Steps to reproduce:

1. Provision a cloud node (with an instance cap of at least 2)
2. Set the node to Temporarily Offline
3. Run a job that would require the relevant node label

Buggy behavior:

Observe that no new instance is provisioned, despite the entry in the Build Queue

Expected behavior:

I expected an additional instance to be spun up to meet the demand in the queue

Investigation

https://github.com/jenkinsci/ec2-plugin/commit/7a3973e1129054029d5f260e6e4df2ddd17dcba

The above commit nerfs a check for whether a node is offline during the checkInstance() call. Thus an offline node that otherwise meets the critera will be selected for use.  If the node is in fact offline, the use will fail and provisioning will be re-attempted with the same results.

Since a node is always found to be "available", no new EC2 nodes will be started to handle the load.