Graceful Rolling Push? about asgard HOT 2 CLOSED

Thanks for the suggestion, phstc. What I would recommend instead is never to run any application with only a single instance if you require high availability, especially if you want to perform a rolling push. The rolling push in Asgard is modeled after a conventional data center cluster of machines, where you need at least two instances in order to guarantee uptime during the disruptive upgrade of each machine in the cluster.

What you're suggesting can be performed simply by increasing your ASG size to 2 and waiting for both instances to take traffic before you do a rolling push. Keeping the size at 2 is a good practice in the cloud if availability is important to you. Cloud instances have a tendency to fail randomly when least convenient for you. Better to be prepared.

Alternately, you could perform a red/black push using Asgard's Cluster screen, which would create a new ASG with a new instance before you delete the old ASG. The red/black push model is what we generally use at Netflix, because it's faster than a rolling push (for ASGs bigger than 1 instance), it guarantees homogeneous ASGs even if errors occur during deployment, and it makes it possible to roll back a bad deployment quickly by switching traffic instead of waiting for replacement instances to launch.

from asgard.

Hi @joesondow

I would like to perform the instances replacement when I deploy a new version of my app. When I start a new instance based on my app image, it automatically updates the app. It can occurs many times per day.

The Rolling Push in the ASG doesn't fit well for us, even with 2 instances, because it sends the terminate signal to the instance which still in the ELB, before the removal from ELB. Users accessing this instance (ELB enabled sticky session) may get errors until the ELB updates its instances.

Alternately, you could perform a red/black push using Asgard's Cluster screen

You mean that you have two ASG, one enabled ASG1 and another disabled ASG2, when you perform a deploy you enable ASG2, when the instances are up and running you disable ASG1? Next time the opposite ASG2 then ASG1?

Or always when you deploy, you create a new ASG and disable the old one?

How do you keep tracking of your instances usage (CPU, Memory, Disks etc), since the instances terminate a lot? We use New Relic and CloudWatch, do you recommend other tools?

from asgard.