calculator.dtDockerTitle
詳細ガイド 近日公開
Docker Container Size Estimatorの包括的な教育ガイドを準備中です。ステップバイステップの解説、数式、実例、専門家のヒントをお届けしますので、もうしばらくお待ちください。
The Docker Container Size Estimator calculates approximate Docker image size based on choice of base image (scratch, alpine, distroless, debian, ubuntu), runtime (Node.js, Python, Go, Java, Ruby), package dependencies, application code, and optional build tools. Container size has substantial practical impact: pull time (CI/CD pipeline speed and autoscaling responsiveness), storage costs (DockerHub, AWS ECR, GCR registry fees), security surface (more code = more potential vulnerabilities), and developer productivity. A 50 MB image deploys in seconds; an 800 MB image takes minutes and feels sluggish across the entire development workflow. Docker image composition follows predictable patterns. The base layer (operating system or minimal runtime environment) provides foundational libraries. Runtime layer (language interpreter or compiled binary support) adds 0-170 MB depending on language. Package dependencies (npm, pip, apt-get) typically range from 5-50 MB per package depending on the ecosystem and specific libraries. Application code is usually the smallest component (10-100 MB for most apps). Build tools (compilers, dev dependencies) add 100-500 MB if left in the final image without multi-stage builds. Base image choice dramatically affects total size. Scratch (0 MB) works for compiled static binaries (Go, Rust) and produces 10-30 MB total images. Alpine (7 MB) is the most popular minimal Linux base — fast pulls but uses musl libc which can cause compatibility issues with some Python wheels and Java applications. Distroless (Google, 25 MB) includes no shell or package manager, providing strong security through minimization. Debian-slim (80 MB) balances size and compatibility for most use cases. Full Debian (120 MB) or Ubuntu (78 MB) are appropriate when you need broad library compatibility but are oversized for most production workloads. This calculator helps engineers right-size their containers before building, understand the cost contribution of each component, and identify optimization opportunities. Enter base image, runtime, package count and average size, application code size, layer count, and whether to include build tools. The calculator provides total size, size category (small/medium/large/very large), estimated pull time at 50 Mbps, and a visual breakdown by component. Use the comparison feature to test alternative base images and runtimes to find the optimal balance for your application.
Total Size = Base Image + Runtime + (Package Count × Avg Package Size) + App Code + Build Tools + (Layers × 0.5 MB)
- 1Step 1 — Select Base Image: Choose the foundational OS layer. Decision factors: Scratch for static binaries (smallest but no shell/utils), Alpine for general use (smallest with package manager), Distroless for security (no shell, very small), Debian-slim for compatibility (most common production), full Debian/Ubuntu only when broad library compatibility required.
- 2Step 2 — Select Runtime: Add language runtime on top of base. Static-compiled languages (Go, Rust) need no runtime — select 'none.' Interpreted languages need runtimes: Node.js, Python, Ruby, Java. Choose slim variants when available (python-slim, java-slim) for ~40% size reduction with mostly equivalent functionality.
- 3Step 3 — Enter Package Information: Count direct + transitive packages your app depends on at runtime (production only, not dev dependencies). Find via 'npm ls --production' (Node), 'pip list' (Python), 'go list -m all' (Go). Estimate average package size or use defaults: 15 MB Node, 10 MB Python, 5 MB Go.
- 4Step 4 — Enter App Code Size: Measure your built application output. For Node: 'du -sh dist/' or 'du -sh node_modules/.cache/' to see compiled size. For Python: source code size (usually small) plus any compiled extensions. Includes static assets (images, CSS) shipped with app.
- 5Step 5 — Enter Layer Count: Open your Dockerfile and count distinct layer-creating instructions (RUN, COPY, ADD). Each adds ~0.5 MB metadata overhead. Optimization: combine multiple RUN commands with && to reduce layer count. Production Dockerfiles typically have 5-10 layers in final image.
- 6Step 6 — Include Build Tools: Check this box if your Dockerfile keeps build tools (gcc, make, dev headers) in the final image. Adds ~200 MB. Typically indicates missing multi-stage build optimization. Removing build tools with multi-stage can reduce image size by 30-50%.
- 7Step 7 — Review Total Size and Recommendations: Calculator shows total MB, category (small to very large), estimated pull time at 50 Mbps connection, and pie chart breakdown of components. Use comparison feature to test alternative base images or remove components to see size impact. Target small (<100 MB) or medium (<300 MB) for production.
Typical mid-sized Node.js production image
Breakdown: Alpine 7 MB + Node 20 runtime 140 MB + 10 packages × 15 MB = 150 MB + app 25 MB + 8 layers × 0.5 MB = 326 MB total. The 150 MB in packages is the largest component — most optimization opportunities involve reducing package count or using lighter alternatives (replacing lodash with native ES2022, replacing moment with date-fns).
Optimal Go container — compiled binary in scratch image
Go compiles to a single static binary with no runtime dependencies. Scratch base (0 MB) + compiled binary (15 MB) + 3 layers (1.5 MB) = 16.5 MB total. Pulls in 3 seconds, deploys instantly, zero unused attack surface. This is the gold standard for container size optimization. Rust and C++ static binaries achieve similar results.
Bloated — needs multi-stage build optimization
Breakdown: Debian 120 MB + Python 55 MB + 20 packages × 15 MB = 300 MB + app 50 MB + build tools 200 MB + 12 layers × 0.5 MB = 731 MB. The 200 MB build tools (compilers, dev headers) are unnecessary in production. Multi-stage build (compile in fat image, copy artifacts to slim runtime) would cut to ~370 MB. Switching base to debian-slim or python-slim adds another 30-50 MB savings.
Reasonable Java production image — Java runtimes are inherently large
Java JREs are large (170 MB for Java 17 base). Java-slim variants reduce to 100 MB. Application JARs add bulk. Total ~472 MB is reasonable for Java but feels large to engineers used to Go (16 MB) or Node (300 MB). Java's GraalVM native-image can compile to small static binaries (~50 MB) but with build complexity tradeoffs.
Sizing CI/CD pipeline storage requirements before deploying new microservices
Comparing base image trade-offs for security vs convenience in production deployments
Calculating registry storage costs (DockerHub, AWS ECR, GCR) for organizations with hundreds of images
Optimizing Kubernetes pod cold-start time by reducing image pull duration
Estimating bandwidth costs for edge deployments where images are pulled to many distributed locations
| Image | Size | Notes |
|---|---|---|
| scratch | 0 MB | Empty base for static binaries (Go, Rust) |
| alpine:latest | 7 MB | Minimal Linux with package manager (musl libc) |
| wolfi-base | 12 MB | Chainguard's modern alpine alternative (glibc) |
| distroless/static | 2 MB | Google's distroless base, no shell |
| distroless/nodejs | 55 MB | Node.js runtime in distroless |
| distroless/java17 | 180 MB | Java 17 runtime in distroless |
| debian:slim | 80 MB | Minimal Debian with apt |
| debian:latest | 120 MB | Full Debian with utilities |
| ubuntu:latest | 78 MB | Ubuntu LTS minimal |
| node:20 | 1 GB+ | Full Node.js with build tools — avoid for production |
| node:20-alpine | 140 MB | Production-ready Node.js |
| python:3.11 | 1 GB+ | Full Python with build tools |
| python:3.11-slim | 120 MB | Production Python |
| python:3.11-alpine | 60 MB | Smallest Python (compatibility caveats) |
Why does container size matter?
Larger images mean: slower CI/CD pipelines (every push pulls the image), slower autoscaling (cold starts increase as image size grows), higher registry storage costs (especially for organizations with many services), larger attack surface (more code = more potential CVEs), longer rollback times during incidents. Reducing from 800 MB to 200 MB can cut CI pipeline times by 30-50% and improve developer iteration speed substantially.
What's the smallest practical Node.js image?
node:20-alpine (~140 MB) or use multi-stage build with distroless/nodejs as final stage (~80 MB). For maximum compression, consider Bun (alternative JavaScript runtime, slim Docker images) or compile-to-Deno using static binaries. AWS Lambda's Node.js runtime is roughly 40 MB total. For most Node.js production apps, 100-200 MB is achievable with reasonable effort.
Is alpine always best?
No — alpine uses musl libc instead of glibc, which can cause compatibility issues with some packages. Common problems: Python wheels (numpy, scipy, pandas often need glibc), Java JIT compilation issues, some Node.js native modules. Distroless or debian-slim are safer for Python; alpine excels for Node.js, Go, and Rust. Test alpine vs slim in your specific stack before committing.
What is a multi-stage build?
Dockerfile feature that uses multiple FROM statements. Build stage uses fat image with all compilers and dev tools; final stage uses slim runtime image and copies only the compiled artifacts from the build stage. Typical pattern: build in node:20 (heavy), copy dist/ to node:20-alpine (light) for production. Results in 40-70% size reduction compared to single-stage builds.
How do I find what's making my image large?
Use 'docker history <image>' to see layer-by-layer size contribution. 'dive' tool (CLI) provides interactive layer exploration showing exactly which files contribute to each layer. 'docker images' with --format option helps inventory image sizes across your stack. For deep analysis, container-diff (Google) compares two images and shows differences.
What's wrong with using full debian or ubuntu?
Nothing inherently — but they include hundreds of packages most applications don't need (man pages, documentation, multiple shells, development utilities). For production runtime, you typically need only libc, your runtime, and your app. Slim variants strip the unnecessary content. Use full debian/ubuntu only when you genuinely need broad utility availability (development sandboxes, CI runners).
Should I optimize for size or build speed?
Both, with size taking priority for production images and build speed for development. Production: optimize for small final image (multi-stage builds, slim bases, layer optimization). Development: optimize for fast incremental rebuilds (BuildKit caching, .dockerignore, multi-stage cache mounts). These goals are typically complementary — well-structured Dockerfiles achieve both.
プロのヒント
Use multi-stage builds to dramatically reduce final image size — build in a fat image with all tools, copy only the artifacts to a slim runtime image. Typical 60-80% size reduction. Combine with .dockerignore to exclude tests, docs, .git directory, and development files. Most Node.js apps can go from 1 GB+ images to 200 MB with multi-stage + alpine. Most Python apps from 1 GB+ to 250 MB with multi-stage + slim.
ご存知でしたか？
The smallest known production Docker image is just 3.5 KB — a 'hello world' Go program in a scratch base image. Docker initially launched in 2013 with images that were 500 MB+ for basic Linux. Over a decade of optimization tools, distroless bases, multi-stage builds, and language ecosystem improvements (Go static binaries, Bun's slim runtime) have made 50-200 MB production images standard. Some companies report deploying entire microservice fleets averaging under 100 MB per service.