3 ms·

AWS _had_ architected away from single-region failure modes. There are only a few services that are us-east-1 only in AWS (IAM and Route53, mostly), and even th

AWS _had_ architected away from single-region failure modes. There are only a few services that are us-east-1 only in AWS (IAM and Route53, mostly), and even they are designed with static stability so that their control plane failure doesn't take down systems. It's the rest of the world that has not. For a long time companies just ran everything in us-east-1 (e.g. Heroku), without even having an option to switch to another region.

MichaelZuo 1y ago
The parent seems to be implying there is something in us-east-1 that could take down all the various regions?
JCM9 1y ago
So the control plane for DNS and the identity management system are tied to us-east-1 and we’re supposed to think that’s OK? Those seem like exactly the sorts of things that should NOT be reliant on only one region.
- cyberax 1y ago
  It's worse than that. The entire DNS ultimately depends on literally one box with the signing key for the root zone. You eventually get services that need to be global. IAM and DNS are such examples, they have to have a global endpoint because they apply to the global entities. AWS users are not regionalized, an AWS user can use the same key/role to access resources in multiple regions.
  - harshaw 1y ago
    not quite true - there are some regions that have a different set of AWS users / credentials. I can't remember what this is called off the top of my head.
    - cyberax 1y ago
      These are different AWS partitions. They are completely separate from each other, requiring separate accounts and credentials. There's one for China, one for the AWS government cloud, and there are also various private clouds (like the one hosting the CIA data). You can check their list in the JSON metadata that is used to build the AWS clients (e.g. https://github.com/aws/aws-sdk-go-v2/blob/1a7301b01cbf7e74e4ceed0041d31a68d5fea7dc/service/ec2/internal/endpoints/endpoints.go#L98 https://github.com/aws/aws-sdk-go-v2/blob/1a7301b01cbf7e74e4... ).

MichaelZuo 1y ago

The parent seems to be implying there is something in us-east-1 that could take down all the various regions?

JCM9 1y ago

So the control plane for DNS and the identity management system are tied to us-east-1 and we’re supposed to think that’s OK? Those seem like exactly the sorts of things that should NOT be reliant on only one region.

cyberax 1y ago
It's worse than that. The entire DNS ultimately depends on literally one box with the signing key for the root zone. You eventually get services that need to be global. IAM and DNS are such examples, they have to have a global endpoint because they apply to the global entities. AWS users are not regionalized, an AWS user can use the same key/role to access resources in multiple regions.
- harshaw 1y ago
  not quite true - there are some regions that have a different set of AWS users / credentials. I can't remember what this is called off the top of my head.
  - cyberax 1y ago
    These are different AWS partitions. They are completely separate from each other, requiring separate accounts and credentials. There's one for China, one for the AWS government cloud, and there are also various private clouds (like the one hosting the CIA data). You can check their list in the JSON metadata that is used to build the AWS clients (e.g. https://github.com/aws/aws-sdk-go-v2/blob/1a7301b01cbf7e74e4ceed0041d31a68d5fea7dc/service/ec2/internal/endpoints/endpoints.go#L98 https://github.com/aws/aws-sdk-go-v2/blob/1a7301b01cbf7e74e4... ).