High availability on remote site

[Frank Leonhardt]

On 15/08/2013 13:18, Mark Felder wrote:
> On Thu, 15 Aug 2013 18:19:35 +0700
> Olivier Nicole <olivier.nicole at cs.ait.ac.th> wrote:
>
>> Hi,
>>
>> I have been assigned to offer HA on a 3 tiers architecture.
>>
>> Data storage tier will be MySQL, so replication is easy.
>>
> Keep in mind that MySQL replication has plenty of its own issues. It
> does not replicate every SQL command to the slave. Guaranteeing that
> data on both servers is identical is also a very tricky process. You
> might want to first browse through the sections here to get an idea:
>
> http://dev.mysql.com/doc/refman/5.5/en/replication-features.html
>
>   
>> HA should be implemented only on the Data storage tier, Active/Active,
>> but one of the sites is remote!
>>
>> When everything is working, each application accesses the local MySQL
>> tier, but when the local MySQL becomes unavailable, it should be able
>> to automatically move to the other database server.
>>
>> I have no access to the application, so I cannot modify it to test if
>> local MySQL is working. So I should have an HA mechanism that enforces
>> changing the IP address on the database server.
>>
> This is easy. Use HAProxy. It can test to see if your local MySQL
> instance is up and running and if it detects it is not it will
> automatically pass connections to the remote site's MySQL server.
>   
>> If both servers are installed at different places, with different
>> addresses, would there be a way beside establishing an IP tunnel/VPN
>> between both places to have all machines in a single subnet?
>>
> This seems unnecessary. Why do you need them to be on the same subnet?
>
>> An image is here http://www.cs.ait.ac.th/~on/HA.gif
>>
>> I am really bothered by the IP tunnel, but that's the only way I see to keep HA.
>>
> Hopefully I've answered this question for you and you see that you
> shouldn't need these to be on the same subnet.
> _______________________________________________
>

WHS, especially regarding the built-in replication of a mySQL database 
being problematic. I tried this a few years ago and decided it wasn't 
worth the candle (for my needs). It came down to the application 
software needing to be sensitive to the situation - to understand it 
needed to use a backup server, and to treat it as read-only. The 
implication is that mySQL could be some kind of distributed cluster 
until you got to it in detail. Or perhaps I was missing a point 
somewhere. If you get a "perfect" cluster going please do tell me know how.

Incidentally, in the end I just used rsync - much less fuss but only 
good as a backup, really (which is what I really wanted).

Regards, Frank.