I was at Apple around 15 years ago working as a sysadmin in their hardware engineering org, and everything - and I mean everything - was stored on NFS. We ran a ton of hardware simulation, all the tools and code were on NFS as well as the actual designs and results.
At some point a new system came around that was able to make really good use of the hardware we had, and it didn’t use NFS at all. It was more “docker” like, where jobs ran in containers and had to pre-download all the tools they needed before running. It was surprisingly robust, and worked really well.
The designers wanted to support all of our use cases in the new system, and came to us about how to mount our NFS clusters within their containers. My answer was basically: let’s not. Our way was the old way, and their way was the new way, and we shouldn’t “infect” their system with our legacy NFS baggage. If engineers wanted to use their system they should reformulate their jobs to declare their dependencies up front and use a local cache, and all the other reasonable constraints their system had. They were surprised by my answer but I think it worked out in the end: it was the impetus for things to finally move off the legacy infrastructure, and it worked out well in the end.
At some point a new system came around that was able to make really good use of the hardware we had, and it didn’t use NFS at all. It was more “docker” like, where jobs ran in containers and had to pre-download all the tools they needed before running. It was surprisingly robust, and worked really well.
The designers wanted to support all of our use cases in the new system, and came to us about how to mount our NFS clusters within their containers. My answer was basically: let’s not. Our way was the old way, and their way was the new way, and we shouldn’t “infect” their system with our legacy NFS baggage. If engineers wanted to use their system they should reformulate their jobs to declare their dependencies up front and use a local cache, and all the other reasonable constraints their system had. They were surprised by my answer but I think it worked out in the end: it was the impetus for things to finally move off the legacy infrastructure, and it worked out well in the end.