Not that simple, and he manages way more machines than me.
I just spoke with a technician on the phone. I actually wanted to reach the sales department, but it seems we concluded that there is probably a partial telephone issue with sales; their callback system sometimes goes nowhere.
So we talked for a while, I explained the problem/need and asked why a 99.95 % SLA on hardware triggers an intervention faster than a 99.99 % SLA on the public cloud.
When a problem is hardware‑related it’s detected much more easily (no ping, fire, …) so the 24/7 team intervenes without being asked.
But for an undetected hypervisor issue (because the status showed "active" when it wasn’t), the admin team decided it was working, period. So no alert, no intervention.
The ticket is sent to support, whose only purpose is to apply the SLA rule that isn’t technical. They told me they had forwarded it to the admin team, but since there was no alert…
At one point I performed a "hard reset" hoping it would lock up and potentially raise an alert. I think that’s when it showed up. There was an intervention, but not in a rush, that’s clear 
The technician I spoke with opened an internal ticket to suggest looking at a GTI/GTR instead of an SLA for the public cloud (though I’m not sure that would fully solve the problem the admin team missed).
I also suggested hiring contractors during holidays because there may have been staff on leave this weekend. I proposed four extra people to fill the gaps, financed by a GTI to be included in version 4 of the instances in exchange for a price increase if needed.
I told him there’s currently a gray area between what we think we’re being sold and reality: OVH is supposed to make the effort to intervene according to the stated SLA, not deliberately wait until Monday and only partially reimburse the failed instance.
If I opened the ticket, it’s because I had reason to believe the hypervisor was down, not just an instance system issue. We need to be listened to a bit more, even if the monitoring says everything is fine.
Next step: the sales department, if I can reach them. And no, @Gaston, not between 8 am and 9 am in the morning—I’m on UTC‑3 and for sales they’re the ones who should be calling, in theory.
After that there is indeed 24/7 support, but it’s €250 per month.
I’d like to have it, but it’s still a bit pricey for me.
Great, that call was necessary in your case 
Thanks for sharing the information with us. It really clarifies several aspects.
Regards.
These days money is the least important 
and how happy you sleep knowing there’s always OVH staff by your side, that has no price 
It's true but go tell that to the clients when you have to present the increase 
That’s the whole problem, and that’s actually (partly) why I’m leaving IT (I just got accepted into my training centre to become a nursing aide!).
Clients want 99.99% uptime but at the cost of a shared‑instance bill…
When I explain to them that such a level of availability requires deploying an entire cluster, which I then bill accordingly, suddenly it’s not such a big deal.
Even so, if a “simple” solution (like a single bare‑metal server) fails, it doesn’t stop them from pressuring us to “get it back up as fast as possible”… Yeah, except you’re not paying for HA, just a basic offering…
Especially since most clients don’t need 99.99%… Apart from the “yeah but you get it, Google penalises the site if there’s downtime”… Well, build yourself a HA/LB setup on AWS in dev‑ops mode, and pay the price that comes with it (because AWS isn’t cheap!)…
Anyway, getting back to the original topic, as raised, the big issue is that admins have to detect that an instance is buggy… and that’s not always easy…
And even then, it’s not guaranteed they’ll intervene “in an emergency” on a PCI‑instance, as mentioned in the thread, that’s reserved for hardware failures (easy to detect).
The solution, indeed, is a €250‑per‑month support plan, but that brings us back to the original problem: it starts to add up for a freelancer… which forces you to raise your prices, which makes customers complain (I don’t get it, it works without it)…
In short, hardware prices have collapsed, but the human cost of 24/7 on‑call duty remains very real, and customers often have a hard time hearing that.
A care aide and IT are completely different fields. I hope that if a granny calls you on a Saturday morning, you won’t tell her you’ll get back to her on Monday 
The technician told me they wanted to avoid raising the rates.
Individually, a €28 instance has a very low cost, and that cost doesn’t allow sending an alert on a Saturday when their monitoring hasn’t seen anything. We need to find an intelligent way to achieve this, even during holidays. The trick is probably to be able to filter the “noise” of tickets that aren’t hypervisor failures, for example, instead of trying to detect a failure, the ticket should trigger a series of very thorough automated tests on the instance, and if the tests fail, it should raise the alert with follow‑up on the intervention.
And indeed, it seems that for hardware, currently, counter‑intuitively it’s simpler/faster to obtain an intervention.
Hello, I got the “commercial” service online.
I won’t comment on what I think about it.
But I conclude that there would indeed be a problem if an OVH service goes down and is not detected by their monitoring.
The only way to alert them is confined to business hours (standard support) while the problem itself may require an intervention to try to meet the SLA. And the intervention guarantee only activates with a support plan at €250 / month, which blows up the price.
Their administrators seem not to try to intervene as long as everything is green on the monitoring.
It’s like calling the fire brigade from their station; they don’t smell smoke so everything seems fine.
A failure on a dedicated server is more easily detectable, a 99.95% SLA actually results in shorter outages than a 99.99% SLA on the Public Cloud…
To me, a 99.99% SLA is purely a sales argument, since the compensation is anyway ridiculous if uptime isn’t met.
For motivated people, in a team, building your own cluster is often more interesting.
Whether for finances, performance or availability.
But it means having sufficiently large needs to make the whole infrastructure worthwhile.
And having a small team to manage on‑call duties.
+1
I'm with you
That's roughly what I understand—while the Public Cloud makes it easy to spin up servers, that advantage is undermined by the risk of receiving no support for up to 62 hours in the event of a failure.
In contrast, a dedicated server is oddly less risky, even when you have just a single dedicated instance.
Only one dedicated server, there are a lot a lot of other potential problems!
I wrote an article on my site about this, so I’ll avoid spamming here
At a minimum, virtualize the server + a PBS server to back up the entire VM once or multiple times a day. This lets you quickly redeploy the services on another server if needed.
Personally, the bare‑metal all alone (non‑virtualized), I don’t want to hear about it any more
Yes, obviously, a VM on Proxmox or something else.
Well, yes!
The more I go, the more I also virtualize (but with KVM/QEMU/virsh without Proxmox); it’s true that it brings flexibility without too much complexity.
The problem now is this sudden price increase on the hardware (RAM and SSD).
I really like using Proxmox, it simplifies things enormously, a nice GUI.
And tools like Proxmox Backup Server to easily back up the VM, restore it, etc...
But anyway, the basic principle remains the same, virtualize the services in production, to make them as little dependent as possible on the underlying physical machine.
I’m with you, being able to have ultra‑fast recovery, cloning, snapshots... helps us cut costs (and headaches 
), come on, it’s a survival tool nowadays.
I have used virssh and Proxmox, and honestly I also stick with Proxmox; when you do a live migration and automatic failover between nodes with a few clicks... Proxmox is more convenient for me.
I tried Proxmox quickly, but the learning curve feels quite steep compared to simple console management.
Well, of course I’m sure Proxmox’s features are awesome; you just have to find the time, you know.
Totally agree. Proxmox is very powerful, but it has a learning curve and takes time to master. For a simpler and more direct use, sometimes the console and tools like Virtualmin or Cockpit are quicker to handle. The important thing is that the tool adapts to your workflow, not the other way around. When you have a spare moment, it’s worth it, but it’s not essential if what you have already works
Cheers 