Because containers share the host Linux kernel, running as UID 0 inside the container means the process runs with root privileges on the host kernel. If an attacker exploits an application vulnerability (such as a path traversal or remote code execution), they can exploit container breakout vectors to take over the host operating system. Always create an unprivileged user and declare USER nonroot.

Placing COPY . . before running dependency installation (e.g. npm install, pip install, or go mod download) causes Docker to bust its layer cache on every single commit. The builder must reinstall all third-party dependencies from the network on every build, increasing CI/CD pipeline runtimes from 15 seconds to 10 minutes.

When a container starts with node server.js or python main.py directly, the interpreter becomes PID 1. Linux kernels treat PID 1 specially: signals like SIGTERM are ignored unless explicitly handled by the process. When Docker or Kubernetes stops the container, the process never receives the signal, fails to close database connections gracefully, and is forcefully SIGKILLed after 10 seconds. Wrapping execution with tini -- resolves this cleanly.

Passing private npm tokens, GitHub personal access tokens, or AWS keys using ARG SECRET_KEY bakes the secret into the Docker image metadata and layer filesystem. Anyone with pull access to the image can inspect historical layers via docker history or dive. Use BuildKit secret mounts (--mount=type=secret) so sensitive keys never persist in layer history.

When executing docker build, the CLI client tars the entire directory and uploads it to the Docker daemon as the build context. Without a .dockerignore, gigabytes of local node_modules, .git history, local temporary builds, and sensitive local .env files are uploaded, drastically slowing down builds and introducing catastrophic secret leakage.