[Bug]: arm64 image ships ~2.4GB of build-time waste (core dumps + duplicate Chromium cache), root-caused to QEMU CI emulation

crawl4ai version

0.9.2

Expected Behavior

The published arm64 Docker image should only contain files needed at runtime. Build/test artifacts (crash dumps from the crawl4ai-doctor sanity check) and intermediate install artifacts (the root-owned Playwright browser cache, superseded by the copy made for appuser) should be cleaned up before the image layers are committed.

Current Behavior

unclecode/crawl4ai:latest (arm64) reports as 9.06GB via docker images. Inspecting a freshly-pulled container shows two separate sources of dead weight, totaling ~2.4GB of its ~5.5GB live filesystem:

1. Core dumps in /app (~926MB)

-r--r--r-- root root 510M /app/qemu_headless_shell_20260715-085214_28.core
-r--r--r-- root root 208M /app/qemu_headless_shell_20260715-085211_34.core
-r--r--r-- root root 208M /app/qemu_headless_shell_20260715-085211_32.core

• These are dated at image build time, not container runtime, and are root-owned/read-only, so end users can't remove them from a running container.

2. Duplicate Chromium install (~1.5GB)

1.5G  /root/.cache/ms-playwright       (dead - root never runs; container runs as appuser)
958M  /home/appuser/.cache/ms-playwright  (actually used at runtime)

• playwright install --with-deps installs the browser to /root/.cache (running as root during build), then a later RUN step cp -r's it to /home/appuser/.cache for the actual runtime user, but never deletes the original. Both copies ship in the final image.

Is this reproducible?

Yes

Inputs Causing the Bug

N/A — not input-dependent. Present in every pull of the arm64 image, not triggered by user config.

Steps to Reproduce

Steps to Reproduce:
1. docker pull unclecode/crawl4ai:latest
2. docker run --rm -it unclecode/crawl4ai:latest bash
3. ls -lh /app/*.core
4. du -sh /root/.cache/ms-playwright /home/appuser/.cache/ms-playwright

Code snippets

docker history — confirms duplicate layer cost:

$ docker history --no-trunc unclecode/crawl4ai:latest | grep -E "chown -R root|cp -r.*ms-playwright"
972MB   RUN chown -R root:root ${APP_HOME} && chmod -R a-w ${APP_HOME}

1GB     RUN mkdir -p /home/appuser/.cache/ms-playwright \
          && cp -r /root/.cache/ms-playwright/chromium-* \
                   /root/.cache/ms-playwright/chromium_headless_shell-* \
                   /home/appuser/.cache/ms-playwright/ \
          && chown -R appuser:appuser /home/appuser/.cache/ms-playwright



docker-release.yml — confirms cross-arch QEMU build:


$ grep -A3 "platforms:" .github/workflows/docker-release.yml

          platforms: linux/amd64,linux/arm64
          cache-from: type=gha
          cache-to: type=gha,mode=max

OS

Linux (Raspberry Pi OS, arm64

Python version

3.12

Browser

Chromium (bundled headless_shell, via Playwright build 1228)

Browser version

149.0.7827.0

Error logs & Screenshots (if applicable)

docker history shows the chown -R root:root $APP_HOME && chmod -R a-w $APP_HOME layer at 972MB, matching the ~926MB of .core files in /app. du -sh on a running container shows /root/.cache/ms-playwright (1.5G) and /home/appuser/.cache/ms-playwright (958M) as separate, duplicate browser installs

Start with the arm64 build configuration in .github/workflows/docker-release.yml and locate the steps that install and copy Playwright browsers during the image build. Reproduce the reported sizes with docker history and du, then verify that the arm64 image has no /app/*.core files or unused /root/.cache/ms-playwright copy while the appuser browser cache remains usable.