[AWS] Setting AWS credentials to another account in the env var through /etc/profile.d will cause the cluster in INIT mode

[Michaelvll]

A user encountered this issue, where they set the AWS credentials through AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY using the /etc/profile.d, and their sky status -r will always show the cluster in INIT mode.

To reproduce:

1. sky launch -c test-cred --cloud aws --cpus 4
2. ssh test-cred; sudo echo 'export AWS_ACCESS_KEY_ID=xxx; export AWS_SECRET_ACCESS_KEY=xxx' > /etc/profile.d/aws_keys.sh; sudo chmod +x /etc/profile.d/aws_keys.sh; ray stop
3. sky stop -y test-cred; sky launch -y -c test-cred; sky status -r

After step 2, the ray autoscaler started with sky launch will use the wrong AWS credential, causing the ray status to show empty Healthy nodes: (reference: https://github.com/ray-project/ray/blob/d2fc4823126927b2c54f89ec72fa3d24b442e6a3/python/ray/autoscaler/_private/autoscaler.py#L396)

> RAY_ADDRESS=127.0.0.1:6380 ray status
======== Autoscaler status: 2023-08-21 23:10:19.630174 ========
Node status
---------------------------------------------------------------
Healthy:

Pending:
 (no pending nodes)
Recent failures:
 (no failures)

Resources
---------------------------------------------------------------
Usage:
 0.0/4.0 CPU
 0B/8.83GiB memory
 0B/4.41GiB object_store_memory

Demands:
 (no resource demands)

That causes sky status -r's check with ray status fail to fetch the cluster status.