Why does ordering of Dockerfile instructions matter, and how do you exploit layer caching for a Java build?

Each instruction is a cached layer keyed by its content and the previous layers; a change invalidates that layer and everything after it.

So you copy and resolve dependencies before copying source — for Maven, COPY pom.xml then mvn dependency:go-offline, then COPY src — so editing code only busts the cheap source/compile layers while the expensive dependency layer stays cached. Putting COPY . . too early is the classic mistake: any file change re-downloads all dependencies, making every build slow.