When downgrading from 1.42 to 1.39 everything works with the exact same setup (no changes, just downgrading by manually uploading the hpi).

The second problem is a result of the first problem, but for some reason it would spin up 10+ instances almost instantly (could be that new setting "launch new instances right away"?) - the error was the same, that is unable to connect to "null:22"

Setup:

• VPC with private/public subnets - private subnets outgoing via natgw (https://registry.terraform.io/modules/terraform-aws-modules/vpc/aws/1.51.0)
• jenkins nodes are launched of a custom ami (debian + openjdk8/encrypted boot) with their own custom instance-profile
• jenkins nodes are launched in private subnet with no public ip - a security group that explicitly only gives the jenkins-master access via port 22

Hope this info helps, else let me know  

A couple of questions :

• did you update the IAM policy connected to the user/role used for raise new node ? (
  ...
  Effect": "Allow",
  "Action": [
  "iam:ListInstanceProfilesForRole",
  "iam:PassRole"
  ],)
• Is Jenkins master with Public ip ? 
• Jenkins master and slaves are in the same VPC  and subnet? 
• What is the configuration of the ec-plugin (use public dns, ...), the option enabled?

For the number of the nodes started, now the plugin is much more "reactive" to the status of the queue

Hitting this issue as well. I was able to work around it by not using spot instances.

thoulen

• yes, those IAM actions are in place
• jenkins master is with private IP but reachable over a public IP behind an ALB
• jenkins master and nodes are in the same VPC, private subnets, but might be in different subnet-ids depending on launched AZ
• public-dns is not enabled

joshuaspence might be correct, we are launching spot ec2 c4.xlarge - so might be an issue with spot instances

I have the same symptoms with spot instance private IP's not being identified by the ec2-plugin but ours also have public IP's assigned in the configuration.

I downgraded to 1.39 so it behaves in the meantime.

We are using spot instances too. Configuring master to connect to slaves using neither public IP or private IP works. Always

Jan 21, 2019 6:41:58 AM hudson.plugins.ec2.EC2Cloud
INFO: Failed to connect via ssh: There was a problem while connecting to null:22

The plugin is indeed unusable for spot instance use case since 1.40, there was always different error in each version.

On Jenkins 2.150.2 EC2 plugin 1.42 we get this behavior intermittently. We oscillate between this error and the error in JENKINS-55639.

With the null:22 error we additionally get the horrible side effect that the script continues to connect to the master node and runs the init script. Just by chance the init script I had in place wasn't destructive to the master. Now I have a little addition to my init script at the top:

master=jenkins-master
host=$(hostname -s)
if [ $host == $master ]
then
 echo "This is the master node! Exiting!!!"
 exit 1
else
 echo "Server is a spot node, apparently. Better yet, it's NOT the master node. Continuing..."
fi

In the 1.43  will be added an option to specify how to connect to slave (private or public)

Is there an estimate of when 1.43 will be released?

1.43 has been released