Elastic Beanstalk stuck in an invalid state

I have Elastic Beanstalk with 2 env running. Envs are exactly the same, one for prod and one for test. During a deploy using codepipeline, which I use regularly, the test env got broken, and it is now in a WARN state with this message:

Environment health has transitioned from Pending to Warning. Initialization in progress (running for 3 minutes). There are no instances. Unable to assume role "arn:aws:iam::XXXXXXXXX:role/aws-elasticbeanstalk-ec2-role-dev". Verify that the role exists and is configured correctly.. The role exists and so the instance that has been initialised. Also, in the Enhanced instance health panel, there aere no instances despite there's a running EC2 instance created by this beanstalk environment.

Same if I try to rebuild environment. If I retry to deploy the code it takes 20 mins before to complete and go again in the WARN state. Code is not deployed anyway (I checked on EC2 with ssh). IN Beanstalk I can see the message "Elastic Beanstalk is updating your environment." which last for around 30 mins before to disappear. The related cloud formation stack shows a CREATE_COMPLETE state.

Does anyone know what the problem is or can point me in the right direction to debug it? I am sure it is not a problem with the deployed code, as I tried to deploy the prod working brancj with the same result.