As mentioned above, it should have been .test
https://www.icann.org/resources/pages/name-collision-2013-12...
— We bought .dev, now what we do with all those people in the wild misusing it?
— Well, most of them misuse it with our browser, let's break at least their hacks early and loudly.
Source: The horse's mouth. I'm the guy who came up with the idea of launching .dev and .app as HTTPS-only TLDs, and I'm the one who had them added to the HSTS preload list.
It’s a huge change that surely warranted special attention before making it happen?
So yes, we didn't anticipate how many people weren't following the best practices, but that would have been hard to determine prior to doing the thing anyway. There were also lots of people who had the mindset of "We won't change anything until it stops working", so in some sense a lot of it was unavoidable. See e.g.: https://github.com/laravel/valet/issues/204 https://github.com/laravel/valet/issues/294 https://github.com/laravel/valet/issues/431 (note that Laravel users were responsible for a non-trivial fraction of the total problems experienced, and that we only discovered all this post-HSTS-preloading). The problem was repeatedly pointed out and the maintainers refused to fix it until it actually broke. So, inevitably, it broke, and then they fixed it.
I've always hated Google for egoistically claiming this tld, and ICANN for letting them.
It enables zero config local development configuration, whereby at the time it would make your locally running dev server available on .dev
So instead of having to spin up a local server and then visiting say 127.0.0.1:3000,
I could instead just visit myappname.dev and it would show me what would previously show on localhost or it would spin up a server first for that app then show it to me.
They switched to .test in response to google’s change.
Official site: https://pow.cx
Thread on change from .dev to .test https://github.com/basecamp/pow/issues/386
Visit http://pow.cx
Also, just because someone is using a fake domain doesn't preclude that from being created as a real domain name farther down the line. That's why you shouldn't use fake domain names. This problem has been known since at least the 90s and is not a good habit to get into. Them now being real makes them actually more useful (and not reliant on potentially unsynced local-only config).
I mean, I THINK this is "properly configured", but it also isn't a huge deal to avoid. Just was annoying when it started happening. Didn't FEEL like I was misconfigured. :-)
In this particular case it sounds like your resolver was set up to try an external resolution first and then append an internal domain if the external resolution failed. And as you say, this clearly worked just fine for a long, long time. Then it suddenly started failing one day, for reasons unrelated to anything you changed.
At this point, most people would find it reasonable to blame the external change for breaking their fully functional, correctly working, "properly configured" setup. Some, perhaps contrarian or perhaps more cautious, would note that the "properly configured" approach only worked so long as external systems played ball. I think it might be the case that you were bitten by this assumption that seemed safe at the time, leading to the awkward and uncomfortable conclusion that your systems were indeed misconfigured.
I would argue that, of the two options, option B is the less onerous one, and less restrictive on the future growth of the Internet. It's not that hard to set up some aliases to be able to SSH quickly into the right hosts without having to manually type out longer paths.
Or you can look at the matching username on GitHub ( https://github.com/CydeWeys ) and see that I'm a member of the Google org and owner of the Nomulus repo ( https://github.com/google/nomulus/graphs/contributors ), the software running our TLD registry and which was announced here: https://opensource.googleblog.com/2016/10/introducing-nomulu...
Or I could just dig out my GPG key and sign a message.
So I respect your healthy skepticism, but I promise I'm really me, and not just someone pretending to be me! And I'll also note that, in all my time on HN, I've never once seen someone pretending to be anyone they weren't, at least not on seasoned accounts like mine. You can go back through my 5 years of comments on here, and every time I say I'm a specific person out there in the real world, it's always the same consistent person.
Not disputing your account in any way, but please don't propagate the widespread myth that only people who own an email address can send from that email address. If you wanted to use your well-known email address as authentication, you'd need to offer to reply to a mail sent to that address.
Though, admittedly, it would be easier for a layman to verify ability to reply to an email than to verify all those features, so point taken.
Yeah, except it broke more than that.
A lot of folks use/d ".dev" and and ".prod" as internal sub-domains in their actually-owned domain (dev.example.com, prod.example.net).
For convenience you could however use the resolve.conf's "search" option to simply things, so at the CLI one could type "ssh webserv01.dev" and the resolver would would then append the company's domain to get the FQDN for the query.
Except once Google made their changes "webserv01.dev" now could go out to the Internet—especially if you had it in a browser and it tried to be "clever".
See https://jdebp.eu/FGA/dns-use-domain-names-that-you-own.html for more explanation.
Say I own "throw0101a.com", and then use "dev.throw0101a.com" and "prod.throw0101a.com". (Or you own CydeWeys.com.)
Previously, when .dev and .prod were not TLDs, it was fairly safe to type "ssh websrv01.dev" and "ssh dbsrv02.prod" because if the queries leaked onto the Internet they'd fail.
Now, with the post-Google TLD changes, if you type one of those things, and the local DNS happens to not be configured properly (i.e., the resolve.conf 'search' options is not present), then strange things can happen.
Further, if you put "websrv01.dev" in a browser now, it may go off into the Internet and try to be clever about auto-complete instead of just doing a local query.
Relying on DNS search lists to find the right host is a bad security practice that has caused security incidents even outside the context of the creation of new TLDs. It's best to always use fully-qualified domain names. A lot of people responsible for implementing DNS search lists in the first place now regret having ever created them.
More info at https://www.icann.org/en/system/files/files/sac-064-en.pdf (particularly section 4.1.3 and the preceding logic leading up to it).
In more dynamic environments, the config may have the domain as its own setting and each service as just the hostname, but the software must combine them before use, or better yet combine them in the config if variable expansion is possible in the config language they are using (ex: db_server="db-01.${domain}" with domain being defined near the top).
The more people follow decent practices at home, the fewer businesses will accidentally break because on of the admins thought it'd be alright because it works for them at home. If you set your DNS domain correctly you can also save yourself some typing effort because DNS will automatically append the network name (so you can http://test instead of http://test.internal or http://test.hamu.co). As an added bonus, you can get valid TLS certificates for your internal network devices without messing with a certificate authority of your own!
.test doesn't quite cut it.
It does seem much, much more likely that .internal will be definitely reserved for this purpose than that it will ever be delegated as an actual real TLD. If you have to pick a fake TLD to use that isn't one of the 4 mentioned in RFC 2606 that has the best chance of actually being reserved for this purpose in the future, then .internal is it.
For example, if you buy "example.com", just set your public DNS (assuming your registrar provides one) to resolv it to 127.0.0.1, then add your internal hostnames and IP addresses to your internal DNS. If you do it that way, "my-server.example.com" will simply fail to resolve unless you're on your internal network and you don't have to worry about any issues with using the reserved *.internal domain.
For example if you have a home network and a testing network, you could have one on home.example.com and the other at lab.example.com, in which case your servers would be server.home.example.com and server.lab.example.com. If you use DHCP on those networks, you simply set the domain and search-domain options and you can just enter "server/" on the devices that moves between them.
You only need to register example.com with a registrar, then you can use whatever subdomains you want wherever you want.
Good grief, usually it's because of a hairpin nat. People do that to themselves. They damage their own L3 networking and then decide that they need to damage also their entire DNS as a workaround.
It's a regular mind virus, because it's easy to implement split-horizon DNS but enormously expensive to remove it. People get used to it on one company and go and spread it on another company.
Just do a snat+dnat. These networking boxes are so expensive because they are meant to handle it, so let them do their job already. Or go IPv6 and get rid of DNAT altogether.
Why not?
Now, there has actually been an int TLD for a long time, but you probably don't visit sites in that TLD very often because it is for international organisations like the UN.
So if your configuration blocks all those actual sites well, too bad right?
However, back that long ago the trusted Public CAs were not actually forbidden from issuing certificates for names that don't belong to anybody on the public Internet. This was a bad idea, but it was not yet (at that time) forbidden. So you could pay Thawte a pile of money and get a certificate for "exchange2" your backup MS Exchange server or maybe "linux.build" your Linux build server.
And since people were using this for internal names, you'd get people asking their CA for certificates like "exchange2.int" - for the internal backups MS Exchange server right?
Obviously there can't be any effective way to demonstrate control over internal names, since you do not in fact have control over them, you've just hijacked them.
And so the end result is that there were actual publicly trusted CAs issuing certificates for names in a real public TLD without checks, because they assumed it was internal when it was actually not.
These days the CAs are required to issue only for names in the actual Internet DNS hierarchy (plus TOR) and only after seeing one of the accepted proofs of control nicknamed the Ten Blessed Methods.
Meanwhile: There is only one namespace, do not try to hijack little pieces for yourself that don't belong to you. If you want to reserve some names so that your DNS doesn't "break" then you can buy names like anybody else.