Server Fault is a question and answer site for system and network administrators. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I run a fairly heavy trafficked website and due to some unfortunate incidents the machines that are in my cloud at Linode went down. And I have only a single Load Balancer machine exposed to the outside world (one IP).

Also my site is a candidate for over 6,000 static pages that can be mirrored. Now my DNS is CloudFlare.

What can I do to maintain a static mirror of my site and route to it incase my site goes down.
Since I am running from Linode I don’t have anything like Route53 to detect downtime at an IP address and point to another one.

What are the strategies that people use statically mirror site and work against downtime?

share|improve this question

A couple different things jump to mind first and foremost:

First, you already have a static mirror of your site that is designed for just this use-case: Cloudflare. As well as providing your DNS, I assume you have them set up as a CDN to dull the brunt of traffic that is coming towards you. Cloudflare has a feature called Always Online which is intended to do just what you are looking for: provide a static copy of the website up even if the "origin" - in this case, your load balancer and/or the servers behind it - go down. Make sure that you have that set up properly first, before you worry about a more complicated solution. Always good to get 80%+ of the problem out of the way with 2% of the work! In fact, you may be able to simply rely on Cloudflare to take care of the problem for you completely. Do some reading into Cloudflare Always On first, as it is going to be a lot simpler for you implement than anything to follow, since Cloudflare is already set up in your infrastructure. If after you have read up, it won't do enough for you, read on.

Now, there are a couple different things you need to think about when worrying about making your site available through different kinds of outages. The first thing is the goals. Are you trying to maintain your sites availability in an outage only, or would you prefer to also maintain a second site that you use to load balance between locations? What kind of system outages are you trying to prevent? How much time and/or money are you willing to invest in minimizing downtime?

After you have set some of the goals, you can now look at the different kinds of solutions that are out there. Generally speaking, all of the different strategies for minimizing downtime involve keeping one or more "extra" locations synchronized with the content in the main location, preferably in a different hosting provider and network to prevent downtime that spreads across an entire company. Failover is generally done through manipulation of DNS records. Larger companies will sometimes use IP-level solutions like anycasting or route manipulation to accomplish the tasks - which comes with several benefits - but doing so is expensive and extremely difficult to get right.

There are lots of companies out there that can help you change your DNS records automatically when a single IP becomes unavailable, but you can do it fairly easily yourself by utilizing the Cloudflare API (or the API of whatever your DNS provider is, if you change in the future.) All that is required is a second system in a separate location as wherever your website is hosted that continually checks your site to make sure it is up. If not, it hits your DNS providers API and changes the DNS records of your site to point to your backup location. This means that you will have a worst-case (on paper) downtime of the monitoring interval + the DNS TTL. In practice, DNS can be quite aggressively cached, and even short (<30sec) TTLs can take up to a couple hours to be completely cleaned out by all clients around the world. Mobile devices, in particular, are known for being troublesome with this. There are lots of tutorials out there on how to use different monitoring systems to accomplish this task - a quick search for "cloudflare failover" got me these two which use nagios and monit, respectively, but I am sure there are lots more easily accessible.

Of course, any kind of failover requires a place to fail over to! There are a bunch of different requirements for doing so, depending on your particular application's specifications and the requirements for synchronization. Some sites which are all static content can simply be copied each time they are updated to both locations, either by hand, by an automated script which pushes or pulls from the master to the slaves (cron + rsync is your friend!), or other methods like block-replication (DRBD) or a shared file system (GlusterFS). Other sites with dynamic content will require both this kind of file-level synchronization and database replication in a master-slave setup. Note, databases can cause all sorts of problems if you are trying to accept writes at both locations, so do plenty of research into master/master database replication using your particular database technologies if you are planning to have both datacenters concurrently available. It is not uncommon to setup the slave as a read-only replica even when failed over to in order to keep from having to sync the data back from the promoted slave when the main center is available again.

There are a lot of different things to think about when considering this kind of high-availability setup. If you tell us a bit more about the specifics of your application, I'm sure we can add more specific advice.

share|improve this answer
Thanks for this. Just to note Always Online from CloudFlare is not a guaranteed services. They cache heavily used pages. Not all pages. I was thinking more like a s3 hosted static version that can be switched. But the unique proposition I have since CF is the DNS and reverse proxy provider is to switch proxing IP in the event a Datacenter goes down. This also means that CloudFlare does not go down at all. I didn’t know CloudFlare had an API to manipulate this. Thanks for that! My intention is not to replicate DB. That’s too much for me. Just a Jekyll or S3 hosted solutions – Quintin Par Jan 26 '13 at 8:28
Then I would look into the kind of failover mechanisms I mentioned, and mirror your website with a cronjob and rsync, or by hosting all the content on a distributed filesystem like GlusterFS. If you are not replicating the DB, make sure the pages you are copying won't include any dependencies on it! – Scrivener Jan 26 '13 at 17:41

Instead of buying the NodeBalancer fits-all ready-to-use feature from Linode, you might as well just buy another regular Linode, and implement load balancing and caching by yourself.

You can use nginx, and have it act as a proxy and balancer to your real web-site.

Depending on whether you require your website to change every couple of hours/days or not, you can use several nginx features to save the content from your upstream Linodes.

One feature you may find very useful is proxy_cache. Another one is proxy_store.

The proxy_cache set of directives are very flexible such that nginx can be configured to automatically store and expire all the pages, or automatically serve stale pages only the the upstreams are unavailable (e.g. take a look at proxy_cache_use_stale). Whereas proxy_store can potentially be implemented together with a manual clear-all rm -rf script, depending on your needs.

Of course, if you're already paying 20$/mo for your load balancer at Linode (and provided you are not over budget), then you might as well cancel that, and look into CloudFlare, Incapsula and other similar offerings, some paid versions of which can be configured to cache all kinds of content (including that dynamically generated, starting at 10$/mo at Incapsula, for example).

share|improve this answer

Best Speed, Best Reliability, Most Secure

If your site is static then you should consider hosting it purely on CDN, then you don't need load balancers , dedicated or vps servers. Good CDNS scale as much you need and your only charged for the send amount or per Xrequests depending which company you go with. In regards of the non www cname issues I believe that cloudsflare has a workaround, Rackspace and Amazon you'd need a vps to redirect from non www to www. on the cdn.

CDN would offer more performance than a vast number dedicated servers or vps. Also its by far the most reliable and secure.

If you have some php files like contact forms then you can host them on a vps 64mb-256mb by using ajax js.

Further you mention mirroring,cdns mirror files all over the world this one of the main reasons there so fast and they use fail proof raids, and other failover redundantances... cdns cant really fail. But if you wanted a mirror say to a vps then you just use the api and cron the backup.

DNS.. Just pickone that has anycast/active failover

share|improve this answer

Cloudflare can already do this for you. Though a service called Always Online™ (scroll down about half way).

If you want your own solution, use a proxy/load balancer. Something like haproxy works very well for this. Though, as this is a static file request, nginx will work very well too. So, if one goes down, the proxy simply stops requesting from the down server. But note that using a load balancer alone is not a failover, you need additional web servers ready to serve the extra load when the other fails.

DNS failover which you mentioned is not recommended if you are looking to stay within one datacenter. See Why is DNS failover not recommended? That post will also outline some additional solutions for you as well.

share|improve this answer
That article refers to using DNS failover /internally/ inside a datacenter which, I agree, is completely idiotic. He already has a load balancer, however, and the outage he mentioned affected not just the load balancer but also the servers within the same datacenter. For multi-center failover, your only choices are DNS-level or IP level, and IP level is far too complex for his needs. – Scrivener Jan 25 '13 at 5:24
@Scrivener I don't see where OP indicated that the load balancer is also down, albeit "machines" is quite vague. As such, I'm assuming sticking to one datacenter is still within his expected scope. But I have added scope and specifics to my answer to prevent confusion. – Grumpy Jan 25 '13 at 20:30
Always Online is not a guaranteed mirror service. A lot of people make this mistake. Always Online attempts to cache the most used page of that website. Not All pages. There is a significant difference. – Quintin Par Jan 26 '13 at 6:12

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.