High Availability of a web server on a distributed cloud

[Fernando Duran]

>Yep, they've had 3 outages in about a year.
>
>
>>Note that with a load balancer (nginx/wahtever) you don't need to reassign IPs since it's not a "master/slave" or stand-by solution; all servers are online with their own IP/DNS name and the LB distributes the load and if one goes down the node gets out of rotation (no traffic goes to it). It's much much better imho to not have different roles (master/slave), changing IPs, split brain and timeout/response time problems and just have the same single configuration in the same one server image that can be spun in a minute. Of course now the LB is a single point of failure.
>>
>>
>>
>
>
>Yes, it is the LB that is the problem, how do you deal with that ?

With more of everything of course!  https://twitter.com/DEVOPS_BORAT/status/274366602252804096

A bit more seriously, it depends on how much money/time you want to spend and what's the target maximum downtime you can afford. For truly no-SPF note that you'd also need a duplicate of everything in a different data centre in a very different location or possibly provider, so it depends where it makes sense to stop.

For the LB an imperfect solution is to have 2 or more with different DNS entries, or go with the master/slave option I criticized a minute ago. Note that a LB is a single simple app so unlike regular servers it's going to be much more reliable and won't have out-of-memory problems, high disk I/O issues, buggy software problems etc, basically you are risking a hard hardware failure (power supply or disk typically) which happens "once every few years" (pulled that out of the air). For a web site with two nodes, this can be acceptable. Or you can close your eyes and trust the "high availability" of the LB provided by AWS/Rackspace.

Cheers,
---------------------
Fernando Duran
http://www.fduran.com



>
>
>
>
>
>
>Dave 
>Cheers,
>> 
>>---------------------
>>Fernando Duran
>>http://www.fduran.com
>>
>>>
>>>Dave
>>>
>>>
>>>Dave Cramer
>>>
>>>
>>>On Mon, Jan 14, 2013 at 3:26 PM, Fernando Duran <liberosec-FFYn/CNdgSA at public.gmane.org> wrote:
>>>
>>>
>>>>>________________________________
>>>>> From: Dave Cramer <davecramer-Re5JQEeQqe8AvxtiuMwx3w at public.gmane.org>
>>>>>To: tlug-lxSQFCZeNF4 at public.gmane.org
>>>>>Sent: Monday, January 14, 2013 2:58:55 PM
>>>>>Subject: Re: [TLUG]: High Availability of a web server on a distributed cloud
>>>>
>>>>>
>>>>>
>>>>>Well I don't want to use a single provider such as amazon. Which makes things more difficult.
>>>>
>>>>More difficult how? if it's because you're more dependent of a particular provider/solution I understand. 
>>>>
>>>>
>>>>Cheers,
>>>>
>>>>
>>>>---------------------
>>>>Fernando Duran
>>>>http://www.fduran.com
>>>>
>>>>
>>>>
>>>>>
>>>>>
>>>>>So the problem is essentially how to duplicate their elastic ip's
>>>>>
>>>>>
>>>>>
>>>>>Dave Cramer
>>>>>
>>>>>
>>>>>On Mon, Jan 14, 2013 at 2:51 PM, Fernando Duran <liberosec-FFYn/CNdgSA at public.gmane.org> wrote:
>>>>>
>>>>>If the application is HTTP a relatively simple solution is to use nginx as a load balancer http://blog.jsdelivr.com/2013/01/nginx-load-balancing-basics.html plus some hearbeat/monitoring to detect/alert when a node goes down.
>>>>>>
>>>>>>High availability is an excellent area to learn about networking/scripting/trade-off decisions etc but it's very hard to get right, esp. the first time. I think is very worth it using the cloud provider's solutions, like Amazon's (or Rackspace's) load balancers and autoscaling.
>>>>>>
>>>>>>---------------------
>>>>>>Fernando Duran
>>>>>>http://www.fduran.com
>>>>>>
>>>>>>
>>>>>>>________________________________
>>>>>>> From: Dave Cramer <davecramer-Re5JQEeQqe8AvxtiuMwx3w at public.gmane.org>
>>>>>>>To: tlug at ss.org
>>>>>>>Sent: Monday, January 14, 2013 8:57:39 AM
>>>>>>>Subject: [TLUG]: High Availability of a web server on a distributed cloud
>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>It is relatively simple to distribute an application to a number of servers and use haproxy to switch ip's. What I can't figure out is how to switch make sure that the IP that points to ha-proxy can be moved easily if that machine fails ?
>>>>>>>
>>>>>>>
>>>>>>>DNS round robin doesn't exactly work. I have a very limited understanding of BGP is it possible to do without BGP ?
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>Dave Cramer
>>>>>>>
>>>>>>> 
>>>>>>
>>>>>>--
>>>>>>The Toronto Linux Users Group.      Meetings: http://gtalug.org/
>>>>>>TLUG requests: Linux topics, No HTML, wrap text below 80 columns
>>>>>>How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists
>>>>>>
>>>>>
>>>>>
>>>>> 
>>>>--
>>>>The Toronto Linux Users Group.      Meetings: http://gtalug.org/
>>>>TLUG requests: Linux topics, No HTML, wrap text below 80 columns
>>>>How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists
>>>>
>>>
>>>
>>>
>>--
>>The Toronto Linux Users Group.      Meetings: http://gtalug.org/
>>TLUG requests: Linux topics, No HTML, wrap text below 80 columns
>>How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists
>>
>
>
> 
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists