HN Explorer

5 comments

[ 3.3 ms ] story [ 21.1 ms ] thread
samwww
good article. the big question really is how (in)dependent azs really are.
willvarfar
The article says amazon say it was caused by ebs replication between zones - one imagines some kind of thundering herd.

Seems to explain it. No mystery here.

fvbock
one would assume thought that that thundering herd would have to go through some kind of queue to avoid resources-starveing/dos-ing yourself...

i hope that the postmortem they promise will shed some light onto the architecture.

vokoda
Insulation means a failure can't spread from one availability zone to another. It doesn't prevent the same type of failure from occurring multiple times across different availability zones.
wladimir
This has interesting parallels with the financial crisis of 2008. People also first assumed several national economies were 'insulated' from each other. This assumption broke down very quickly.

We are very good in underestimating the multitude of ways in which complex systems are connected.