I have a boto3 client :

boto3.client('kms')

But it happens on new machines, They open and close dynamically.

    if endpoint is None:
        if region_name is None:
            # Raise a more specific error message that will give
            # better guidance to the user what needs to happen.
            raise NoRegionError()

Why is this happening? and why only part of the time?

One way or another you must tell boto3 in which region you wish the kms client to be created. This could be done explicitly using the region_name parameter as in:

kms = boto3.client('kms', region_name="us-west-2")

or you can have a default region associated with your profile in your ~/.aws/config file as in:

[default]
region=us-west-2

or you can use an environment variable as in:

export AWS_DEFAULT_REGION=us-west-2

but you do need to tell boto3 which region to use.

os.environ['AWS_DEFAULT_REGION'] = 'your_region_name'

In my case sensitivity mattered.

you can also set environment variables in the script itself, rather than passing region_name parameter

os.environ['AWS_DEFAULT_REGION'] = 'your_region_name'

case sensitivity may matter.

For Python 2 I have found that the boto3 library does not source the region from the ~/.aws/config if the region is defined in a different profile to default.
So you have to define it in the session creation.

session = boto3.Session(
    profile_name="NotDefault",
    region_name="ap-southeast-2"
)

print(session.available_profiles)

client = session.client(
    'ec2'
)

Where my ~/.aws/config file looks like this:

[default]
region=ap-southeast-2

[NotDefault]
region=ap-southeast-2

I do this because I use different profiles for different logins to AWS, Personal and Work.

I believe, by default, boto picks the region which is set in aws cli. You can run command #aws configure and press enter (it shows what creds you have set in aws cli with region)twice to confirm your region.

Alternatively you can run the following (aws cli)

aws configure --profile $PROFILE_NAME

it’ll prompt you for the region.

notice in ~/.aws/config it’s:

[default]
region = ap-southeast-1
output = json

[profile prod]
region = ap-southeast-1
output = json

[profile profile name] in the square brackets

If you are using lambdas, then you would likely want to use the region the lambda is deployed in.
You can use the following

import boto3
import json
import os

def lambda_handler(event, context):
    region = os.environ['AWS_REGION']
    print('Lambda region: ', region)
    kms = boto3.client('kms', region_name=region)

For those using CloudFormation template. You can set AWS_DEFAULT_REGION environment variable using UserData and AWS::Region. For example,

MyInstance1:
    Type: AWS::EC2::Instance                
    Properties:                           
        ImageId: ami-04b9e92b5572fa0d1 #ubuntu
        InstanceType: t2.micro
        UserData: 
            Fn::Base64: !Sub |
                    #!/bin/bash -x

                    echo "export AWS_DEFAULT_REGION=${AWS::Region}" >> /etc/profile

We have the configured regions stored in our ~/.aws/config file. Here is a pure python way of reading the correct region from this file based on the profile name:

def get_aws_region(profile_name: str) -> str:
  config = configparser.ConfigParser()
  config.read(f"{os.environ['HOME']}/.aws/config")
  profile_section = config[f"profile {profile_name}"]
  return profile_section["region"]

regions = [
            'eu-north-1', 'ap-south-1', 'eu-west-3', 'eu-west-2',
            'eu-west-1', 'ap-northeast-3', 'ap-northeast-2'
            'ap-northeast-1', 'sa-east-1', 'ca-central-1', 
            'ap-southeast-2', 'eu-central-1', 'us-east-1', 'us-east-2',
            'us-west-1', 'us-west-2']
for r in regions:
   kms = boto3.client('kms', region_name= r)

If you use AWS Lambda your code will work while you deploy it,
because the Lambda is deployed in a specific region.