AWS makes this impossible in a few places such as a fair number of ACM use-cases.
I think your cert-per-session idea is interesting. We’d need significant throughput and processing boosts to make that happen, probably at least on the order of 10X computing speeds and 10X transmission speeds across the board minimum. These operations are computationally intense and add data to the wire so, for example, a simple Lemmy server with hundreds of users slows to a crawl and a larger site eg Mastodon goes to dialup speeds or worse. You can test at home by trying to generate an x509 self-signed cert before connecting to a website every time.
If you think it’s just too easy but people are still discussing it, please entertain the notion that you may have oversimplified the situation in your assessment and that as assumptions become clarified you may yet soon understand a horror that apple can’t quite grok.
Did we read the same article? DNS-01 challenges require updates to DNS. This means you need an API for your DNS. This means you now have to worry about DNS permissions in your application cert workflow. We’ve just massively increased blast radius! Or you could do it manually but that’s already failed.
All of this is straightforward with infrastructure-as-code. While I don’t struggle with that, I’ve watched devs and sysadmins both stare blankly at this kind of thing for days at a time.
Ahh yes the: we can’t have self signed certificates for security reasons but also can’t open up the environment to the web, and we dont have our own CA server, trifecta.
Solution: awkward, manual, certificate import process from a 3rd party vendor.
I’m sorry, but has no-one heard of https://letsencrypt.org that issues certificates via API for free?
I would not be surprised if certificates at some point will be issued for each session.
AWS makes this impossible in a few places such as a fair number of ACM use-cases.
I think your cert-per-session idea is interesting. We’d need significant throughput and processing boosts to make that happen, probably at least on the order of 10X computing speeds and 10X transmission speeds across the board minimum. These operations are computationally intense and add data to the wire so, for example, a simple Lemmy server with hundreds of users slows to a crawl and a larger site eg Mastodon goes to dialup speeds or worse. You can test at home by trying to generate an x509 self-signed cert before connecting to a website every time.
I’m sorry, but have you ever needed to manage some certificates for a legacy system or something that isn’t just a simple public facing webserver?
Automation becomes complicated very quickly. And you don’t want to give DNS mutation access to all those systems to renew with DNS-01.
You can delegate to isolated nameservers with DNS-01, there’s no need to have control over the primary zone: https://www.eff.org/deeplinks/2018/02/technical-deep-dive-securing-automation-acme-dns-challenge-validation
Yes, and that is where we enter the complicated territories…
How complicated is it to have a CNAME? /s
If you think it’s just too easy but people are still discussing it, please entertain the notion that you may have oversimplified the situation in your assessment and that as assumptions become clarified you may yet soon understand a horror that apple can’t quite grok.
Did we read the same article? DNS-01 challenges require updates to DNS. This means you need an API for your DNS. This means you now have to worry about DNS permissions in your application cert workflow. We’ve just massively increased blast radius! Or you could do it manually but that’s already failed.
All of this is straightforward with infrastructure-as-code. While I don’t struggle with that, I’ve watched devs and sysadmins both stare blankly at this kind of thing for days at a time.
Ahh yes the: we can’t have self signed certificates for security reasons but also can’t open up the environment to the web, and we dont have our own CA server, trifecta.
Solution: awkward, manual, certificate import process from a 3rd party vendor.