Time to test a new hosting company .This time I opted for German Hetzner, and their 4.80€ VPS CX1. Note that this price is VAT inclusive, so it is one of the very cheapest on the market (many providers present VAT 0% prices) for a consumer.
1 snapshot included in the price (paid extra option at OVH and many others)
DDoS protection and the usual goodies
I was going to benchmark this and move one instance there right away, but the always flexible Germans do not open new accounts during weekends. They validate the massive 4.80€ transaction manually, before issuing the virtual server. And yes, this is 2016. Also their support seems to be available only during extended office hours. For the record servers never break during office hours.
I’ll get back to this topic some time later when they have finally booted my server…Not sure when I have the time to do it, as I would have had some today, but it’s Saturday. It would be nice to read any comments while waiting.
I was client of hetzner during 6/8 months , dedicated servers and was ok , untill “someone” was flooding my server (wtf??) . And hetzner blocked my servers for 72h . I am talking about …4/5 years ago . Right now i dont know how it is …
Right now i am testing a ovh vps ssd 2 for my discourse instance , all things are ok , fast install , easy install but the dns settings…no comment , are very dificult to do .
Well, it took them about 2 office hours to deliver my servers, so they are quite efficient when they are onsite.
CPU model: Intel(R) Xeon(R) CPU E5-26xx (Sandy Bridge)
Number of cores: 1
CPU frequency: 1999.999 MHz
Total amount of RAM: 992 MB
Total amount of swap: MB
System uptime: 2:52,
I/O speed: 594 MB/s
Bzip 25MB: 8.31s
Download 100MB file: 69.3MB/s
Good I/O speed in the benchmark, but the Bzip result is roughly the same as with Digital Ocean. The reported clock speed is 200MHz / 10% faster. One more thing that I noticed is that ping times are somewhat slow (~100ms category) from my location. Usually I ping to central Europe in 40-60ms.
Found a one hour slot to do the migration. It was a pleasant experience, as there was nothing needed outside usual Discourse installation process. No funky business with boot scripts or blocked ports, as with Scaleway. Site feels good - perhaps a bit snappier and more responsive than with ScaleWay.
So over all this seems like a decent budget offer. Now I shall start the long term monitoring and let’s see how reliable they are.
Thanks to your posts here, @ljpp, I’ve just done this myself (pulling my data from Scaleway). As with you, once Hetzner actually provisioned the server (almost three hours) the migration went without a hitch.
On the east coast of the U.S., I’m seeing load times about 20% – 25% faster than Scaleway’s. (Of course, it costs about 30% more, too!)
One month is too short period of time to make any conclusions, but so far no issues what so ever. Did the Discourse 1.7 upgrade on both of my boxes there - these definitely rebuild a lot faster than ScaleWay’s.
I’ve been a Hetzner customer for some years, my D. instance on their small-mid VPS (2GB RAM) is running fast and responsive. I’m based in Europe, BTW, so their datacenter location in Germany is a plus, for latency and for legal reasons.
Recently I discussed here an issue that my site is not reachable via http. After some investigation, this was not a Hetzner issue. Domain auto-renewal had failed and it was suspended. I deleted those posts as they had disinformation regarding the root cause.
Still a valuable experiment to see if Hetzner handles support tickets outside office hours, and they do not. Also, I dont seem to be update or close my own support ticket, which is odd. The server and connection works though.
Hetzner’s CX servers include at least one Snapshopt for free, which is great, and it just takes a few seconds to create one. But they issue this warning:
Power off your vServer before taking a snapshot to ensure data consistency. If you take a snapshot of a running vServer, data in the memory cannot be saved. To restore a snapshot, the vServer must be switched off.
Can someone translate what this means for discourse (and, Wordpress, for that matter)? I understand that if I take a Snapshot only includes whatever is saved to the disk, but does it matter that I don’t get what was in RAM (and swap?)? Or when does it matter?
BTW: have you seen the revamped Hetzner-website? Was about time Now, I hope, the next redesign will be the beautiful backend…
I’ll post a screenshot here, so we can compare before and after
You’re probably right about not needing the RAM/swap data. All the data you need for a restore should be on the filesystem, with the exception of any caching going on with the database.
The problem with taking a snapshot of a “live” system is that the data can change while the backup is happening. For example, if the backup gets halfway through the database, and then something changes, the backup will carry on regardless. But when you go to restore, the database file will probably be corrupt.
I think other cloud providers will have exactly the same warning - Digital Ocean certainly does.
The way Discourse deals with this problem for its own backups is to put the database in read-only mode, and then use the database’s built-in backup methods.
The only real way to put an entire operating system in “read only mode” is to shut it down, hence their recommendation.
Okay, so if I take backup in discourse and then a snapshot, I should be fine?
By “fine” I mean, I if I then fiddle with some server settings and the thing becomes unresponsive, I can restore it from the snapshot and then restore discourse from the backup and everything will be as it was.
(For wordpress, I assume the same applies: do an internal backup in addition to the snapshot)
The safest thing would be to have a completely external system storing your backups as well - like S3/Dropbox/Google-Drive etc. - that way you’re not entirely reliant on a single company and their computer systems.
Then I’d say you’re pretty safe - in a worst case scenario where the hosting provider loses your server and its snapshot, you’ve still got your data safe
Since Digital Ocean decided to burn my hard earned credits, I moved my little sandbox Discourse instance to Hetzner.de (CX11). Looks like Moore’s Law still applies, as the CPU performance has improved over the years
CPU model: Intel Xeon Processor (Skylake, IBRS)
Number of cores: 1
CPU frequency: 2294.608 MHz
Total amount of RAM: 1939 MB
Total amount of swap: MB
System uptime: 26 min,
I/O speed: 969 MB/s
Bzip 25MB: 3.39s
Download 100MB file: 128MB/s
There was one new issue that I discovered – the discourse-setup failed on error message which suggested that port 443 would be unreachable. This was on a clean default configuration, with no firewall rules defined. I did not find an obvious reason why this occurs, going though the relevant topics here, disabled CloudFlare caching, but the issue persisted.
Then I just configured the app.yml manually and it worked perfectely, with Let’s Encrypt, https:// and everything.
.This time I opted for German Hetzner, and their 4.80€ VPS CX1. Note that this price is VAT inclusive, so it is one of the very cheapest on the market (many providers present VAT 0% prices) for a consumer.
Now, I hope, the next redesign will be the beautiful backend…