Having a different path every time here was a conscious design decision - having separate paths per run allows us to make sure runs don't interfere with each other and makes cleanup trivial.
Having some kind of 'global' wrapper for this brings in a bunch of issues:
Responsibility for creating and removing the wrapper (including global locking for preventing races),
Prescribing some behavior of what is the current global instance of a test cluster in case there's multiple ones running,
None of this is impossible (Kubectl already does this and we rely on this for test cluster access), but it's work and requires defining behaviour that will stick around forever in a user's home directory or general global installation. We can do this, but only if it's worth the effort.
Having a different path every time here was a conscious design decision - having separate paths per run allows us to make sure runs don't interfere with each other and makes cleanup trivial.
Having some kind of 'global' wrapper for this brings in a bunch of issues:
None of this is impossible (Kubectl already does this and we rely on this for test cluster access), but it's work and requires defining behaviour that will stick around forever in a user's home directory or general global installation. We can do this, but only if it's worth the effort.