Isn’t the point of using Amazon EC2 exactly that you get guarantee
that your virtual machine will not be affected by normal problems
related to hosting it on a single machine. I mean they say that they
have a copy of your virtual machine on more than on server on maybe
more than one location. I may be wrong here but from what I have read,
you should not worry about your instance shutting down or slowing down
drastically because what happens if I only have one instance and I
reply on that?
Instances on EC2 can and sometimes do go down. I have not had it happen
to me yet. However if you look at the forums there you will see
occasional posts like “i’m having network problems” or “i’m having slow
response” etc and sometimes the reply from Amazon is that something was
not right on the physical box the instance was running on and they have
rebooted that box and they should now be fine.
To make our solution as resilient as possible we will run nginx on
multiple instances in multiple availability zones, and these will
connect to other multiple upstream instances in multiple availability
zones We are expecting that from time to time one or more instances
will go down or suffer from performance problems not related to our
app/coding but rather the ec2 infrastructure. After all technically ec2
is still considered beta.
So with this in mind we have very big needs for monitoring upstream
servers closely (and hopefully Francois is on track for this with his
great ideas about enhancing nginx moniroting) and things like ‘fair’
which take into account the response times of the upstreams would be
VERY helpful. This way if an upstream is suffering (but not meeting the
failure requirements of nginx to take the upstream out of rotation) than
‘fair’ can at least keep those requests to a minimum. With the right
enhancements to ‘fair’ that is.
Be a better friend, newshound, and
know-it-all with Yahoo! Mobile. Try it now.