Redundant infrastructure

[Derek Atkins]

Geert Janssens <geert.gnucash at kobaltwit.be> writes:

>> > Of course this doesn't help with the service redundancy.  If there IS a
>> > local issue (hardware, power, network) then the service will go offline
>> > until it can be repaired.  Granted, I have a large-scale UPS and a
>> > natural-gas-powered backup generator so there is no longer a local power
>> > outage issue.  However HW and ISP issues are a bit more out of my
>> > control.
>> 
>> I could provide a mirror site for redundancy with my shared hosting (at
>> Dreamhost), if that would help. Perhaps just a "hot backup" that could be
>> enabled if you did lose connectivity or so far a while, rather than working
>> up a more complicated High Availability system. I assume a brief outage
>> (eg, hour?) of the bug tracker would not be critical to life.
>
> That's very kind. For the record I have two redundant servers myself that can 
> be configured to run as backups/mirrors/whatever of the gnucash 
> infrastructure.
>
> This is something I'd like to pick up some time later, when gnucash 2.7/.28 
> are taking less of my time. Those are top priority now as several distros are 
> starting to drop gnucash due to the webkit obsolescence.

Linas and I have talked about it as well.

The problem, of course, is that it's easy to set up load-balancing
redundancy (with two separate systems in separate locations) both
serving requests..  But this really only works for static (or
quazi-static) content.

I'm not sure how we would accomplish that with the Wiki or even GIT,
where there's a read-write backend.  For something backed by a database
you would need to run multiple instances of MySQL with concurrency
behind them.  I'm not enough of a MySQL guru to set that up.

Similarly, I'm not sure how we would replicate the Mailman instances.
We could certainly set up a redundant MX record, but SMTP already holds
mail for ~5 days so should handle relatively short outages.

The other option is to have a "hot spare" running, which is taking
updates from the primary but isn't serving any content.  But I don't
know how to handle an "automated cutover" in the case of a failure.

All the solutions I know about are for multiple instances in the same
data center.  I have no idea how to do it in a globally distributed
manner.

> Geert

-derek

-- 
       Derek Atkins, SB '93 MIT EE, SM '95 MIT Media Laboratory
       Member, MIT Student Information Processing Board  (SIPB)
       URL: http://web.mit.edu/warlord/    PP-ASEL-IA     N1NWH
       warlord at MIT.EDU                        PGP key available