Or because the cloud has offered such a good value proposition for cheap and easy scaling with demand.
And anyway, it's unfair to rail on people who "should have had a redundancy plan" when the service they pay money for is one with a redundancy service included in it (availability zones) which has unexpectedly also failed.
Our point stands, for engineers to consider all likely scenarios when building redundancy and not assume anyone – even Amazon – can provide 100.0% uptime.
Your point appears to be "If you don't have 100% uptime then it's all your fault and you should have planned for it you lazy idiot, everyone should blame you. Also you can never have 100% uptime so people should stop blaming Amazon.". Do you have more of a point than that?
At what point did they make the call that the outage was too serious and they would lose all data since the last backup and start migrating? Had they pre-planned for it, or was it ad-hoc? Will they stay where they are now and use Amazon US East as their failover or migrate back in due course? Or rearchitect to handle this in future?
(thanks, re: name).
Is it just luck that the problem which happened was one they prepared for instead of one they didn't?
AZs - according to Amazon, "are distinct locations that are engineered to be insulated from failures in other Availability Zones."
that did not really seem to work.
if you're deployed in one zone and shit hit's the fan: "your fault". if you assume amazon does as advertised and live in several AZ and these go down apparently more or less at the same time: 'amazons fault'...
i read that amazon plans to post a 'postmortem' on this... i'd be really eager to know how AZs are actually designed/sperated. not to be able to point fingers (maybe just a little bit), but to just _know_ where i am deploying stuff to...
The genesis of the article was the press implying that to use the cloud your only choice was to trust AWS provided 100% up-time, and this is a position we disagree with.
Shit is going to happen with your host. To say it's a problem with the cloud is unfair.
In this case, it looks like most users of AWS are in need of more geographic redundancy, but in terms of localized data redundancy (term?), it appears AWS is a pretty solid solution.
I was under the impression if one of the clusters were to be unavailable, the nearest mirror would resume responsibility. This should include distribution services as well.
