No, you’re right. There’s a much bigger story here. I was just trimming out a lot of it since I don’t normally run into people who can follow along easily. Akamai, for example, uses the proprietary AKAMAICDN record to allow the functionality of a CNAME. For example: foo.com AKAMAICDN’s to foo.com.edgekey.net (edgekey.net of course being the Akamai edge server suite). So someone using Akamai can do that to allow them to use the apex (but will still very likely have a www.foo.com CNAME foo.com setup to catch people who did a www anyway) Cloudflare uses CNAME flattening to “cheat” the CNAME rules by doing the CNAME DNS lookup internally and pretending to be authoritative for the request.
You don’t typical have your webpage itself delivered by CDN, you have your static assets delivered by CDN. Why can’t you put your static assets in a subdomain that gets a CNAME?
You can most certainly put static assets in a specific subdomain (and in fact, that’s how most setups are), but the CDN itself often requires handling the entire request at the beginning. You don’t want, for example, an A record at the apex pointing directly at your origin servers (terrible idea for security & performance; kind of defeats the purpose of the CDN), instead you want the user to connect to an edge server and have that edge server immediately serve the static content while the origin is contacted by the edge server for any non-static content that the user needs. This allows the CDN to do their cloud magic while your origin servers can do as little work as possible with as few people as possible. Effectively, you can block all requests to your servers that are not from your CDN. Many CDNs these days are also a major security feature.
So this is right at the limits of my understanding - I’m not a web admin, and I’m not a DNS dev; I’m a webapp backend dev. I’m coming at this from the perspective of best-practices for API and service design.
I’m generally talking about webpages that are services, both because that’s what I work on, and because that’s when trust actually matters to me. In these cases, a lot of page content is dynamic and cannot be hosted by a CDN.
For domains that are hosting services, it makes sense to have a subdomain for each service or class or service, but the only client that makes the URL visible is the browser so I want my landing page to not require typing in a subdomain, to reduce friction.
Subdomains are then defined by their purpose to the user, not by the type of resource they’re hosting. The typo of resource is determined by the accept header. If a client wants to access a resource using HTML or by xml or by json or by txt, they specify that in the header, and the webserver returns the resource in the requested representation.
Using a subdomain for a specific representation rather than for a specific utility seems lazy, which sets off red flags, if I’m required to enter it.
Edit, just in case it’s useful to explain what I mean by subdomain “purpose”.
To use a fictional version of Google where they didn’t have separate branding for all their services.
Search: search.google.com
Video: video.google.com
Music: music.video.com
Gmail: mail.google.com
Productivity suite: office.google.com
Etc.
These are what I feel subdomains are best used for. Clients can use headers to control how the resources for each of those services get represented. You don’t have representation-specific subdomains. If you only offer one service, then you don’t need subdomains; you can still have them but don’t force your users to use it, and probably still best to name it after the purpose, in case the company expands into other product domains.
It’s like the difference between “package by feature” and “package by layer”. Having a representation-specific subdomain is packaging by layer.
No, you’re right. There’s a much bigger story here. I was just trimming out a lot of it since I don’t normally run into people who can follow along easily. Akamai, for example, uses the proprietary AKAMAICDN record to allow the functionality of a CNAME. For example: foo.com AKAMAICDN’s to foo.com.edgekey.net (edgekey.net of course being the Akamai edge server suite). So someone using Akamai can do that to allow them to use the apex (but will still very likely have a www.foo.com CNAME foo.com setup to catch people who did a www anyway) Cloudflare uses CNAME flattening to “cheat” the CNAME rules by doing the CNAME DNS lookup internally and pretending to be authoritative for the request.
You can most certainly put static assets in a specific subdomain (and in fact, that’s how most setups are), but the CDN itself often requires handling the entire request at the beginning. You don’t want, for example, an A record at the apex pointing directly at your origin servers (terrible idea for security & performance; kind of defeats the purpose of the CDN), instead you want the user to connect to an edge server and have that edge server immediately serve the static content while the origin is contacted by the edge server for any non-static content that the user needs. This allows the CDN to do their cloud magic while your origin servers can do as little work as possible with as few people as possible. Effectively, you can block all requests to your servers that are not from your CDN. Many CDNs these days are also a major security feature.
So this is right at the limits of my understanding - I’m not a web admin, and I’m not a DNS dev; I’m a webapp backend dev. I’m coming at this from the perspective of best-practices for API and service design.
I’m generally talking about webpages that are services, both because that’s what I work on, and because that’s when trust actually matters to me. In these cases, a lot of page content is dynamic and cannot be hosted by a CDN.
For domains that are hosting services, it makes sense to have a subdomain for each service or class or service, but the only client that makes the URL visible is the browser so I want my landing page to not require typing in a subdomain, to reduce friction.
Subdomains are then defined by their purpose to the user, not by the type of resource they’re hosting. The typo of resource is determined by the
acceptheader. If a client wants to access a resource using HTML or by xml or by json or by txt, they specify that in the header, and the webserver returns the resource in the requested representation.Using a subdomain for a specific representation rather than for a specific utility seems lazy, which sets off red flags, if I’m required to enter it.
Edit, just in case it’s useful to explain what I mean by subdomain “purpose”.
To use a fictional version of Google where they didn’t have separate branding for all their services.
Search: search.google.com
Video: video.google.com
Music: music.video.com
Gmail: mail.google.com
Productivity suite: office.google.com
Etc.
These are what I feel subdomains are best used for. Clients can use headers to control how the resources for each of those services get represented. You don’t have representation-specific subdomains. If you only offer one service, then you don’t need subdomains; you can still have them but don’t force your users to use it, and probably still best to name it after the purpose, in case the company expands into other product domains.
It’s like the difference between “package by feature” and “package by layer”. Having a representation-specific subdomain is packaging by layer.