How to use multiple domains in a load-balanced environment

[nichols-green]

When used in a load balanced situation with two nodes and a load balancer, the public_url env var breaks the load balance config.

server 1 is addressed as mysite-1.mydomain.org
server 2 is addressed as mysite-2.mydomain.org
The LB is addressed as mysite.mydomain.org.

In the Directus config, I've set the public_url to mysite.mydomain.org. This works for the customer facing site and the email invitations. This setup fails when attempting to reach the -1, -2 sites directly. This is essential as the load balancer health check attempts to reach the downstream servers by their -1, -2 addresses.

Essentially we can run the site without health checks and everything works, or we can enable health checks as is our policy and the site breaks. The load balancer isn't able to get a response from the health check endpoints, and therefore has no active downstream servers to route to.

Lastly, I set the public_url variable to / and everything worked as expected. The load balancer health checks worked, and the site was reachable at all three addresses. I thought this was the solution, until I tried to invite a member of my team to the site and the links in the email were all just relative links prefixed with /

I would suggest separating thi public_url variable into public_url and host_url, or something along those lines to allow the use of / as the site host and still have mysite.mydomain.org in the email prefix and other areas where that would be used.

To Reproduce

Deploy a site to two or more servers, with a loadbalancer. Assign the servers addresses, and assign the loadbalancer an address. Set the public_url env var to the value of the load balancer address. Ensure health checks are off. The site should work as expected, but the two direct addresses will not work.

Set the public_url value to /, and the site should be working as expected. Invite someone to the site, and check the email links. The links will not be complete.

[rijkvanzanten]

In the Directus config, I've set the public_url to mysite.mydomain.org. This works for the customer facing site and the email invitations. This setup fails when attempting to reach the -1, -2 sites directly. This is essential as the load balancer health check attempts to reach the downstream servers by their -1, -2 addresses.

This shouldn't necessarily matter in Directus itself. That public URL is only used for things like password-reset emails, and other user-facing names. That being said, it doesn't affect the /server/health endpoint at all (Directus itself runs fine regardless of what value you use for PUBLIC_URL). I'm a little confused as to why that would affect the health check at all. How are you health checking currently?

[nichols-green]

Are we sure this isn't caused by some other network configuration or environment change besides the public url flag?

I was asking this question myself and so I tested switching the value of the public_url variable from the site address to / and back. The custom endpoint returns service unavailable when the public_url is set to anything other than / or the address I am attempting to reach. I get the same result when attempting to reach the /server/health endpoint.

This is currently an open feature request. If enough people upvote it we'll look into implementing it if and when possible. As of now, the only way to get an accurate timeline or release date is to sponsor this task. We recognize that this may be an important feature/fix, but we are a small open-source organization with a lot to triage and complete. If you can't sponsor this ticket, then the next best thing is to increase its priority by giving it a 👍 . Or contribute a pull-request... after all, this is OSS!

Happy to contribute! This one feels a little beyond me, but I'll give it a shot. After all, the last one felt a little beyond me as well and that worked out beautifully.

[nichols-green]

Also, your example controller doesn't make use of ServerService, and I suspect it is within this service that there is a dependency on public_url, either directly, or while loading other deps.

[rijkvanzanten]

Happy to contribute! This one feels a little beyond me, but I'll give it a shot. After all, the last one felt a little beyond me as well and that worked out beautifully.

I think it could be doable by adding a new "HOST_URL" env var here

directus/api/src/env.ts

Line 19 in e95404a

PUBLIC_URL: '/',

that defaults to "PUBLIC_URL" (so we don't require the user to set both if they don't need it), and then effectively find and replace all uses of PUBLIC_URL with the new HOST_URL (except for the one that creates the links in the emails) 🤔

[nichols-green]

OK I was wrong. The custom endpoints/health endpoint failure was due to a misconfigured uploads directory. So this failure is only related to the loading of dependencies. Should be a pretty quick fix.

[rijkvanzanten]

Ahh yeah that explains!

[nichols-green]

This was solved somewhere between RC.76 and RC.96 and is no longer a concern.