I'll generally link to - and quote - TFM in the process, with a view to providing a dupe-target. (If one doesn't exist already)

IMO one of the key advantages of docker is that you can automate the dev->test->prod deployment. As such, I think we'll increasingly see it in prod, for all the reasons we see virtual machines in prod.

Yep, that's it. But I submit it is (or can be) a journey

Easy way to get rep on a site that doesn't make it easy perhaps?

I would agree - we all know legacy is an ongoing headache that we'd like to be rid of. But sometimes additional constraints apply that the IT guy in the middle has no control over. Even if the answer is simply "Upgrade, because this problem isn't present in X" - that's the start of a business case that helps me get the funding to do it.

I don't agree - whilst "good practice" means "don't keep legacy" - there are plenty of places where additional constraints apply, meaning that you do have "legacy" sticking around far longer than is optimal.