Getty's on serial consoles need to be consistent

[Mark Shuttleworth]

In our cloud images today we launch a getty on ttyS0, as long as it's not in a container. We don't launch a getty on ttyS1-n even if they exist.

In MAAS, which also uses cloud images, it would often be useful to put getty's on the serial port that is mapped to remote serial access, such as IPMI SOL. It is however difficult to know which is the correct getty.

Broadly speaking I think we should have a consistent approach to getty's. That might mean:

 * launch a getty on each ttySn that passes an stty test (as per ttyS0 currently)
 * allow cloud-init to prevent some of those, via vendordata or userdata or default behaviour per-cloud

or

 * launch no getty's on ttyS, but
 * allow cloud-init to create them based on vendordata or userdata or default behaviour per-cloud

[Mark Shuttleworth (sabdfl)]

Not sure if this should also be tracked against the cloud image builder.

[Ben Howard (darkmuggle-deactivatedaccount)]

One big thing to consider with this is that some clouds (i.e CloudSigma and Joyent) use a serial console as a meta-data channel. In at least one case, activating a getty on the meta-data channel tty will result in the instance being unbootable. For this reason, I think that baking them into the Cloud Images is probably a bad thing.

[James Falcon (falcojr)]

Tracked in Github Issues as https://github.com/canonical/cloud-init/issues/2477