fuel-dev team mailing list archive

[Dmitry Borodaenko <dborodaenko@xxxxxxxxxxxx>]

On Fri, Mar 7, 2014 at 4:03 AM, Sergey Vasilenko
<svasilenko@xxxxxxxxxxxx> wrote:
> I do not sure, that it's a good idea.
> Usualy, most of anything new things, we developing and testing under simple
> configuration. And after it scale to HA configurations.

I think it is actually a very BAD idea to develop using a
configuration that is significantly different from production.

Every time you increase the time interval between introducing a bug
(i.e. developing) and finding a bug (i.e. testing), the cost of fixing
the bug increases exponentially. You no longer remember what you've
changed, you piled other changes on top of incorrect code, you
impacted other engineers who encountered your bug and now have to
figure out that it wasn't their changes causing problems, and so on.

> Simple configuration gives us low time of deploy,

Using HA configuration will make us finally pay some attention to the
time it takes to deploy HA and fix it. It's not a fundamental problem,
we're actually doing something wrong here and we should figure it out.

> possibility of don't use
> buggy Galera, songle-node AMQP. Works with "simple"
> configuration we can don't distractions to HA ussues.

These are not distractions, you will encounter all these issues before
you can release. And it will be much easier to fix them immediately
after they are introduced, not 1 week before code freeze.

> One of most typical
> examples -- migration to the next openstack version.

It is even more important for Icehouse. If we encounter Icehouse bugs
that break HA before Icehouse is released (5 weeks from now), we might
get them fixed upstream instead of having to carry our own patch
series after the release.

-- 
Dmitry Borodaenko