Can't start Sagemaker Notebook Instance with a CodeCommit repo - amazon-web-services

I linked a repo in CodeCommit to Sagemaker. However when I try to start an instance with that repo it fails and I get a message:
fatal: unable to access 'https://git-codecommit.us-east-1.amazonaws.com/v1/repos/MyRepo/': The requested URL returned error: 403
I think maybe it has something to do with the IAM role. Is there some policy I should add to the AmazonSageMaker-ExecutionRole. I am completely new to this so please excuse any incorrect usage of terms here.

This is presumably caused by your IAM role ; if you used the UI to create a role, I think by default, it creates a role requiring resources to have "sagemaker" in their ARN (Codecommit repos, S3 buckets...). So you have 2 options:
Try renaming your repo eg "myrepo-sagemaker"
Or edit (ask your admin if you don't have the ability) your SM IAM role to be allowed to access MyRepo

Related

AWS cross account access for code commit in build AWS code job source

I have two AWS accounts A, B.
All my code commit repositories are present in account A.
Now I want to create the AWS code Build job in account B for repositories in account A.
I am trying to figure out to get the list of AWS repositories in account B from account A while selecting the source for creating a code build job.
I am not sure how to get the list of repositories from account A to account B in the source Repository field.
I have followed the below tutorial only till the second topic.
https://docs.aws.amazon.com/codecommit/latest/userguide/cross-account.html
Any help will be appreciated.
You can configure access to CodeCommit repositories for IAM users and groups in another AWS account. This is often referred to as cross-account access.
Mainly you be need to do the following:
Will need to create a policy and role for the repository with the needed permissions.
Create a policy and attach to your CodeBuild Role allowing the access on the Resource for the created Role
eg.
"Resource": "arn:aws:iam::REPO_ACCOUNT:role/MyCrossAccountRepositoryContributorRole"
This will enable the CodeBuild to access the needed CodeCommit repository.
This page explain this very well: Configure cross-account access to an AWS CodeCommit repository using roles.
Also, check this blog post that explain a little more detailed what you want:
AWS CodePipeline with a Cross-Account CodeCommit Repository.

Why is iam:PassRole required in this project?

I cloned this solution azure-devops-on-aws and used dotnet lambda deploy-serverless ... to deploy the MyLizardApp to my personal AWS account.
During the learning curve, I created an S3 bucket my-lizard-test, IAM user group MyLizardGroup with user lizard-user and group policy MyLizardApp-Policy. Included in the policy are these services:
API Gateway (full access, all resources)
CloudFormation (full access, all resources)
Lambda (full access, all resources)
S3 (full access, all resources)
(Eventually) the deployment succeeded and I had a Lambda application serving the simple razor page showing the time.
I then copied the LambdaEntryPoint.cs, aws-lambda-tools-defaults.json and serverless.template files to my own dotnet core webapp (also a razor project) and attempted to deploy it to the same AWS account with the same command. The only changes made were the namespace of the LambdaEntryPoint class (reflected in the serverless.template file) and the .csproj file to include:
<AWSProjectType>Lambda</AWSProjectType>
and:
<PackageReference Include="Amazon.Lambda.AspNetCoreServer" Version="5.0.0" />
The dotnet lambda deploy-serverless ... command failed with the message:
User: arn:aws:iam::123456789120:user/lizard-user is not authorized to perform: iam:PassRole on resource: arn:aws:iam::123456789120:role/MyLizardAppServiceRole (Service: AWSLambdaInternal; Status Code: 403; Error Code: AccessDeniedException; Request ID: 12345678-1234-1234-1234-123456789012; Proxy: null)
I got the command to succeed by adding the IAM service to the MyLizardApp-Policy with the PassRole (all resources).
Why was this necessary for my personal app and not the demo solution from github? If the answer is not clear, what should I be looking for as differences? My personal app is not significantly different from the demo solution and I don't think the functional differences (in C#) would matter.
Whenever an AWS Service assumes (uses) an IAM Role, the service must have iam:PassRole permission to grant permission to use the Role. This required to prevent users from gaining too much permission.
For example, imagine a normal (non-Admin) user who launches an Amazon EC2 instance. When launching the instance, they can nominate an IAM Role to be assigned to the instance. If this user was permitted to select any IAM Role, they could select an Admin role and assign it to the EC2 instance. They could then login to the instance and use the credentials to make API calls as an Admin. This is an unwanted "privilege escalation".
Similarly, when an AWS Lambda function executes, it uses an IAM Role to obtain permissions. The iam:PassRole permission is used to control which roles a user can assign to the Lambda function.
So, there is something in that project that is trying to use an IAM Role and needs appropriate permissions.
First of all, we need to know what PassRole is:
iam:PassRole is the permission that controls which users can delegate an IAM role to an AWS resource.
As I can see in the repo, there is a file for CodeDeploy which already have credentials so maybe you are using CodeDeploy.
But btw, you are using an instances to deploy a Lambda function, and you need to pass the role to that Lambda so that is what PassRole do
AWS Services cannot directly assume service-linked roles. The role must be passed to the service by a user with the iam::PassRole permission.
The role-passing needs to be done only once, when a resource (e.g. EC2 instance) is created. After that the resource can assume the role repeatedly.
EC2 Instance profile is implemented this way. When a user launches an instance, it passes a role to the instance to act as an instance profile (it in addition needs iam:AddRoleToInstanceProfile for this case).
Other service-linked roles are also passed in this way.
Do not confuse it with the iam::CreateRole permission. A user may freely create service-linked roles, but is unable to pass the role to a service when needed.
In the management console, and to some extent in the CLI commands, role-passing is implicit, so you may encounter it without clear error messages when using non-root accounts.
As for why sometimes you need this permission, ands sometimes you don't, that is because when you use the root user, it will have AdministratorAccess which basically allows all actions on all resources.
If you create a new IAM user or account with blank permissions then you will need to add this permission manually.

AWS CodeCommit Permissions Errors in CodePipeline

I am attempting to follow this example of setting up an AWS Pipeline for use across multiple accounts. I have the four different accounts set up. I've followed through on each step of the process successfully. No commands are generating any errors. The pipeline completes successfully. I can then connect to the pipeline and commit my code changes. In short, every single step up to the final one works as written in the documentation.
However, I'm then presented with an error on the initial trigger of the code commit:
Insufficient permissions
The service role or action role doesn’t have the permissions required
to access the AWS CodeCommit repository named dbmigration. Update the
IAM role permissions, and then try again. Error: User:
arn:aws:sts::12345678912:assumed-role/my-pipeline-CodePipelineRole-1UPXOXOXO1WD0H/987654321
is not authorized to perform: codecommit:UploadArchive on resource:
arn:aws:codecommit:us-east-2:123456789:dbmigration
The AWS Account I used to create the pipeline is not the root account, but an IAM Administrator login with admin privileges across the account. I've tried adding AWSCodeCommitFullAccess and AWSCodePipelineFullAccess, which I would have thought would have been part of Administration anyway. However, that didn't change anything.
My assumption is I've done something horribly wrong, but I'm not able to identify what that is. Any suggestions for better troubleshooting, let alone suggestions on how to fix it would be most welcome.
The code used to create the pipeline, again, run using the IAM login, Administrator, from a fourth AWS account, is as follows:
aws cloudformation deploy --stack-name my-pipeline `
--template-file db-migration-master.yml `
--parameter-overrides ProjectName=dbmigration `
EmailAddress=grant#scarydba.com `
DevAccountId=98765432123 `
TestAccountId=123456789012 `
ProdAccountID=210987654321 --capabilities CAPABILITY_NAMED_IAM
All the templates are from the linked article and not modified or customized.
Based on the comments.
The error message indicated that the role my-pipeline-CodePipelineRole-1UPXOXOXO1WD0H/987654321 was missing permission codecommit:UploadArchive which:
Grants permission to the service role for AWS CodePipeline to upload repository changes into a pipeline
The solution was to add the codecommit:UploadArchive to the role as an inline policy.

Why I get AWS repository ARN could not be added to restrict access?

I create new policy which restrict access to repository.
I copy repository ARN:
But get error:
Why so I get this error and how to fix it?
UPD
I have added similar policies before:
Can not understand what is wrong now (
heh... answer my own question: I must type Region, Account, Repository name manually. They are not detected automatically from ARN

AWS Elastic Beanstalk: Environment Fails to Launch

I have created a few environments before so I know how the Amazon EBS works however lately I have been having the following issue while creating an environment:
The instance profile aws-elasticbeanstalk-ec2-role associated with the environment has no role. Please attach a role to the instance profile.
I follow the steps to create a new environment. When it gets to the Permissions page, I create a new role as there are no existing ones. Then I follow the rest of the steps and eventually it starts to launch. After a couple minutes, I get the error above. Any help towards this will be helpful.
To solve this issue, I created a new role from the IAM Manager console. I selected Amazon EC2 as my Service Role Type. I attached the AWSElasticBeanstalkFullAccess policy. Then when creating a new environment, I chose the new role I created.
Though I am bit late in answering this issue, posting here if someone faces this error now.
In case your user has all the required permission to create role and BS has already created the "The instance profile aws-elasticbeanstalk-ec2-role".
The reason of this error is due to roles only and when we try to launch EC2 from aws BS, it creates a role naming "aws-elasticbeanstalk-ec2-role" with required permissions.
But, if there is already a role with Trusted entities "AWS service: ec2" so BS will check permission in that already existing older role.
So go to Roles > Search ec2 related roles and select the role which is active.
And just add AWSElasticBeanstalkFullAccess permission to that ec2 role, and this issue will be resolved.