frontend 이미지는 nginx 공식 이미지라 root로 시작해 80번을 열고 worker를 nginx 사용자로 내린다. 그래서 Kubernetes에서 runAsNonRoot와 drop ALL을 둘 수 없다(k3s 성숙도 점검 2026-11-11, 7.5절). 운영 매니페스트는 allowPrivilegeEscalation: false만 두고 PSA restricted 경고를 낸다. runtime base를 nginxinc/nginx-unprivileged:1.27.5-alpine(digest 고정)으로 바꾼다. 예전 base와 같은 NGINX 1.27.5·Alpine이고 wget도 있어 HEALTHCHECK가 그대로 돈다. 파일을 고치는 동안만 root로 있다가 USER 101로 돌아간다. 생성되는 server 블록은 listen 8080이고, EXPOSE·HEALTHCHECK·dev-stack 포트 매핑도 8080으로 맞춘다. 버린 대안: Debian판 nginx-unprivileged. wget이 없어 HEALTHCHECK가 늘 실패한다. 80번을 그대로 두고 NET_BIND_SERVICE만 남기는 안도 버렸다. drop ALL을 못 하고 PSA restricted도 통과하지 못한다. 위험·한계: 이 이미지는 8080을 열므로 운영 매니페스트의 포트를 같이 바꿔야 한다. deploy/production/frontend-nonroot가 그 일을 하고, 기본 경로에는 넣지 않는다. dev-stack의 443·/etc/nginx/tls 부분은 이미지에 없는 파일을 쓰는 예전 상태 그대로다. 이 repo의 pnpm 게이트와 전체 docker build는 돌리지 않았다(heavy lock 대기가 길어 취소). 검증: desktop에서 이 Dockerfile의 runtime 단계를 그대로 쓰고, 빌드 단계만 운영 이미지 7a4e201의 dist와 conf(listen 8080으로 바꿈)로 대신해 이미지를 만들었다. --user 101 --cap-drop ALL --security-opt no-new-privileges --sysctl net.ipv4.ip_unprivileged_port_start=1024로 띄우면 GET /config.json 200, GET / 200, CapEff 0, NoNewPrivs 1이고, Docker health는 healthy다.
106 lines
4.8 KiB
Docker
106 lines
4.8 KiB
Docker
# syntax=docker/dockerfile:1
|
|
#
|
|
# The frontend deployment artifact. The repository had none — `dist/server.mjs`
|
|
# is a preview server that applies neither the security headers nor the cache
|
|
# policy `config/hosting/` declares — so a deployment had nothing to run.
|
|
#
|
|
# Two stages: the build produces `dist/` and, from the serving contract, the
|
|
# nginx configuration that matches it; the runtime is nginx with both.
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# build
|
|
# ---------------------------------------------------------------------------
|
|
# Pinned by digest: the release-provenance gate requires an immutable runner
|
|
# identity, and a floating tag cannot give one.
|
|
FROM node@sha256:3638d9a6fe4030bd716be989438248074489337ba3275657f93595428be4fc03 AS build
|
|
WORKDIR /src
|
|
|
|
# The profile is baked at build time (scripts/generate-runtime-config.ts), so it
|
|
# has to be chosen here rather than at `docker run`. `dist/config.json` stays a
|
|
# separate file in the image, which is what makes build-once/promote possible:
|
|
# a deployment can replace just that file without rebuilding the bundle.
|
|
ARG APP_PROFILE=production
|
|
ENV APP_PROFILE=${APP_PROFILE}
|
|
|
|
# The bundle and the nginx locations must agree on the prefix the deployment
|
|
# serves this under: "/" at a domain root, "/dev/" behind a path prefix.
|
|
ARG VITE_ROUTER_BASE_PATH=/
|
|
ENV VITE_ROUTER_BASE_PATH=${VITE_ROUTER_BASE_PATH}
|
|
|
|
# `CI=true` turns on the release-provenance gate (scripts/lib/build-environment.ts),
|
|
# which refuses to build without an identity for the artifact. That is the point:
|
|
# a deployed bundle that cannot say which commit it came from is not traceable,
|
|
# and the checklist asks exactly that. Supplied as build args so the caller —
|
|
# a pipeline or the deploy script — owns the values.
|
|
ENV CI=true
|
|
ARG VITE_BUILD_ID
|
|
ARG VITE_COMMIT_SHA
|
|
ARG RELEASE_ID
|
|
ARG CI_RUNNER_IMAGE
|
|
ARG SOURCE_DATE_EPOCH
|
|
ENV VITE_BUILD_ID=${VITE_BUILD_ID}
|
|
ENV VITE_COMMIT_SHA=${VITE_COMMIT_SHA}
|
|
ENV RELEASE_ID=${RELEASE_ID}
|
|
ENV CI_RUNNER_IMAGE=${CI_RUNNER_IMAGE}
|
|
ENV SOURCE_DATE_EPOCH=${SOURCE_DATE_EPOCH}
|
|
|
|
# The two values a deployment is allowed to supply (scripts/generate-runtime-
|
|
# config.ts OVERRIDES); everything else is fixed by the profile. API_BASE_URL
|
|
# has to be absolute — the runtime canonicalises it with `new URL(value)` — so
|
|
# even a same-origin deployment names its own origin here. The committed
|
|
# production profile ships a placeholder (https://api.example.com/), which is
|
|
# what a deployment that forgets this would silently serve.
|
|
ARG RUNTIME_API_BASE_URL
|
|
ARG RUNTIME_TELEMETRY_ENDPOINT
|
|
ENV RUNTIME_API_BASE_URL=${RUNTIME_API_BASE_URL}
|
|
ENV RUNTIME_TELEMETRY_ENDPOINT=${RUNTIME_TELEMETRY_ENDPOINT}
|
|
|
|
RUN corepack enable
|
|
|
|
# Dependencies first so a source-only change does not re-resolve them.
|
|
COPY package.json pnpm-lock.yaml ./
|
|
RUN corepack pnpm install --frozen-lockfile --ignore-scripts
|
|
|
|
COPY . .
|
|
RUN corepack pnpm build \
|
|
&& node scripts/generate-nginx-config.ts
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# runtime
|
|
# ---------------------------------------------------------------------------
|
|
# Unprivileged NGINX 1.27.5 on Alpine, the same release and distribution as the
|
|
# previous nginx base: the master runs as uid 101 and listens on 8080, so the
|
|
# container needs no root and no capability. Kubernetes runs it with
|
|
# runAsNonRoot and capabilities drop ALL (deploy/production/frontend-nonroot).
|
|
FROM nginxinc/nginx-unprivileged@sha256:65e3e85dbaed8ba248841d9d58a899b6197106c23cb0ff1a132b7bfe0547e4c0 AS runtime
|
|
|
|
# The file edits below need root; the image drops back to uid 101 at the end.
|
|
USER root
|
|
|
|
# Replaces the packaged default server block; the generated file is the whole
|
|
# server definition, including the BFF proxy locations.
|
|
RUN rm /etc/nginx/conf.d/default.conf
|
|
COPY --from=build /src/dist/nginx.conf /etc/nginx/conf.d/tech-log.conf
|
|
COPY --from=build /src/dist/ /usr/share/nginx/html/
|
|
|
|
# The generated config is served from /usr/share/nginx/html as root, so the two
|
|
# copies above would also publish nginx.conf itself. It is not secret, but it is
|
|
# not a page either.
|
|
RUN rm -f /usr/share/nginx/html/nginx.conf /usr/share/nginx/html/server.mjs \
|
|
&& rm -rf /usr/share/nginx/html/.vite \
|
|
# The build writes config.json 0600, which nginx (running as uid 101) cannot
|
|
# read — the container came up healthy and answered 403 for the one file the
|
|
# SPA needs before it can boot. Normalise what is served to world-readable.
|
|
&& chmod -R a+rX /usr/share/nginx/html
|
|
|
|
USER 101
|
|
|
|
EXPOSE 8080
|
|
|
|
# No `nginx -t` here: proxy_pass names are resolved when the config loads, and
|
|
# `backend`/`keycloak` only exist on the compose network. The container's own
|
|
# startup is the check, and it fails loudly.
|
|
|
|
HEALTHCHECK --interval=30s --timeout=5s --start-period=10s --retries=3 \
|
|
CMD wget --no-verbose --tries=1 --spider http://127.0.0.1:8080${VITE_ROUTER_BASE_PATH:-/}config.json || exit 1
|