In the world of the glue that holds the web together, A is a DNS record that maps a domain name to an IPv4 address. AAAA is a DNS record that maps a domain name to an IPv6 address. A website without an AAAA can only be used over IPv4. (The opposite is also true, a website like http://ip6only.me/ that has only an IPv6 AAAA DNS record cannot be reached over IPv4.)
A user noticed that Reddit recently turned off its AAAA records, and posted asking why it was now an IPv4-only site.
> A website without an AAAA can only be used over IPv4.
Nit: if IPv6 is still enabled server-side, it can be used over IPv6 but you can't rely on public DNS for discovery and may have to set Host headers manually.
Reddit used to have an AAAA (DNS IPv6-equivalent of an A record). They deleted the record. Users have noticed and posted.
> Yea, we found a couple of internal issues that need some debugging. Since it's a holiday week we decided to rollback. Better safe than sorry.
>
> It'll be back, don't worry.
A type DNS records are used for looking up IPv4 addresses, AAAA records are for IPv6. This effectively takes Reddit off of the IPv6 internet and is a bit of a step backward.
This really (continues) to burst the trope here on HN that IPv6 is "easy".
If you look at any discussion of for example expanding IPv4 address space, lot's of "experts" come out of the woodwork claiming IPv6 is the same as IPv4 and/or that it is easy. No it is not just an address space expansion and no it is not easy for many use cases.
Even the big players (Azure / GCP) struggle to fully enable an IPv6 only or IPv6 supporting servicing offering across their products.
They had an internal issue that needed debugging and so they decided it was easier to just disable it all-together for the holidays. This is less an indicator for complexity and more an indicator of apathy. An ipv4 issue would have obviously prompted a much more swift response. In fact considering this is reddit, I'm surprised this hasn't happened more often considering they seemingly go down every other day regardless.
The only ones making this hard are apathetic engineers who keep kicking the can down the road for the next generation. Arguments against ipv6 are like sailers on a sinking ship bickering with the rescuers that while the ship is completely underwater at this point, they can still float on their backs for a day or two, so their help is unecessesary.
IPv6 is inevitable, you can either deal with it now while it's hard, or deal with it later when ipv4 addresses are too expensive and your infrastructure is already so entrenched that transitioning is a nightmare. The latter is the situation Azure & GCP are slowly finding themselves stuck in, and they have only themselves to blame.
Many of the existing internal networks are using private, non-routable IPv4 ranges. And exposing random hosts on the internet is a bad practice, security-wise. The only things that needs routable IPs are VPN endpoints and load balancers.
Together, that means there is no need to ever get rid of IPv4 on the internal networks. Yes, maybe the workstations will have IPv6 for random website access, but that old NAS box will always be at 10.1.2.3.
This was re your "IPv6 is inevitable" comment -- we can get pretty far with IPv4. In particular, as long as you keep using third-party load balancers, you could probably keep using it for decades. And you may never have to migrate your internal infrastructure at all.
> IPv6 is inevitable, you can either deal with it now while it's hard
20+ years on, and IPv6 is still hard. This is unlike other things that were hard 20 years ago; TLS was hard 20 years ago, and now it's not so bad. 20 years ago, databases were hard, now they're not so bad. Etc. But IPv6 is still hard, so it shouldn't be surprising that people are still waiting.
OTOH, IPv6 is widely deployed in (many) mobile networks, so using it can eliminate the NAT layer(s) that are almost certainly present for IPv4 on mobile.
For those of us who have been around the block - we've been hearing from these "experts" that IPv4 is over for decades. It's boring at this point frankly.
Wake me up when old.reddit.com is the default. Actually, strike that. This can't be helped until we all meet with Aaron again, but then we won't need Reddit.
they are trying so hard to actually try to make it as an ad space. They added link your facebook and apple accounts. Seriously who tf is doing that. They want to push their ads product.
Not that they can't hire product managers, they don't hire product managers that understand the audience and what reddit the actual product is. IMO theres better ways to monetize their audience than ads.
How long should a protocol upgrade take? I cant help but feel that IPv6 has a flawed upgrade path.
It would be so much better to upgrade to a protocol that is directly compatible with IPv4 and does not require deployment and maintenance a second network in parallel to IPv4.
That doesn't seem possible. Fundamentally you need more addresses than IPv4 can represent, so legacy programs wouldn't be able to handle the addresses of any system which solved that problem.
Now, you could have an upgraded system which kept a lot more the same from IPv4 and just added more addresses. But I personally think the extra kitchen sink added in IPv6 is only a small part of why it's adoption has struggled.
I don't work at Reddit so I can only guess, but one does not have to convert most of their infrastructure to ipv6. Traffic terminates to load balancers and those would need ipv6 listeners. Everything behind them can remain ipv4 for the next million years or so if one was so inclined. The only gotchas would be anything inside their datacenter that hairpins back to the load balancer for whatever reason and there are easy work-arounds for that by creating WAN or internal vips that are ipv4.
32 comments
[ 3.1 ms ] story [ 50.7 ms ] threadA user noticed that Reddit recently turned off its AAAA records, and posted asking why it was now an IPv4-only site.
Nit: if IPv6 is still enabled server-side, it can be used over IPv6 but you can't rely on public DNS for discovery and may have to set Host headers manually.
> Yea, we found a couple of internal issues that need some debugging. Since it's a holiday week we decided to rollback. Better safe than sorry. > > It'll be back, don't worry.
Nothing to see here, really.
If you look at any discussion of for example expanding IPv4 address space, lot's of "experts" come out of the woodwork claiming IPv6 is the same as IPv4 and/or that it is easy. No it is not just an address space expansion and no it is not easy for many use cases.
Even the big players (Azure / GCP) struggle to fully enable an IPv6 only or IPv6 supporting servicing offering across their products.
The only ones making this hard are apathetic engineers who keep kicking the can down the road for the next generation. Arguments against ipv6 are like sailers on a sinking ship bickering with the rescuers that while the ship is completely underwater at this point, they can still float on their backs for a day or two, so their help is unecessesary.
IPv6 is inevitable, you can either deal with it now while it's hard, or deal with it later when ipv4 addresses are too expensive and your infrastructure is already so entrenched that transitioning is a nightmare. The latter is the situation Azure & GCP are slowly finding themselves stuck in, and they have only themselves to blame.
Together, that means there is no need to ever get rid of IPv4 on the internal networks. Yes, maybe the workstations will have IPv6 for random website access, but that old NAS box will always be at 10.1.2.3.
20+ years on, and IPv6 is still hard. This is unlike other things that were hard 20 years ago; TLS was hard 20 years ago, and now it's not so bad. 20 years ago, databases were hard, now they're not so bad. Etc. But IPv6 is still hard, so it shouldn't be surprising that people are still waiting.
OTOH, IPv6 is widely deployed in (many) mobile networks, so using it can eliminate the NAT layer(s) that are almost certainly present for IPv4 on mobile.
Bit more details would have been nice. Now i fear of poorly assuming they made internal network ipv6 too?
Not that they can't hire product managers, they don't hire product managers that understand the audience and what reddit the actual product is. IMO theres better ways to monetize their audience than ads.
You seem to think the product managers have the user's best interest in mind.
It would be so much better to upgrade to a protocol that is directly compatible with IPv4 and does not require deployment and maintenance a second network in parallel to IPv4.
Now, you could have an upgraded system which kept a lot more the same from IPv4 and just added more addresses. But I personally think the extra kitchen sink added in IPv6 is only a small part of why it's adoption has struggled.