That said, spanner seems like it does what it needs to do far better than any academic system. For example. And there are many, many other academic pursuits that are totally ignored by industry.
This can't be just narrow mindedness -- if there was any utility in, say, most of the type theory that's directly pumped from academia into Haskell, startups would be using it. But they aren't. They're not using that, they're not using academic database concepts, they're not using anything from academia. And virtually everything academic is open sourced at some point, so it's not like that's an excuse.
(Are Plan 9 or Solaris academic systems? They're both the result of industrial R&D groups...)
Solaris was a 30 year series of successive engineering projects focused on providing valuable concrete improvements to what was previously just an academic project, the early unix releases were definitely academic regardless of their future commercial impact.
Docker is an awful example. You actually have your history of containers off. By 1998, they weren't just academic, but already an emerging industry standard in at least IBM i machines.
The only reason Docker exists is because an ex-Plan 9 developer (Al Viro) introduced namespaces into the Linux kernel to make the VFS model more tolerable, and some years later some Google developers lifted Solaris process contracts to make cgroups. The first and central part of Linux containers, then, is a direct result of half-assing an academic system software advance for its usefulness.