7 ms·
Be prepared to incredibly unreliable disks. Or better, ask for new disks for your box, even if this will require some additional cost. Clarification: I had bad
by antirez 13y ago
Be prepared to incredibly unreliable disks. Or better, ask for new disks for your box, even if this will require some additional cost.
Clarification: I had bad experiences with disk reliability, I told different friends, that in turn had very bad experiences including multiple disks in RAID breaking in the same time range.
- sspiff 13y agoWhat makes you say this? I've been running on Hetzner for about half a year now, no issues so far.
- carlio 13y agoPresumably his bad experiences make him say this.
- sspiff 13y agoHe only added this clarification after I asked, so my question is not as redundant as it would seem.
- gmac 13y agoMy 2¢: I've been on Hetzner for a few months, experimentally, and the sporadic 'offline uncorrectable sector' emails I'm getting from smartd are making me nervous.
- FooBarWidget 13y agoMaybe you're not stressing your disks enough. I recently setup a database server and a disk broke (as in becoming unaccessible, not just bad sectors) within a day.
- annnnd 13y agoI seem to recall a research by Google which concluded that stress is not really an important factor in disk reliability... Can't find a link now though. Of course, that doesn't make parent's claim any more true; just pointing out that this is probably not the reason.
- bosie 13y agois this the google research paper you were talking about: https://static.googleusercontent.com/external_content/untrusted_dlcp/research.google.com/en//archive/disk_failures.pdf https://static.googleusercontent.com/external_content/untrus... ?
- WA 13y agoWhy?
- Argorak 13y agoHetzner reuses disks to reduce cost. So a "new" server might come with old disks with shorter life-spans. Thats all okay, especially as they are not a 'premium' host. As long as you are aware of this, they are a good host. As others said, you can ask for new disks at a price. Also be aware that their abuse department shoots first and asks later. I know of servers that were off the grid for multiple days without access because of abuse on a friday - everyone relevant went on weekend. They _are_ cheap and that comes with a few trade-offs. Doesn't make the product bad, necessarily. If you are cash-strapped and need raw power, buy without hesitation.
- kbar13 13y agoI don't see logic behind not powering off an abusive machine, no matter if it's saturday night at 3am. The internet doesn't run on your schedule. If your server has been compromised (or not, heck we all know there are baddies out there), and it's actively hosting a phishing site, taking part in a DoS attack, etc, why shouldn't a provider be allowed to take action to protect their network, other users in general, and at the end of the day, perhaps the reputation of your site as well? Blaming providers for the inability to secure your server is incredibly irresponsible. Unless service was interrupted because Hetzner misread an abuse report for another server (which is highly improbable and most likely is not what you implied in your original statement), you shouldn't ever be pissed that your host took action to prevent abuse from occurring.
- petsos 13y agoYou assume a definitive definition of "abusive".
- kbar13 13y agoand you don't? things that warrant disruption of service and are defined as abuse: 1) any type of DoS attack 2) phishing 3) malicious activity (ssh brute force, port scanning without consent, etc) 4) cp 5) anything that generates a court order with jurisdiction, including DMCA if you have anything to do with the US 6) spam/spamvertised site there are more things that can be defined as abuse, but the others are more or less discretionary, imo.
- orf 13y agoI run a hosting service on Hetzner and we haven't had a disk failure on any of our servers for a year or so.
- carlio 13y agoI have also had many disk failures, including two RAID'd HDs failing within a few days of each other. I think we've lost 5 HDs so far. (We only have 1 server)
- antirez 13y agoI heard this story multiple times, apparently it is possible in the new servers to ask them for new disks and this solves most of the issues. Also the procedure to get the disk replaced is insane and involves you dealing with disk IDs... It's a RAID, then should just check what is the broken disk and replace it, end of the story.
- ElbertF 13y agoYep, I had disk issues on a new box too. Hetzner is great though. Also don't be too worried about high ping times, I recently moved http://readable.cc http://readable.cc away from Linode and the site is much, much faster (and I'm in Australia, the other end of the planet.)
- grk 13y agoI can confirm this, we're replacing at least two disks every month. You really have to design your architecture that allows a server to be taken offline for a few hours for disk replacement. Also, never use raid 0 on Hetzner.
- epic9x 13y agoRAID 0 is generally frowned on in pretty much all circumstances. Is there a specific reason you're using it? (Legitimately curious.)
- oellegaard 13y agoWe have a lot of servers running there and they indeed do fail once in a while. We designed it, so that we can recover very fast, by having a standby server ready. Its usually faster to order a new server, than HD swap.
- stiff 13y agoBesides the mentioned issues with reusing disks, they also had used and maybe still use some faulty series of Seagate drives e. g.: http://forums.seagate.com/t5/Barracuda-XT-Barracuda-Barracuda/ST31500341AS-Seagate-worst-nightmare/td-p/16442 http://forums.seagate.com/t5/Barracuda-XT-Barracuda-Barracud... So beware of having those installed, under high I/O they will fail at a random instant.
- Udo 13y agoA Barracuda on my Hetzner server failed 1 hour ago. This is the second time I had disk trouble at Hetzner - the first time both RAID disks failed at the same time. However, support is acting incredibly quickly, within 5 minutes of sending the support request they got to work on the server.
- annnnd 13y agoBoth disks failing at the same time might indicate that the problem is with the controller or cables. One of our disks (at Hetzner) died on us a few months ago and unearthed an issue with backups. :( Because of missing data we asked for a data cable replacement first (instead of just replacing the disk) - and it worked. Even more, it works without problems since. YMMV of course. BTW, I have worked with many providers but Hetzner is my favourite. Great support and very few problems.
- Udo 13y agoIn my case I think it actually was the drives though, it was the same model Barracuda. Looks like they're replacing them with Toshibas now, so I'm hopeful. > BTW, I have worked with many providers but Hetzner is my favourite. Great support and very few problems. I'll second this. Support is/was often a horror story for me, but so far Hetzner has been absolutely impeccable. Contrast this with Mediatemple, where I once had to engage in a lengthy support battle with a chain of incredibly ignorant and condescending support staff before being escalated to a senior person who finally, finally admitted I had found huge problem with the cluster. This took several days of intermittent outages, dozens of emails, and a lot of energy...
- boyter 13y agoOdd. I am thrashing the disk of my two hetzner servers with with massive amount of writes and reads for http://searchcode.com http://searchcode.com without any issues over 2 years. One takes care of HTTP/DB and the other the search index. The DB server has pumped over 10 TiB of data in the last 80 days (mostly non cached) and the index is rebuilt constantly and well over 1.5 TiB in size. Anecdotal I know but another data point to consider.
- vidarh 13y agoSimilar situation here.
- jbkkd 13y agoOff topic: Your blog posts regarding building a search engine are quite fascinating, thanks for them!
- boyter 13y agoThank you. I couldn't find much online like it so wrote it myself. I am hoping a actual expert does the same some day.
- sker 13y agoThat's one awesome project you have right there. I was just thinking of making something like that for F# with data from Codeplex and Github. But I guess your already-existent product blows my non-existent idea out of the water.
- boyter 13y agoThank you. This sort of feedback keeps me going.
- mkuhn 13y agoA startup I know ran a Dropbox like solution with hundreds of servers mainly on Hetzner servers and had no significant issues with disk reliability although it was not a high stress situation.
- no_gravity 13y agoI can confirm this. Im running many servers with all kind of hosters since the 90s and the only one that had a Harddisk go haywire was the one Server I had with Hetzner.
- ch0wn 13y agoWe also had very bad luck with hard drives at Hetzner. Even the replacement disks we got after one gave up had 25000+ hours of spinning time according to SMART.
- X-Istence 13y agoI've had hard drives go bad in a huge variety of different servers in many different locations. The Planet (now Softlayer), Digital Ocean (VPS ...), NAC.net, Atjeu, and others. These days I just expect it to go bad, so I use ZFS within FreeBSD to watch for errors, and when they happen I ask them to replace a single drive, and then the other. So far it has worked very well for me with no data loss. I've been fairly lucky in that both drives have't failed at the same time.