Skip to content

website.md

Latest commit

 

History

History
90 lines (68 loc) · 3.81 KB

website.md

File metadata and controls

90 lines (68 loc) · 3.81 KB
Raw

diskpush.com

The marketing and documentation site. Next.js App Router, server components throughout, no client JavaScript of its own.

Architecture

  • Static and server-rendered. Every page, including the docs, is readable with JavaScript disabled.
  • Docs are read from this repository's docs/*.md at build time by apps/web/lib/docs.ts, which also rewrites inter-document links so defaults.md resolves to /docs/defaults. There is no second copy of the documentation, so the site cannot drift from the product.
  • Release metadata is fetched server-side and normalised in apps/web/lib/releases.ts, cached for an hour. No database. A repository with no tagged release yet degrades to "not yet published" rather than an error.
  • /api/releases/latest exposes that normalised shape so the download UI never has to know GitHub's response format.

Deployment

Railway, in the shared Profullstack, Inc. project, service diskpush-web.

railway link -w "Profullstack, Inc." -p "Profullstack, Inc." -e production
railway service link diskpush-web
railway up

Build configuration lives in Railway, not in this repo

Railway builds with Railpack, which reads three service variables. They are set on the service and are not in version control, so they are recorded here — a service recreated without them will fail to build:

RAILPACK_INSTALL_CMD = pnpm install --frozen-lockfile --prod=false --filter @diskpush/web...
RAILPACK_BUILD_CMD   = pnpm --filter @diskpush/web build
RAILPACK_START_CMD   = pnpm --filter @diskpush/web start

Three things those commands are doing deliberately:

  • --filter @diskpush/web... keeps the install to the site and its dependencies. A full workspace install would download Electron, which the website has no use for.
  • --prod=false is required because Railway sets NODE_ENV=production, under which pnpm would skip devDependencies — and TypeScript and Tailwind, which the build needs, are devDependencies.
  • The build runs from the repository root, not from apps/web. Setting a service root directory would break the docs pages, which read ../../docs.

.nvmrc pins Node 24; engines.node is a range and not something a builder can pin to.

.railwayignore keeps node_modules and build output out of the upload.

Config as code

Railway deprecated railway.json in favour of .railway/railway.ts (existing files work until 2026-12-01). The migration is not applied here: railway config migrate generates a file describing the whole project's resources, and this is a shared project that also runs other production services. Applying it risks pruning services this repository did not create. The Railpack variables above are service-scoped and carry no such risk.

The installer

/install.sh is a route, not a static file: it reads the repository's own scripts/install.sh and rewrites the site URL baked into it to this deployment's. So the script people pipe into a shell is the one in version control that they can read on GitHub, and an installer served from a preview installs from that preview rather than silently from production.

It is force-static, so the read happens at build time and outputFileTracingIncludes carries the script into the deployment.

Custom domain

diskpush.com and www.diskpush.com are attached and serving.

DNS is at Porkbun. The apex is an ALIAS rather than a CNAME, because a CNAME cannot coexist with the other records at a zone apex; www is an ordinary CNAME, which also has to exist explicitly because the zone has a *.diskpush.com wildcard that would otherwise keep sending it to the parking page. Both have a _railway-verify TXT beside them.

The MX records for Porkbun's email forwarding and the SPF record were left alone; repointing the apex must not take the domain's mail with it.