Hetzner pulled my site over a bogus claim

ASN doesn't help when the abuse desk is a black hole. I've been down that road with Contabo in '14, OVHcloud in '19, now watching it happen to folks at Hetzner.

The real tell is whether they publish an abuse phone number. Hetzner doesn't. OVHcloud doesn't. RackNerd does, and someone actually answers. That's your canary.

Vultr's fine if you're running stateless workloads. Try explaining to their chat why your MariaDB replication needs consistent latency and watch the eyes glaze over.

Following this thread because I just moved from Hetzner Cloud to Vultr Warsaw last month. The ping from Kraków to Vultr Warsaw is 12ms stable, to Hetzner Falkenstein was 25ms with spikes.

Not the same issue but related: Vultr actually answered my abuse question in 4 hours when I was setting up PTR records. Hetzner took 3 days for the same ticket.

The $5 plan is real, not the "unlimited" lie. I monitor it with UptimeRobot and the free tier is enough for one check per 5 minutes.

The PTR record is the good point. I am fighting with the OVHcloud for two weeks now to set the reverse DNS for the mail server. The form is broken and the support is sending the links to the articles I already read.

The Vultr has the API for the PTR and it works from the first time. This is the difference between the company who builds for the engineers and the company who builds for the accountants.

I am still on OVHcloud for the big storage because the price is good for the dedicated server in Gravelines but the support is the lottery. Sometimes you get the human, sometimes you get the loop.

I rewrote my entire deployment pipeline to avoid exactly this class of problem. Terraform state lives in a self-hosted Gitea instance on a Kimsufi box in Roubaix. If OVHcloud pulls that, I've got bigger problems than DNS.

The actual answer is geographic redundancy with split authority. Hetzner Falkenstein primary, Vultr Stockholm secondary, Route 53 doing health checks. Yes, it costs more than a single €4.51 instance. No, I don't care.

Rust-powered health checker pings both endpoints every 30 seconds. Failover is automatic. Complain to both providers simultaneously and you still lose, but at least you lose gracefully.

Actually, (and I say this as someone who keeps a (spare) NUC in my mother's attic in Košice), the multi-provider redundancy is (theoretically) sound but most folks in this thread are running WordPress blogs, not distributed systems.

My (actual) advice for the non-technical: buy two cheap VPS instances, not one expensive one. RackNerd's $10/year plans — https://www.racknerd.com — in two locations, rsync the WordPress nightly, change DNS when the primary dies. Total cost: less than one month of solicitor fees.

The (sad) truth is that Hetzner's €4.51 plan is priced exactly where they can afford to ignore you. That's the (entire) business model.

rsync for WordPress misses the database unless you're dumping it first, and most cheap hosts throttle I/O hard enough that a nightly dump kills performance for ten minutes.

I've done the two-VPS dance for clients. Works until they forget to update the SSL cert on the secondary, or the theme update breaks on PHP 8.1 while the primary is still on 7.4. Now you've got two broken sites instead of one.

If you're going to run redundant, run identical stacks and test the failover monthly. Otherwise you're just paying for a false sense of security.

The identical stack is the key. I see this in Utrecht with the hosting clients: they buy the "hot spare" but never test it, then the failover fails because the MySQL version drifted by two minor releases.

OVHcloud's public cloud has the managed database now, starts at €15/month. Not cheap, but it solves the replication drift. Hetzner has no equivalent unless you're running your own patroni on their cloud instances.

For the original problem: the abuse desk response time correlates directly with overselling ratio. Hetzner's cloud is profitable at €4.51 because they pack 40:1 on the CPU and answer tickets in batch once per day.

This. The overselling is the thing nobody talks about. I had a Hetzner CPX11 that couldn't burst past 30% because the neighbor was mining Monero. Took three tickets to get migrated to another node.

The abuse desk and the performance desk are the same people, and they're underwater. You think they're going to carefully review your DMCA counter-notice when they've got 200 "my site is slow" tickets in the queue?

Actually the CPX11 is the shared CPU and the CX21 is the dedicated vCPU for the €8.21. The price list is public and the overselling is documented in the FAQ if you read it.

Actually I am not defending Hetzner, I am saying the expectation is wrong. You pay for the Toyota, you complain about the Ferrari performance. The abuse desk is slow because the price is low, this is the same at Contabo and RackNerd and everywhere.

Actually the solution is to pay more or to host yourself. There is no third way.

Self-custody is the only answer I trust. In Medellín I pay about 80 USD for a 300mbit symmetric fiber. The electricity is unreliable but the UPS handles 4 hours and the generator handles the rest.

I run Proxmox on a used Dell R730 from the US surplus market. Total cost: $400 for the server, $200 for the UPS. Monthly operating: about $15 including the fiber and the electricity for the generator fuel.

No abuse desk. No DMCA theater. The only takedown notice I get is from my mother when the WhatsApp backup is slow.