LINISNIL

bind mounts are what i’m accustomed to using for local docker dev. they’re the typical go to for being able to use host-native dev tooling to edit source code and ensuring that those changes are immediately reflected in the container environment. they work fine, for the most part.

my biggest issue with them has always been the speed on mac. every company i’ve worked at provisions mac’s for dev machines and the biggest reason for the slowness is because on a mac, docker is actually running in a linux VM. afaik there is no native mac container technology – they’re built for linux. so docker makes this work by running a hypervisor using apples virtualization framework. virtualization is cool but it’s expensive.

when docker desktop starts up, it mounts paths like /Users into the virtual machine and makes that available to processes running inside the container. this is what allows source directories in the mac host machine to be bind-mounted. unfortunately, every i/o operation incurs an extra cost in mapping a read or write operation in the virtual machines virtual file system to a read or write on the actual host file system.

i don’t think there’s much you can do about this cost – it’s always going to be an extra level of indirection so it’s not going to ever match native, no-container i/o speeds. but there are some alternatives that i’m eager to try out in the future

• dev containers is basically taking containerization to the extreme – what if your entire workflow / tooling was inside a container?!
• docker released https://docs.docker.com/desktop/synchronized-file-sharing/ , this tackles this problem by asking the question “what if we can copy/sync file changes really fast into the container?” instead of forcing the vm to reach across file system boundaries