I'm still waiting for Hetzner to support servers (physical and virtual) without public IPv4 addresses. I could easily free up the ~50 public addresses I'm using. One public IP will do, I can reverse proxy everything else.
But there's no support for that. So every time I spin up a 1 vCPU tiny VM, which will never connect to the public internet, I'm wasting an expensive resource. Sorry.
If you are willing to go ipv6-only on Vultr.com it brings the price of their smallest virtual-server option down to $2.50/month (the same server offering with an ipv4 address costs $3.50/month). It's nice to see them offering that kind of discount, but I have no idea whether or not there's anything similar for their more powerful offerings.
The only downside is you cannot do BGP on those IPv6-only hosts, as their BGP speaker is IPv4-only, so you cannot BYOIPv4 to those hosts, unless you route via their private network to another IPv4 enabled host first.
Oh, I wasn't even aware that was a thing. How does that work? Don't you have to buy blocks of IPv4 (which would be rather expensive) and then route them all the the Host. What advantage would Bringing your own IP have anyway?
> If you want some serious redundancy nothing beats running services anycast with BGP across multiple cloud providers.
Unless they're TCP services. Or stateful UDP services. Then you're in for a world of pain if you try to anycast them.
The list of sensibly anycast services is surprisingly low. Stateless UDP services, essentially, which isn't a big list. Especially if you're the kind of person who's using Vultr's cloud, and not a megacorp.
If you have the resources to develop backend state sync, you're probably still better serviced with a (set of) load balancer(s).
No way. Stateful TCP over anycast works fine reliably to many nines.
I run a CDN with anycast BGP, this obviously uses TCP. Works great. I've also done it with a global network of physical datacenters and our own transit/peering network but eventually migated to the cloud for reasons.
This is absolutely awesome. I'm not sure if you're a maintainer of this list, but, this is VERY useful. You just found me a provider who is scratching an itch for a specific niche location which I've had for years. I owe you a beer or three.
Huh, I just deployed one yesterday. And I'm looking at the Vultr "deploy instance" page right now and it's showing both the $2.50/mo and $3.50/mo options, at least in the "New York (NJ)" location.
This! I don't see any reason for _internal infra_ to use IPv4, if it's under your control. At least AWS lets you have "private" IPv4's only. (Dunno about the situation with GCP or Azure, happy to learn about that.) But I'd gladly set up my stuff in IPv6 and expose only the endpoints in IPv4.
Unfortunately the mantra of "ain't broke, don't fix" is law at some companies. Pushing IPv6 will be met with immediate resistance because it brings no benefits but does add the risk of something not supporting it yet.
Been pretty happy with them over the years. Pricing is good. They have a presence in Germany and in Finland, so uptime is usually excellent. The physical hardware has been an excellent experience, we've had two (!) broken drives in ~8 years. Both times the drive was replaced within half an hour. We replace most physical machines after 3 years just to be on the safe side.
We've experienced some networking issues, usually not on the Hetzner network, but there have been some peering issues with some ISPs over the years. Generally nothing too bad.
Their Cloud API is a joy to work with. It obviously isn't anywhere near as future rich as AWS, but it's got everything we need, and we can spin up VMs with a couple of simple HTTP requests.
With my previous employer, we deployed several thousand VMs at Hetzner (incidentally, we were one of their biggest customers in Germany). Really can recommend, billing was fair, support was quick and their Infrastructure worked without a hiccup for multiple years.
Im just waiting for them to offer a k8s environment…
Hetzner is great: professional, high quality, and cheap, cheap, cheap.
Their margins are low, however, so I understand it is possible to get fired as a customer if your support burden is too high and your ROI goes negative, so be on your best behavior to keep access to those prices.
> I understand it is possible to get fired as a customer if your support burden is too high and your ROI goes negative, so be on your best behavior to keep access to those prices.
That's not a nice statement.
It sounds like Hetzner will cancel your contract as soon as you submit a support request, which is not true (I personally opened in ~May 5-10 tickets within a few days when trying to boot from UEFI).
I really do not think that they track single customers in relation to their single ROIs.
We have IIRC ~1k VMs there. It's rather unreliable - VMs are migrated often, so you have to expect failure. But as you have to factor failure in in any case, our company's position has been that constant failure avoids complacent design, and so we are very happy with them (and they keep our mitigation systems well fit).
Working with them is nice, the APIs are a breeze, and are really dirt cheap. And they take security very seriously.
currently having 2 root servers in germany and 1 in Finland.
If I remember correctly I had within 8-10 years at least once a full unexpected shutdown of a server (don't honestly remember how long in took until it was back online, it was a few years ago).
A few weeks ago I could not reach for ~4 hours my (new!) server in Finland from my own Internet provider at home (it was super weird - I could ping any other server in Finland/world but not the Hetzner stuff) but at the same time it was absolutely reacheable from my mobile phone and from other ping-test-sites in CH and D => interesting, probably something to do with my ISP but I'm wondering why exactly my Hetzner server was involved, mmmhhhh... .
Maintenance downtimes (few) are scheduled and communicated by email.
Throughput is good - currently writing a web crawler and I can go up from time to time to at least 50MiB/s download if I want to (BUT I try to be "nice" to not get banned :) and anyway my processing queue cannot currently keep up with that rate).
Support is good - servers in Germany want tickets to be written in german, servers in Finland want tickets to be written in English. Both are fast (e.g. initial reaction time usually max ~1h to pick up a ticket, answers to replies usually within minutes), german support can be verbose, finnish support wasn't at all in my case (e.g. if you ask "I have problem X because blahblah so I was wondering if maybe boot on disk #2 is disabled in the BIOS?" they might just reply with "boot activated").
About HW replacement: I had so far only 1 HDD failure (some years ago, in a classic mdraid5 array) => I opened a ticket and pasted the proof of the drive failing together with the output of "smartctl" to identify the drive => got a reply asking for a downtime => agreed to the downtime and shut down the server at that time => drive was replaced and the server was booted => I then resync'ed the mdraid and that was it.
About HW upgrades: opened a ticket asking details about feasibility & pricing => got back a statement 1-4 hours later => sent back a reply agreeing to it => I was asked a few minutes later by the tech team "when" to perform the upgrade => I replied "now" and shut down the server => a few minutes later the server was up and running with the extra 32GB RAM.
So, all in all, as you probably understood, I'm currently happy with them. The costs are ok for me (especially for the servers that have >=10 HDDs - is there a better offer anywhere, honestly asking?), the reliability of the infrastructure is good, the support is good.
I was before (many many years ago) at OVH (good support + reliable, at that time) but it then became very expensive therefore I left them. Nowadays they have a more differentiated offering, but it looks messy to me, and bigger servers like the ones that Hetzner offers (e.g. 8+ CPUs with 3+ HDDs and 1 NVMe/SSD) seem to be extremely expensive to me.
Yeah, me too. Was confused why they needed to have an IP at the beginning, coming from AWS, since they have internal networking now. The public IP doesn't serve any purpose for me, and would perhaps also improve security.
Vultr has v6 only servers which come at a slight discount. I think it will become more common and they can be useful for some tasks like running chat bots and stuff for APIs with v6 support.
But there's no support for that. So every time I spin up a 1 vCPU tiny VM, which will never connect to the public internet, I'm wasting an expensive resource. Sorry.