Description
The ssh_user and ssh_port parameters are complex and undocumented.
Old (released) system. The template create.yml hard coded ssh_user and ssh_port to ubuntu and 22. This was passed in to the instance config. driver.py read these from the instance config and used them to set the ansible_user and ansible_port portions of the connection options. To override them required setting ansible_user and ansible_port in the connection_options portion of the platform definition.
New system. The template create.yml reads ssh_user and ssh_port from the platform, the defaults are ubuntu and 22. This is used during the creation process, the user is set via cloud-init unless the key_inject_method is overridden and the port is used when polling until the instance starts. These are then passed via the instance config to driver.py and used to populate ansible_user and ansible_port for downstream inventory use.
I detailed both to highlight the backwards compatibility issue. My prior practice was to specify the ansible_user in connection_options, for example setting it to admin for a Debian distribution. This configured the inventory for the subsequent ansible stages, commands such as molecule login also worked. Using current master that breaks badly, cloud-init is used during create to set the user to ubuntu. The inventory is then configured to use admin, which breaks everything. The cause is not obvious.
I feel there are also issues with the name. Ansible is moving away from ansible_ssh_user to ansible_user. Setting ssh_port to 5986 for winrm looks weird.
Activity