tree 1496c78ccc2bb61b5645ea99221d0d2c00063091
parent ef6015fa27274e0df5f094904c0cfca17a29737f
author Antonio Barone <syntonyze@gmail.com> 1609248289 +0100
committer Antonio Barone <syntonyze@gmail.com> 1609324628 +0000

Separate replica from master

Not having a specific placement strategy for replica might cause it
to be deployed on the same ec2-instance as the master.

If tasks are running on the same instance, both source and destination
IP are the same when being received by the target instance and this
causes packets to be dropped as per convention for L4 networking[1].

Ensure replica is never deployed to the same ec2-instance as the
master by creating separate ASG for master and replica.

Since the cluster is now composed by different autoscaling groups,
rather than one, the DESIRED_CAPACITY variable cannot be used to specify
the size of the cluster as a whole, as each autoscaling group has its
own size.

The master ASG is defined as such:
* minimum:1 - desired:1 - maximum: configurable via the MASTER_MAX_COUNT

The slave ASG does not yet have the ability to scale (it's in the roadmap
to do[2]):
* minimum:1 - desired:1 - maximum:1

[1]https://aws.amazon.com/premiumsupport/knowledge-center/target-connection-fails-load-balancer/
[2]https://bugs.chromium.org/p/gerrit/issues/detail?id=13619

Bug: Issue 13879
Change-Id: I306610e9b5720363f730765d9f9332d0b7f52814