I had a similar experience when worked with AWS Batch jobs executed in Fargate containers which follow the same principles as ECS in scope of assigning roles and permissions.
If you are going to publish messages into specific topic from the code executed inside of your container, then you should create a role with necessary permissions and then use its ARN in the JobRoleArn
property of your job definition.
For example (there can be minor mistakes in the code below, but I am just trying to explain the concept here):
Role cloudformation:
"roleresourceID": {
"Type": "AWS::IAM::Role",
"Properties": {
"AssumeRolePolicyDocument": {
"Statement": [
{
"Action": "sts:AssumeRole",
"Effect": "Allow",
"Principal": {
"AWS": "*"
}
}
],
"Version": "2012-10-17"
},
"RoleName": "your-job-role"
}
}
Policy attached to the role:
"policyresourceid": {
"Type": "AWS::IAM::Policy",
"Properties": {
"PolicyDocument": {
"Statement": [
{
"Action": "sns:Publish",
"Effect": "Allow",
"Resource": "<arn of the specific SNS topic>"
}
],
"Version": "2012-10-17"
},
"PolicyName": "your-job-role-policy",
"Roles": [
{
"Ref": "roleresourceID"
}
]
}
}
And finally attach role to the Job Definition:
....other job definition properties
"JobRoleArn": {
"Fn::GetAtt": [
"roleresourceID",
"Arn"
]
}
Of course you may structure and format roles and policies in way you like, the main idea of this explanation is that you need to attach proper role using JobRoleArn
property of your job definition.