Multi-stage builds, slim bases, layer ordering and .dockerignore, the techniques that cut image size by 90% and speed up every deploy.
Why size matters more than it seems
Image size is not aesthetic. Every gigabyte is pulled on deploy, stored per host, kept per release for rollbacks, and pruned eventually by someone at 2 a.m. when the disk fills. Big images slow every deploy, stretch rollback windows, and inflate the attack surface (more packages, more CVEs in every scan). The good news: 90% reductions are routine with four techniques, none of which change your application code.
Technique 1: multi-stage builds
The heavy hitter. Build with the full toolchain image; copy only the artifacts into a minimal runtime stage. Compilers, dev dependencies and source never reach production:
