Elastic Beanstalk app fails to deploy without public IPs on instances

Question

I can't get an elastic beanstalk app with a public ELB but private EC2 instances to work.

I created a basic eb config with eb init. This results in the following config:

branch-defaults:
  default:
    environment: test3
    group_suffix: null
global:
  application_name: test
  branch: null
  default_ec2_keyname: null
  default_platform: Node.js
  default_region: us-east-1
  include_git_submodules: true
  instance_profile: null
  platform_name: null
  platform_version: null
  profile: null
  repository: null
  sc: null
  workspace_type: Application

I use the default VPC, but with 2 custom public subnets (to prevent peering CIDR conflicts with another account and an external DB).

I now try to deploy the eb app with the following options:

# powershell
eb create --profile dev `
          --sample `
          --vpc.id vpc-123abc 
          --vpc.ec2subnets "subnet-123,subnet-456" `
          --vpc.elbsubnets "subnet-123,subnet-456" `
          -sr arn:aws:iam::<account>:role/service-role/aws-elasticbeanstalk-service-role `
          --vpc.elbpublic `
          test8

This does not work. The initial instance never passes health checks and the creation of the EB environment is considered a failure after 15 minutes or so. It is never accessible from the internet.

HOWEVER, adding the option --vpc.publicip makes this work! -- I can access the webpage just fine from the internet using the EB environment address. But I shouldn't need public ips on each instance. So what am I missing?

I ran both these commands with different environment names and compared their security groups, ELB settings, etc. I can't find any differences. Why do these instances need public IPs to pass health checks and connect to the ELB?

Answer 1

You currently have the instances deployed to public subnets, without public IP addresses assigned, so the instances have no access to the Internet. For an EC2 instance to have access to the Internet one of the following scenarios must be met:

• Deployed to a public subnet (subnet with a route to an Internet Gateway) and public IP address assigned to the EC2 instance.
• Deployed to a private subnet (subnet with a route to a NAT Gateway).

Elastic Beanstalk instances need to communicate with the Elastic Beanstalk service API. They can't do that without either Internet access, or endpoints configured in the VPC.

Even if you configure those endpoints in your VPC so that the instances can communicate with the AWS services they need to talk to, the servers may still fail to start up if you have them configured to download anything from the Internet like application dependencies or OS updates, in which case you would need to configure them with Internet access.