ping @samuelkarp

Ignoring the flaky test on s390x

@thaJeztah Thanks! I will open the follow-up PR in a few days.

This is great to see! What is the process for getting this into docker-ce? How long does it usually take?

a first release-candidate for Docker CE 18.06 has already been cut, so likely in the version after that; I'll discuss if this is important enough to backport to that version.

without updating the vendored revision of the AWS SDK

@haikuoliu @samuelkarp I missed this earlier; does the above mean that we wouldn't need this change if we'd update the version of AWS SDK? Is there a reason to not update?

@thaJeztah Unless I'm misunderstanding what he meant by that, I believe he's referring to the fact that these isolated regions have their own aws sdk with region specific endpoints (e.g. amazonaws.com.cn), and using a region specific version is a typical requirement for users of those various regions.

This update seems to allow you to specify the endpoint for your isolated region w/o an sdk change (which most importantly isn't viable for end-users since the sdk is baked into the docker binary). The alternative would be to create a docker-ce for each of these isolated regions that has their own aws sdk revision.

Hm, right; I was wondering if a newer version of the SDK had better detection (e.g., if region starts with cn-, use amazonaws.com.cn instead of amazonaws.com)

Building different versions for each region doesn't sound like a feasible option (what if, say Google or Azure uses different regions with their own exceptions)?

@thaJeztah Yes, we can just update AWS SDK in Docker. But the problem is ECS cannot control Docker update, so ECS will probably be blocked by Docker update when there is a new region. ECS Windows uses Docker EE, we don't know how long does it take to include the SDK update in Docker EE.

@thaJeztah AWS regions are divided into different partitions (e.g,. for standard regions, the partition is "aws" while for regions in China the partition is "aws-cn"). Partitions tend to have different endpoint suffixes (e.g., amazonaws.com for aws, amazonaws.com.cn for aws-cn). The AWS SDK contains logic that guesses the endpoint when it encounters a region it doesn't otherwise know about, and that logic defaults to guessing an endpoint in the aws partition.

When the cn-northwest-1 region was launched, older versions of Docker (I believe anything prior to 18.03 or so) incorrectly guess the CloudWatch Logs endpoint as "logs.cn-northwest-1.amazonaws.com" instead of "logs.cn-northwest-1.amazonaws.com.cn", making the awslogs driver unusable in that region. This change seeks to provide a way to override the endpoint if the SDK guesses incorrectly.

@thaJeztah

Follow up PR to docker.github.io: docker/docs#7003

@thaJeztah This missed docker-ce 18.06, correct? Not seeing it in the release notes.

I see the 18.09.0-ce-beta1 contains this fix. One qustion:

• Is this something that can be defined in docker-compose?

FYI this works for me using docker CLI w/ 18.09.0-ce-beta1 when specifying awslogs-endpoint log-opt. Have not tested w/ compose yet.