166 comments

[ 5.2 ms ] story [ 213 ms ] thread
migadu.com is also down, connected?

edit: also can't get access to ovh vps via ssh

Lot of websites will probably be impacted
Migadu is indeed using OVH servers.
Azure is also having issues with deploying and starting virtual machines.

EDIT: To those having issues starting/deploying Windows virtual machines in Azure and are using ARM to do so: change the OS type to Linux instead of Windows. This seems to resolve the following issue:

> Error: No version found in the artifact repository that satisfies the requested version '' for VM extension with publisher 'Microsoft.WindowsAzure.GuestAgent' and type 'CRPProd'

That may be an interesting effect. People test their failover solutions if one of their provider is down when the provider is not actually down. But when it is actually down, everyone else is executing their failover plan at the same time, and compete for other clouds resources.
Hence the BS of cloud marketing imagine AWS US-East going down there is no capacity to absorb that anywhere.
Do you have any numbers/sources to back that up? I find that hard to believe, mainly because US-east1 goes down so much. It's never been hard-down across three AZs level of being down with no EC2 no VPC no nothing, but S3 being down in the whole region is pretty freaking close!
AWS us-east-2 (Cincinnati) is generally pretty reliable; it pays to use your head and not simply use the defaults.
There was a leaked dock that detailed AWS datacenters a while back and US-east dwarfed everything else.
Everything else in AWS maybe, but that's a different claim than the rest of the cloud (Azure, GCP, Oracle...) couldn't absorb that load.
For each cloud provider spare capacity is portion of their actual capacity so considering anything beyond GCP is not even a thing. Now since most of the spare capacity is being utilized as spot instances there is not that much in reality as many spot workloads are actually cost optimizations and would shift to other types if spot instances are not available.
Interesting indeed but I doubt that "everyone else" is executing their failover plan, simply because most don't have a failover plan including running on another cloud provider
(comment deleted)
looks like funny routing.. packets to ovh.nl go to Asia, and then suddenly find a faster way back to me

   3  0.ae21.xr4.1d12.xs4all.net (194.109.7.169)  7.324 ms
      0.ae21.xr3.3d12.xs4all.net (194.109.7.173)  5.929 ms
      0.ae21.xr4.1d12.xs4all.net (194.109.7.169)  5.864 ms
   4  0.et-1-1-0.xr1.tc2.xs4all.net (194.109.5.7)  7.562 ms  5.487 ms
      0.et-7-1-0.xr1.tc2.xs4all.net (194.109.5.5)  5.551 ms
   5  asd-s8-rou-1041.nl.as286.net (134.222.94.216)  6.034 ms  5.759 ms  5.608 ms
   6  ae11.cr6-ams1.ip4.gtt.net (213.200.117.178)  6.620 ms  7.938 ms  6.670 ms
   7  80.231.85.162 (80.231.85.162)  7.348 ms  6.170 ms  6.320 ms
   8  if-ae-45-2.tcore2.av2-amsterdam.as6453.net (80.231.152.50)  310.705 ms  257.678 ms  257.278 ms
   9  if-ae-14-2.tcore2.l78-london.as6453.net (80.231.131.160)  257.217 ms  256.759 ms  259.044 ms
  10  if-ae-2-2.tcore1.l78-london.as6453.net (80.231.131.2)  258.850 ms  256.723 ms  256.735 ms
  11  if-ae-12-2.tcore2.mlv-mumbai.as6453.net (180.87.39.22)  257.349 ms  255.042 ms  257.338 ms
  12  if-ae-16-2.tcore1.svw-singapore.as6453.net (180.87.12.226)  248.550 ms  257.559 ms  329.240 ms
  13  if-ae-2-2.tcore2.svw-singapore.as6453.net (180.87.12.2)  307.149 ms  255.582 ms
      be101.mrs-mrs1-sbb1-nc5.fr.eu (54.36.50.135)  181.205 ms
  14  be101.mrs-mrs2-sbb1-nc5.fr.eu (54.36.50.159)  180.612 ms *  179.738 ms
  15  * * par-gsw-sbb1-nc5.fr.eu (54.36.50.228)  186.377 ms
Maybe they accidentally misconfigured the data siphon to the Chinese state to not only send duplicate packets but the originals too!

(This is an unsubstantiated thesis that has no collaborating evidence)

Singapore != China
Have you asked China this question lately?
Seems to be back 07:41 UTC.

I became able to reach our servers in London, and www.ovh.com, from UK locations and from DigitalOcean.

Still down from here
Both my Montreal based servers are still down. Seems like their resellers are also, unsurprisingly, impacted
2 10.80.71.15 (10.80.71.15) 1.095 ms 10.80.70.253 (10.80.70.253) 0.919 ms 10.80.71.15 (10.80.71.15) 1.067 ms

3 138.197.249.124 (138.197.249.124) 1.365 ms 138.197.249.102 (138.197.249.102) 1.354 ms 138.197.249.100 (138.197.249.100) 1.549 ms

4 138.197.251.140 (138.197.251.140) 1.179 ms 1.168 ms 138.197.251.136 (138.197.251.136) 1.114 ms

5 * 138.197.249.217 (138.197.249.217) 1.167 ms 1.155 ms

9 vl1272.lon1-eri1-d1-a75.uk.eu (213.186.32.253) 2.594 ms be101.lon1-eri1-g2-nc5.uk.eu (91.121.215.119) 2.654 ms 2.466 ms

14 (51.89.234.11) 2.102 ms 2.091 ms 2.076 ms

"No impact expected" - famous last words
Tough timing, considering their IPO is tomorrow [1].

[1]: https://ipo.ovhcloud.com/sites/default/files/2021-10/OVHclou...

Pure coincidence.
not really, from Octave’s twitter it seems that DDoS attempts have increased which is usual right before an IPO. They were installing new gear because of the higher DDoS.
Could you please elaborate on this? I’ve never heard of that being related to an IPO before, so I’m wondering if this is a standard practice for competitors to get known groups to do this sort of thing, or if this is an economy destabilization measure that’s in the playbook of some nation state actors. Any insight there?
*NATION state actors. Damn mobile keyboards. Stupid app wouldn’t let me edit the reply either, hence this.
The term is state actor, anyway.

A state is a sovereign political division, or a subdivision within a federation (France, Arizona).

A nation is a community of people with a common culture, language, ethnicity etc (the Navaho, the Welsh, the Japanese).

When these are the same entity, it's a nation state (Japan), but there are many cases where they aren't the same (Belgium, China, USA, Russia).

> A nation is a community of people with a common culture, language, ethnicity

Nitpick: The belief of the existence of the said common ground is actually more important than its actual existence. France is considered a nation state even though there are several ethno-linguistic groups in France. A few illustrations:

- less than half of the French population during the French revolution spoke French as their native language.[1]

- The second Nobel Prize of Literature from France (Frédéric Mistral) didn't wrote its work in French, but in Occitan[2].

- My family has lived in France for as long as genealogy could trace back, yet half of my great-grandparents only learnt French in school, they didn't spoke it at home.

[1]: I don't have an online source for that, nor the exact figure: I read it in Frernad Braudel's L’Identité de la France

[2]: https://en.wikipedia.org/wiki/Occitan_language

"Nation-state" is an little bureaucratic Americanism because if you talk about a malicious state actor there everyone assumes you mean Florida.
It's more likely to be a "DDoS until you pay a ransom" sort of deal where the attacker thinks the company will value uptime more highly during that period.
You've clearly never run a web service of any sort... you can expect DDoS in response to pretty much anything you do. Not seeing in increase in DDoS attacks before an IPO would be more concerning.
Like other said. Its blackmail for money. Essentially "would be a shame if your website is down during your IPO, maybe you should pay us so we don't DDoS you to hell".

I imagine this is especially common here since one of the services offered by OVH is DDoS protection. Hence being down from a DDoS would be even worse press.

So ovhcloud seems up, where's that hosted?
Stuff is apparently starting to come back up. Mostly still inaccessible though.
If this is an unforced error that would be monumentally stupid.
> Those last few days, DDoS attacks intensity has grown a lot.

not unforced, prolly ransom requests before IPO

Yes, that would be the expected thing. Around an IPO you should have your house in order otherwise you will get burned.
Wasn't the fire in one of the Strasbourg's datacenter the night after the announcement of the IPO ? They have bad luck.
they also have garbage it support
This is one of those things I keep hearing but in my case I experienced the opposite. I have been using OVH for a long time, and their paid support is actually really good.

One of our projects used to get ddosed and as a small company the only reason we managed to survive was thanks to OVH. When the attackers kept changing their attacks, we were direct to talk to their engineers and they helped fix the issue, all we had to do was send a bunch of packet captures to them.

Over the years since then we had a few issues and every single time their customer support was really helpful.

I have used paid support from google, azure and amazon, but I have had a better customer service experience with OVH overall.

Maybe competitors doing their worst to cripple the IPO? Are hosting/CDN firms evil enough?
Ah, I was wondering why my scraper was failing a lot this morning. good to know
probably bgp? It seems bgp is always the issue?
Well, that's a bit like saying SMTP or IMAP is always the issue when your email stops working.

If they screwed up routing, BGP was very likely to be involved, but the cause was probably human error, failure of process design, failure of system design, etc etc.

Our whole OVH infra is down, composed of OVH baremetal + SoYouStart an OVH sister company.

http://travaux.ovh.net/

is also down which is their WIP tasks dashboard, having worked there I can tell that something definitely went very very wrong

Update:

We recovered our OVH sandbox env (yay) but prod is still very down, guess they are rolling back DC by DC

Update 2:

Our call center provider has VoIP provided by OVH as well, they are down as well... It's going to be a very long day...

It looks like only IPv4 is affected. IPv6 is working fine to a couple of test endpoints.

   3  2a02:c28:1:6506::106  2.131 ms  2.124 ms  2.117 ms
   4  2a02:c28:11:6::100  15.695 ms  15.689 ms  15.682 ms
   5  2a02:c28:1:1900::19  15.998 ms  15.991 ms  15.985 ms
   6  2a02:c28:0:1819::18  15.259 ms  15.051 ms  16.975 ms
   7  2a02:c28:0:1718::17  16.968 ms  16.403 ms  15.619 ms
   8  2a02:c28:0:1731::31  15.406 ms  18.620 ms  18.554 ms
   9  2001:7f8:4::3f94:2  12.025 ms * *
  10  * * *
  11  2001:41d0:aaaa:100::5  21.622 ms  38.813 ms 2001:41d0:aaaa:100::3  37.231 ms
  12  * * *
  13  * 2001:41d0::25f1  19.507 ms 2001:41d0::c68  20.892 ms
  14  2001:41d0::513  19.685 ms  19.674 ms 2001:41d0::50d  18.248 ms
  15  2001:41d0:0:50::5:10a1  19.092 ms  19.917 ms 2001:41d0:0:50::5:10a5  20.126 ms
  16  2001:41d0:0:50::1:143f  19.375 ms 2001:41d0:0:50::1:143b  24.609 ms 2001:41d0:0:50::1:143d  18.933 ms
Ugh. You reminded me that while OVH used to hand out a /128, if you needed more address space you had to pay more for “enterprise” IPv6 or something.
Is their enterprise prefix /64? That would be even funnier. :)
lolwut ?

Don't consumer ISPs hand out /64s or /48s ?

Yeah, Comcast hands out /64s on residential connections. /128 is literally just a single address, which is absurd if true.
Unfortunately, it was(is?) true. Not just OVH though, even Scaleway also had these silly /128 shenanigans.
they give a /64 to all dedicated servers
When you use OVH you have to be ready for this kind of stuff. They are cheap for a reason.
Well, even if you don't use OVH you should be ready for this kind of stuff.
Hetzner is also cheap but for some reason they don't seem to suffer from these kinds of issues.
Well they are not French.
i guess facebook must be french then, judging by what happened last week.
Facebook does not go down several times a year and their datacenters don't literally burn down because of how cheap and useless they are.
Thanks, it's about time this reputation on french sloppiness ends.. What would people have said if "move fast and break things" moto was coming from a french company.
I wish they had US location
With them recently announcing that they will start charging tax for US sales, `ash1-speed.hetzner.com` resolving to an IP address (yet unroutable) under a different Hetzner ASN, and this one[0], I believe they are already in for a US region possibly in Ashburn, Virginia.

[0] https://www.bizapedia.com/va/hetzner-us-llc.html

Wow thnx for the info that is exciting news
Hetzner is the cheapest and their cloud offering is solid. Heavily under ratted provider. Price to value, interface, documentation, API, libraries. Everything top notch.
(comment deleted)
I was just going for a peaceful morning walk, trying to start the day with some sunlight. As soon as I sit on the first bench with sun I can find, the dreaded message comes: "everything is down".

Thank you! I had almost forgotten about my beloved provider since the last outage :)

One more and you get to make a sign.

It has been [17] days since the last outage.

Our servers in Strasburg were down for around 10 minutes, now they are fully operational but it looks like some of their sites are still experiencing issues [1]. On the positive side: it was my second day as on-call and I already had the worst incident(no ssh, provider site is down).

[1]: http://status.ovh.com/

I don't think this counts as a worst incident. Since there is nothing you can do you might as well go take a nap and wait for it to pass.

On the other hand something like Gitlab's "We deleted the production database and our backups were not working for months" sounds much more stressful to me.

LOL. OVH has turned into the laughing stock of the industry. Its a shame really, I always used to gravitate towards their offerings vs competition because I liked the fact that they have servers literally everywhere...but what good are servers everywhere when u get a major outage every few months?

Perhaps I'd tolerate the constant outages if they were actually any good at communicating. Their status page sucks. They rarely update it properly, and when they do, its hard to find the information you need and that you know relates to your servers.

If you have to deal with a few outages for a 90% cost reduction, then you just make sure you have a plan. I've found OVH to be mostly reliable.
Things do happen even with AWS.

Minus the fire incident, can't recall any outage. Not the cheapest offering out there but been running many things on it on autopilot. Support is solid too, always had a good experience.

Well, some products/businesses don't need 99.999% uptime and that's fine. Hell, I'm working on a small shop and we had to update some service that is not used constantly by our customers and when I suggested to have a downtime of 5 minutes everybody looked at me as if I were a caveman. The other alternative was to come up with some custom strategy to have two instances of the same service running at the same time and kill one of them when we know the other is healthy. Now, for a service that is constantly used that's a good idea, but for our scenario a 5 min. downtime is fine... we are not Google.
I'm not affiliated with them in any way, but historically, Online.net (now rebranded as "Scaleway") has always been the main competitor of OVH in France. They were known for their "Dedibox" when I was younger.

OVH kinda won because their VPS pricing was always dirt-cheap compared to the cheap dedicated servers provided by Online.net/Scaleway. Since then OVH VPSes got expensive, and Scaleway started offering cheap VPSes, the prices got aligned. I know a few people who used to be OVH-fanboys and they all migrated to Scaleway and Vultr the last 4 years.

But I agree, the competition for unmetered reliable VPSes is getting thinner and thinner… People are even started to consider metered VPS solutions like Vultr…

Rasmus Lerdorf (the author of PHP) did an objective non-sponsored comparison[1] in 2019, I'm using IONOS personally, but since then strato.de appeared. (Both are German) I'm thinking about adding another node from them to avoid relying on only one vendor.

[1] https://toys.lerdorf.com/low-cost-vps-testing

Router maintenance, no impact predicted. Many people saying that if "all servers down everywhere" is not impact, they wonder what actual impact is.

https://twitter.com/ovh_status/status/1448185498812485633

Which is stupid. If I say I'm walking to the store and will be gone for 30 minutes, then get hit by a truck and spend a week in the hospital, nobody would say "30 minutes, huh?"
It seems like (only?) their name servers are down. I got my website back up by avoiding to use these (it's hosted on a dedicated server at OVH). They'll probably have the same kind of issues as facebook did while trying to get things back up.
What could be the best way for status page apart from third party services? I remember one project that used IPFS as distributed status page. https://news.ycombinator.com/item?id=16273609
That third party service has still got to be hosted somewhere though. What happens when the cloud they're running on goes down?

The rule should be host your status page on your competitor's cloud. If you're AWS, host it in Azure, if you're Azure, host it in GCP, if you're GCP, host it in AWS. (Linode, Digital Ocean, OVH, etc can do their own dance.)

Mxroute is also down as a result of this outage.
Octave Klaba (OVH CEO) on Twitter [0] :

  Following a human error during the reconfiguration of the network in our DC in VH (US-EST), we had an issue on all the backbone. We're gonna isolate the DC VH then fix the configuration

  Those last few days, DDoS attacks intensity has grown a lot. We decided to increase our DDoS handling capacity by adding new infrastructures in our DC VH (US-EST). A bad router configuration caused the network crash.
[0]: https://twitter.com/olesovhcom/status/1448196879020433409
(comment deleted)
It's now back: https://status.ovh.com

Short lived incident fortunately...

Does not work for me
Apparently it's only the status page. My mail servers have been down for just about a hour.
Down here. Comcast west coast.
Down for me too. Although, seems like this time only the status page is down.
I can ping the status IP but no HTTPS available from it, so routing is working externally.

  72 bytes from 5.135.138.70: icmp_seq=1 ttl=249 time=276 ms
  72 bytes from 5.135.138.70: icmp_seq=2 ttl=249 time=1041 ms
  72 bytes from 5.135.138.70: icmp_seq=3 ttl=249 time=313 ms
  72 bytes from 5.135.138.70: icmp_seq=4 ttl=249 time=198 ms
  72 bytes from 5.135.138.70: icmp_seq=5 ttl=249 time=226 ms
  72 bytes from 5.135.138.70: icmp_seq=6 ttl=249 time=390 ms
  72 bytes from 5.135.138.70: icmp_seq=7 ttl=249 time=865 ms
Forgot I was using IP over carrier pigeon
HughesNet?
Even worse if you can believe that. Hopefully it is temporary.
How is it still down for me..?

Also, I can access my OVH VPS from an Italian IP, but if I try with a VPN from outside my country, I get error 521.

Seems like the best DDoS attack is to persuade the target to change their backbone configuration. Then they’ll take themselves out.
so. how do these things happen? juniper had/has "commit confirm" for ages. or that would not help for certain things?
Easy to be really sure of a config change that only later obviously was terrible?
So the CEO's twitter feed had information at 08:00 UTC, but the official ovh_status feed didn't even acknowledge the incident until 08:22.

That seems the wrong way round.

The description of the ovh_status account explicitly mentions it's an unofficial account:

> Unofficial OVH Status Feed

Argh. And it has a pretty user image which attempts to say that.

Only the image is conveniently cropped to a circle so that the "Un" part isn't visible and it reads "OFFICI".

Twitter and Reddit are the true status pages
didnt facebook have the exact same issue the other day but theres was all there datacenters rather than a single DC?
Exact same? I doubt it. The only similarity (that we know of) is that there was a router configuration change. In Facebook's case, there wasn't even supposed to be a configuration change, just a look at some data.
> "gonna"

Truly, only a professional outfit could have written a Twitter update with spelling like this.

irony being you think twitter updates is professional
The tweet is in French. That was a rough translation
(comment deleted)
(comment deleted)