API incompatability: ERROR no such request "DestroyMachines" on Client

[Curtis Hovey]

A tester packaged and installed his own 1.16.4. He boostraped without
--upload tools which setup a 1.16.3 server. The call to
terminate-machine produced
    ERROR no such request "DestroyMachines" on Client

This happened because Juju believes that patch-level differences are API compatible. The client does not require the server to be the same version, nor does it warn that the server and client are different.

* This will not happen when the user bootstraps with --upload-tools.

* This is unlikely to happen during bootstraps after the 1.16.4 release of the
  tools to the public clouds.

* This error will be seen for anyone who upgrades their own machine,
  but fails to use juju-upgrade on their deployments.

* This error will *also* happen when 1.18.0 is released and users fail
  to upgrade their deployments

This bug was introduced with r1985 or r1983, possibly both.

[Curtis Hovey (sinzui)]

We expect users to upgrade to 1.18.0 and manage 1.16 environments for a period of days to weeks during phased upgrades. How can we ensure that this change is compatible for Canonical's own deployments?

[John A Meinel (jameinel)]

This should be fixed, I didn't realize we had a different bug tracking it.
See bug #1253643