3 releases 4 tags

RSS feed

• v1.1.0 696a18ddd3

  Compare

  v1.1.0 — Production hardening & first successful APK build on Ubuntu VPS Stable

  zaph0d released this 2026-02-27 02:41:47 +01:00 | 16 commits to main since this release

  📅 Originally published on GitHub: Fri, 27 Feb 2026 01:42:45 GMT
  🏷️ Git tag created: Fri, 27 Feb 2026 01:41:47 GMT

  First successful APK build confirmed ✓

  After deploying to an Intel-based Ubuntu VPS (2 cores, 4 GB RAM), the first complete APK build was produced end-to-end. This release contains the fixes that were required to get there.

  ---

  What changed

  Bug fixes

  nginx: dynamic DNS resolution for the backend proxy

  Without a resolver directive, nginx resolves the backend container's hostname once at startup and caches the IP indefinitely. When the backend container restarts (e.g. after an OOM kill), it often receives a new internal Docker IP — nginx keeps hitting the old address, returning 502 Bad Gateway for every /api/* request. The frontend's own healthcheck routes through the same proxy, so it also fails and the container is marked unhealthy.

  Fix: resolver 127.0.0.11 valid=30s ipv6=off + set $backend_upstream http://backend:3001 variable-based proxy_pass. Nginx now re-resolves the hostname on each request (TTL 30 s) so the proxy recovers automatically after a backend restart without any manual intervention.

  SSE keep-alive heartbeat

  nginx and upstream load-balancers close idle connections after their own read/send timeouts. During a Gradle dependency download on a slow link, the SSE stream can be completely silent for minutes — long enough to be killed. The backend now emits an SSE comment line (: heartbeat) every 15 seconds, keeping the stream alive for the full build duration.

  Gradle JVM heap cap (OOM prevention)

  On a 4 GB host running Docker, an unconstrained Gradle JVM can exhaust available RAM. The Linux OOM-killer then terminates the Node.js backend process, the container restarts, and the nginx DNS-caching issue above turns a temporary blip into a permanent outage. GRADLE_OPTS=-Xmx512m -Xms128m is now set by default — sufficient for a standard TWA release build while leaving enough headroom for the host OS and Docker daemon. Override via GRADLE_OPTS in .env if needed.

  Observability improvements

  HTTP request logging (Morgan)

  docker compose logs backend now shows a timestamped line for every inbound request, making it immediately clear whether requests are reaching the backend at all or being rejected at the nginx proxy level.

  Process-level error handlers

  unhandledRejection and uncaughtException are now logged to stdout before the process exits, surfacing crash reasons in docker compose logs backend for post-mortem analysis.

  ---

  Files changed

  File	Change
  frontend/nginx.conf	Add resolver 127.0.0.11, variable proxy_pass, proxy_send_timeout 600s
  backend/src/routes/build.ts	SSE keep-alive heartbeat every 15 s
  backend/src/index.ts	Morgan request logging, unhandledRejection/uncaughtException handlers
  backend/src/services/builder.ts	GRADLE_OPTS=-Xmx512m -Xms128m default
  backend/package.json	Add morgan dependency
  README.md	Document GRADLE_OPTS, low-memory note, updated feature list
  stats.md	Session 4 log, debug entries 13–16, updated cumulative totals

  ---

  Upgrading

  git pull
  docker compose up -d --build
  

  The frontend image needs a rebuild (nginx.conf changed). The backend image only needs a docker compose restart backend if you don't want to wait for a full rebuild, but rebuilding is recommended to pick up the Morgan dependency.

  Downloads

  • Source code (ZIP)
    0 downloads
  • Source code (TAR.GZ)
    0 downloads