ensure that public urls use https
matrix_playbook_ssl_enabled: true
# Disable the web-secure (port 443) endpoint, which also disables SSL certificate retrieval
devture_traefik_config_entrypoint_web_secure_enabled: false
devture_traefik_container_web_host_bind_port: '127.0.0.1:81'
devture_traefik_additional_entrypoints_auto:
- name: matrix-federation
port: 8449
host_bind_port: '127.0.0.1:8449'
config: {}
(Deprecated) For matrix-nginx-proxy
fronted by another reverse-proxy, you would need some configuration like this:
matrix_playbook_reverse_proxy_type: playbook-managed-nginx
# Ensure that public urls use https
matrix_playbook_ssl_enabled: true
# Disable SSL certificate retrieval
matrix_ssl_retrieval_method: none
# Given that we won't be obtaining SSL certificates here, disable HTTPS
matrix_nginx_proxy_https_enabled: false
# Do not listen for HTTP on port 80 globally (default), listen on the loopback interface.
# If you'd like, you can make it use the local network as well and reverse-proxy from another local machine.
matrix_nginx_proxy_container_http_host_bind_port: '127.0.0.1:81'
# Likewise, expose the Matrix Federation port on the loopback interface.
# Since `matrix_nginx_proxy_https_enabled` is set to `false`, this federation port will serve HTTP traffic.
# If you'd like, you can make it use the local network as well and reverse-proxy from another local machine.
#
# You'd most likely need to expose it publicly on port 8448 (8449 was chosen for the local port to prevent overlap).
matrix_nginx_proxy_container_federation_host_bind_port: '127.0.0.1:8449'
If you'll be fronting with a reverse-proxy that lives on another machine (not on the same one as Matrix), you need to replace 127.0.0.1
in the above configurations with 0.0.0.0
or another network interface.
Using no reverse-proxy on the Matrix side at all
Instead of Fronting the integrated reverse-proxy webserver with another reverse-proxy, you can also go another way -- completely disabling the playbook-managed reverse-proxy. You would then need to reverse-proxy from your own webserver directly to Matrix services.
This is more difficult, as you would need to handle the configuration for each service manually. Enabling additional services would come with extra manual work you need to do.
If your webserver is on the same machine, sure your web server user (something like http
, apache
, www-data
, nginx
) is part of the matrix
group. You should run something like this: usermod -a -G matrix nginx
. This allows your webserver user to access files owned by the matrix
group. When using an external nginx webserver, this allows it to read configuration files from /matrix/nginx-proxy/conf.d
. When using another server, it would make other files, such as /matrix/static-files/.well-known
, accessible to it.
Using your own nginx reverse-proxy running on the same machine
WARNING: this type of setup is not maintained and will be removed in the future. We recommend that you go for Fronting the integrated reverse-proxy webserver with another reverse-proxy instead.
If you'll be using nginx
running on the same machine (not in a container), you can make the playbook help you generate configuration for nginx
with this configuration:
matrix_playbook_reverse_proxy_type: other-nginx-non-container
# If you want https configured in /matrix/nginx-proxy/conf.d/
matrix_nginx_proxy_https_enabled: true
# If you will manage SSL certificates yourself, uncomment the line below
# matrix_ssl_retrieval_method: none
# If you're using an old nginx version, consider using a custom protocol list
# (removing `TLSv1.3` that is enabled by default) to suit your nginx version.
# matrix_nginx_proxy_ssl_protocols: "TLSv1.2"
You can most likely directly use the config files installed by this playbook at: /matrix/nginx-proxy/conf.d
. Just include them in your own nginx.conf
like this: include /matrix/nginx-proxy/conf.d/*.conf;
Using your own reverse-proxy running on the same machine or elsewhere
WARNING: this is difficult to set up, likely not very well supported and will be removed in the future. We recommend that you go for Fronting the integrated reverse-proxy webserver with another reverse-proxy instead.
To reverse-proxy manually for each service, use configuration like this:
# If your reverse-proxy runs on the same machine:
matrix_playbook_reverse_proxy_type: other-on-same-host
# Or, if it runs on another machine:
# matrix_playbook_reverse_proxy_type: other-on-another-host
# Or, optionally customize the network interface prefix (note the trailing `:` character).
# For other-on-same-host, the interface defaults to `127.0.0.1:`.
# For other-on-another-host, the interface defaults to `0.0.0.0:`.
# matrix_playbook_service_host_bind_interface_prefix: '192.168.30.4:'
With this configuration, each service will be exposed on a custom port. Example:
- Synapse will be exposed on port
8008
- Grafana will be exposed on port
3000
- synapse-admin will be exposed on port
8766
You can capture traffic for these services and forward it to their port. Some of these services are configured with certain default expecations with regard to hostname, path, etc., so it's not completely arbitrary where you can host them (unless you change the defaults).
For each new playbook service that you enable, you'll need special handling.
The examples/
directory contains examples for various servers: Caddy, Apache, HAproxy, Nginx, etc.