Run Tailscale on a Heroku dyno, shoving connections through proxychains.
It really shouldn't be. However, I discovered that in my project to use Django and PostgreSQL,
pyscopg did not respect the ALL_PROXY
environment variable. In order to get the network
communication to go through the Tailscale SOCKS5 proxy, I needed to manually force it. Enter
proxychains-ng.
This is based on https://tailscale.com/kb/1107/heroku/.
Thank you to @rdotts, @kongmadai, @mvisonneau for their work on tailscale-docker and tailscale-heroku.
To set up your Heroku application, add the buildpack and TAILSCALE_AUTH_KEY
environment variable:
$ heroku buildpacks:add https://github.com/aspiredu/heroku-tailscale-buildpack
Buildpack added. Next release on test-app will use aspiredu/heroku-tailscale-buildpack.
Run `git push heroku main` to create a new release using this buildpack.
$ heroku config:set TAILSCALE_AUTH_KEY="..."
$ git push heroku main
...
To have your processes connect through the Tailscale proxy, you need to update your
Procfile
. Here's an example for a Django project with a Celery worker:
web: proxychains4 -f vendor/proxychains-ng/conf/proxychains.conf uvicorn --host 0.0.0.0 --port "$PORT" myproject.project.asgi:application
worker: proxychains4 -f vendor/proxychains-ng/conf/proxychains.conf celery -A myproject.project worker
To test a connection, you can add the hello.ts.net
machine into your network.
Follow the instructions here. You
may need to modify your ACLs to allow access to the test machine. For example, I have
a separate Tailscale token that is tagged with tag:test
. My ACL looks like:
{
"hosts": {
"hello-test": "100.101.102.103"
},
// Access control lists.
"acls": [
// Only allow the test tag to access anything.
{"action": "accept", "src": ["tag:test"], "dst": ["hello-test:*"]}
]
}
To verify the connection works run:
heroku run -- heroku-tailscale-test.sh
You should see curl respond with <a href="https://hello.ts.net">Found</a>.
The following settings are available for configuration via environment variables:
TAILSCALE_ACCEPT_DNS
- Accept DNS configuration from the admin console. Defaults to accepting DNS settings.TAILSCALE_ACCEPT_ROUTES
- Accept subnet routes that other nodes advertise. Linux devices default to not accepting routes. Defaults to accepting.TAILSCALE_ADVERTISE_EXIT_NODES
- Offer to be an exit node for outbound internet traffic from the Tailscale network. Defaults to not advertising.TAILSCALE_ADVERTISE_TAGS
- Give tagged permissions to this device. You must be listed in "TagOwners" to be able to apply tags. Defaults to none.TAILSCALE_AUTH_KEY
- Provide an auth key to automatically authenticate the node as your user account. This must be set.TAILSCALE_HOSTNAME
- Provide a hostname to use for the device instead of the one provided by the OS. Note that this will change the machine name used in MagicDNS. Defaults to the hostname of the application (a guid). If you have Heroku Labs runtime-dyno-metadata enabled, it defaults to[commit]-[dyno]-[appname]
.TAILSCALE_SHIELDS_UP"
- Block incoming connections from other devices on your Tailscale network. Useful for personal devices that only make outgoing connections. Defaults to off.TAILSCALED_VERBOSE
- Controls verbosity for the tailscaled command. Defaults to 0.
The following settings are for the compile process for the buildpack. If you change these, you must trigger a new build to see the change. Simply changing the environment variables in Heroku will not cause a rebuild. These are all optional and will default to the latest values.
TAILSCALE_BUILD_TS_VERSION
- The target version Tailscale package.TAILSCALE_BUILD_TS_TARGETARCH
- The target architecture for the Tailscale package.TAILSCALE_BUILD_EXCLUDE_START_SCRIPT_FROM_PROFILE_D
- Excludes the start script from the buildpack's.profile.d/
folder. If you set this to true, you must callvendor/tailscale/heroku-tailscale-start.sh
. This likely should go into your.profile
script (see Heroku docs). Starting the script in your.profile
file would allow you to better control environment variables in respect to the executables. For example, a specific dyno could changeTAILSCALE_HOSTNAME
before tailscale starts.TAILSCALE_BUILD_PROXYCHAINS_REPO
- The repository to install the proxychains-ng library from.
If you decide you want to customize the proxychains.conf
configuration file, you can copy the
file from conf into your project. If you copy it to the base directory of your application,
ProxyChains will find it automatically. If you copy it to a specific directory, such as conf,
you'll need to specify the path.
For example, if your conf file exists at <project>/conf/proxychains.conf
your web command
would need to be:
proxychains4 -f conf/proxychains.conf <process>
Switching to a Tailscale database or service can be troublesome. Especially if you interact
with the resource during the Release phase of Heroku's deployments
such as basic SQL migrations. This is because you don't want to use the proxychains4
wrapper if you don't have Tailscale running, and you can't have Tailscale running if you
don't have a valid Tailscale auth key and the database/resource configured in your tailnet.
I suggest working these problems out in reverse allowing for a fallback to a connection outside of your tailnet. Once you've done the final switch over, you can remove access to your database/resource from outside of the tailnet.
- Configure database/resource to be accessible in and outside of your tailnet
- Create a Tailscale auth key (reusuable1, not ephemeral, and appropriately tagged)
- Add the auth key and the Tailscale database/resource url to your Heroku app's environment variables.
heroku config:set TAILSCALE_AUTH_KEY=<tailscale_auth_key> \
TAILSCALE_DATABASE_URL=<tailscale_database_url>
- Add the heroku-tailscale-buildpack
heroku buildpacks:add https://github.com/aspiredu/heroku-tailscale-buildpack
-
(Optional) Test your integration.
- Add the Tailscale test machine in your tailnet
- Create a test tag that can only access the hello.ts.net machine via your ACLs
- Create a reusable ephemeral auth token that has the test tag applied to it.
- Temporarily change your application to use the test auth key.
- Trigger a build. This should include this buildpack.
- Run a one-off dyno to confirm that the setup is correct.
heroku run heroku-tailscale-start.sh
- Restore the previous version of Tailscale auth key.
-
Modify your application to try to use the Tailscale database/resource and fallback to the non-tailnet version. If you're using python, the following script may help:
import os
import dj_database_url
def tailscale_resource_key(base_key):
"""Fetch the resource key for a Tailscale service.
It checks for an environment variable with the TAILSCALE_ prefix
and if it exists and TAILSCALE_AUTH_KEY is defined, it uses that key.
Else it returns the value that was passed in.
This is useful for configuring different services to use tailscale
without having to do everything all at once.
"""
tailscale_auth_key = "TAILSCALE_AUTH_KEY"
tailscale_resource_key = f"TAILSCALE_{base_key}"
return (
tailscale_resource_key
if os.environ.get(tailscale_resource_key) and os.environ.get(tailscale_auth_key)
else base_key
)
DATABASES = {
"default": dj_database_url.config(env=tailscale_resource_key("DATABASE_URL"))
}
- Push your code to your Heroku application, triggering a new build.
git push heroku
- You're now running your application connecting to your resources via Tailscale.
It is possible to serve your web application on Heroku to only users in your tailnet, but that is outside the scope of this particular buildpack.
That said, this buildpack makes it possible to do so. We solved this by
creating a separate dyno in the Procfile
called admin_web
which runs a script
that does the following:
- Attempt to restore certificate files from cache.
- Revoke previous machine with the desired hostname.
- Update current tailscale process to use the desired hostname.
- Validate and potentially re-issue certificates for hostname.
- Store certificate files in cache.
- Serve localhost process on Tailnet for the hostname.
- Start web process to serve requests on localhost.
The certificates need to be persisted beyond instances of the dyno to avoid requesting certificates for the same full DNS path from Let's Encrypt. Let's Encrypt will limit you to 5 requests per week.
The certificate files you'll need to persist are:
.local/share/tailscale/certs/$TAILSCALE_HOSTNAME.crt
.local/share/tailscale/certs/$TAILSCALE_HOSTNAME.key
To serve your application your admin_web
script should do something similar to:
# Expose the local 8000 port to the tailnet.
# The URL will be https://<machine-name>.<tailnet>.ts.net
tailscale serve https / http://127.0.0.1:8000
# Generate the current serve status so the logs include the URL.
tailscale serve status
# Start the web server process
exec proxychains4 -f vendor/proxychains-ng/conf/proxychains.conf uvicorn --host 127.0.0.1 --port "8000" project.asgi:application
Footnotes
-
You want reusable auth keys here because it will be used across all of your dynos in the application. ↩