When Your Docker Image Bloats from 900MB to... Wait, How Did We Get Here?

A

Adil Sher

Author

Aug 2, 2026
4 min read
11 views
When Your Docker Image Bloats from 900MB to... Wait, How Did We Get Here?

I was debugging a deployment issue at 2 AM last week when I realized our production Docker image had somehow ballooned to nearly a gigabyte. A gigabyte. For a Node.js API that should've been maybe 150MB lean. That's when I came across this article about image size optimization, and it hit me like a wall, we were doing everything wrong in our build process.

The title "900MB 8MB says it all" is deliberately stark, and for good reason. It's not hyperbole; it's a wake-up call. The difference between a bloated container and an optimized one can literally be that dramatic. And when you're running dozens of replicas across multiple environments, that's the difference between reasonable infrastructure costs and hemorrhaging money on cloud storage and bandwidth.

The Problem Nobody Talks About Until It's Too Late

Most developers, myself included, ship their first Docker images without much thought. You build locally, it works, you push it to a registry, done. But nobody's really watching what actually gets baked into that image.

The issue is subtle. Your build context includes everything: node_modules, build artifacts, temporary files, development dependencies, git history inside the image, unnecessary system packages. Each layer in your Dockerfile adds up. Before you know it, your image is pulling down half your hard drive just to start a container.

I've seen this happen in real projects. We had a Python service that included the entire pip cache. Another team was shipping their entire .git folder inside their Go binary. These weren't malicious mistakes, they were just oversights from developers (like me) who hadn't internalized the consequences of container bloat.

Where the Waste Actually Happens

There are three main culprits in my experience:

First, multi-stage builds aren't being used. This is the low-hanging fruit. If you're using a heavy builder image (like node:18 or python:3.11) and shipping that same image to production, you're carrying around compilers, dev headers, and build tools you'll never execute. Multi-stage builds let you compile in one image and copy only the artifacts to a clean final image.

Second, layer caching is being ignored. Docker builds layers sequentially, and if you're doing RUN apt-get install followed by COPY . ., then modifying your code, you're rebuilding dependencies every single time. The order matters enormously.

Third, nobody's cleaning up. After installing packages, running build steps, or compiling code, the intermediate artifacts stick around. A simple RUN apt-get install && apt-get clean && rm -rf /var/lib/apt/lists/* cuts gigabytes from Debian-based images.

A Practical Example from My Own Work

Here's what I started doing after this realization:

# Builder stage - heavy, with all dependencies
FROM node:18-alpine AS builder
WORKDIR /app
COPY package*.json ./
RUN npm ci --only=production

# Final stage - lightweight
FROM node:18-alpine
WORKDIR /app
COPY --from=builder /app/node_modules ./node_modules
COPY src ./src
COPY package.json ./
RUN npm prune --production
USER node
CMD ["node", "src/index.js"]

Notice what we're doing: building dependencies in a separate stage, then copying only what we need. We're using Alpine Linux (around 40MB vs 900MB+ for standard Node images). We're running as a non-root user. We're not copying .git, test/, or README.md into production.

I ran this against one of our services, and the image went from 680MB to 85MB. Same functionality, 8x smaller.

What Actually Matters Here

The original article's point isn't academic. Container size directly impacts:

  • Cold start times: Smaller images pull faster. In Kubernetes, that's the difference between handling a spike and getting crushed by it.
  • Registry storage costs: If you're shipping hundreds of these daily, registry storage becomes real money.
  • Network bandwidth: Every deployment, every node pull, every container rollout, it all adds up.
  • Security surface area: More layers, more packages, more potential vulnerabilities.

I think about this differently now. A 900MB image isn't a quirk; it's technical debt we're pushing onto operations and your cloud bill every single day.

The Follow-Up Question I'm Wrestling With

Here's what I want to know: why aren't image size and optimization built into our CI/CD pipelines from the start? Most teams have automated tests, linting, and security scanning, but how many actually fail a build if an image exceeds a reasonable threshold?

I'm adding image size checks to our GitHub Actions now, if an image comes in over 200MB for our services, the build fails with a report. It forces the conversation upfront instead of discovering it in production.

What's your biggest Docker image actually running? Let me know in the comments, I'm curious what the real-world range looks like.


Source: This post was inspired by "900MB 8MB says it all. Starred." by Dev.to. Read the original article

Written by Adil Sher

Full stack developer building high-traffic platforms, AI services, and custom web applications. Explore my portfolio, learn about my background, or get in touch.

Related Articles

Stop Jumping to Kubernetes Before You Can Debug a Linux Server
DevOps & Cloud Nov 9

Stop Jumping to Kubernetes Before You Can Debug a Linux Server

I made this mistake. I remember sitting in front of a Kubernetes cluster that wasn't working, staring at pod crash logs that might as well have been written in Klingon. I didn't understand what `kubectl logs` was showing me because I'd never actually ssh'd into a bare Linux box a...

Linux Architecture Isn't Just Theory, It's Your Superpower in Production
DevOps & Cloud Nov 8

Linux Architecture Isn't Just Theory, It's Your Superpower in Production

I spent three years shipping containerized applications before I actually understood how Docker worked. I could write Dockerfiles, orchestrate services with Kubernetes, and troubleshoot port conflicts, but when things broke at 2 AM, I was guessing instead of reasoning. The turnin...