Host port 3000 is Gitea (git.soroushasadi.com proxies to :3000). The
portfolio was publishing 3000:3000 AND the deploy had a "Free Port 3000"
step that force-removed every container on :3000 — which evicted the
Gitea container.
- compose: publish 3020:3000 instead of 3000:3000
- deploy: delete the "Free Port 3000" step entirely; compose recreates
only our own named container and must never touch other stacks
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The previous name-based stop only removed soroushasadi-site, but the
container holding :3000 can have a different name (old Next.js build or
an orphan from a previously-named compose project), so the bind kept
failing. Now we remove every container publishing :3000 by filter, then
also remove by our known name as a fallback.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Port 3000 was already allocated by the previous container from the
Next.js era. --remove-orphans only removes containers within the same
compose project, so the old one survived. Explicitly stop+rm the named
container before docker compose up so the port is always free.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
.gitignore has '/data' which Windows git (case-insensitive) silently
matched '/Data/', so AppDbContext.cs was never committed and the Docker
build (Linux, case-sensitive) failed with CS0234 'Data' not found.
Renaming the directory to 'Database/' sidesteps the collision.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Full rewrite of the portfolio site from Next.js 14 to .NET 10:
- ASP.NET Core 10 Razor Pages, no Node.js dependency
- EF Core 10 + SQLite (same schema as before — data survives upgrade)
- Cookie authentication (same single-password model)
- Resend contact form via HttpClient
- Bilingual FA/EN via locale cookie + BasePageModel
- All UI ported to Razor Pages with Tailwind CDN + custom CSS
- Vanilla JS: particles, typewriter, cursor, animations, portfolio modal
- Dockerfile: SDK 10.0-alpine → aspnet 10.0-alpine (no npm/Node needed)
- CI/CD: dropped NPM_TOKEN, ADMIN_SESSION_SECRET — pure dotnet publish
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The npm ci layer was cached with a broken result (node_modules/next missing).
Changing the RUN command text invalidates that cache entry and forces a fresh
install. The added post-install check will show the exact npm ci error if next
is still missing, instead of failing silently.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace `npm run build` with `node node_modules/next/dist/bin/next build`
to bypass PATH/symlink resolution issues on Alpine. Also adds a diagnostic
`ls node_modules/.bin/next` so CI logs show whether the binary is present.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
The cross-stage COPY --from=deps /app/node_modules caused
sh: next: not found in the builder stage. Fix by merging deps
and builder into one stage — npm ci and next build run in the
same layer so node_modules is always present when building.
Also add libc6-compat (required by Next.js SWC binaries on Alpine)
to both builder and runner stages.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Mirror exposes images as mirror.soroushasadi.com/<image> directly,
not via /repository/docker-group/ path. Also node:20-alpine is already
cached on the server; node:20-slim was never pulled.
Dockerfile: apk instead of apt-get, addgroup/adduser instead of
groupadd/useradd (Alpine BusyBox compatibility).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Previous commit introduced a typo (soroushasad vs soroushasadi).
All references now use:
npm -> http://mirror.soroushasadi.com/repository/npm-group/
docker -> mirror.soroushasadi.com/repository/docker-group/ (no protocol in image refs)
Also restore Dockerfile ARG NPM_TOKEN + COPY .npmrc that were lost
when the soroush-cicd skill regenerated the file, and set the
NODE_IMAGE ARG default back to the mirror path.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
docker-compose.yml:
- Add top-level name: soroushasadi so Compose uses a stable project name
regardless of which temp directory the runner checks out into. Without
this, each run gets a different project name and the old container is not
recognised as an orphan — it stays bound to port 3000 and the new
container fails to start.
- Fix healthcheck: wget is not in node:20-slim; use Node 20's built-in
fetch instead.
- Set NODE_IMAGE default to the Nexus docker-group mirror.
Workflows:
- Add GIT_SSL_NO_VERIFY=true to checkout steps (self-signed cert on
git.soroushasadi.com).
- Pass NODE_IMAGE to build step in deploy.
- Add git remote remove guard to CI checkout (prevents failure on
second run in the same workspace).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
soroush.asadi