Collector crashes on duplicate address + frontend shows no data #214
Description
Issues Found During VPS Migration (2026-02-28)
1. Collector create() → upsert() (Fixed on VPS, needs commit)
File: src/server/workers/collector.ts:401
The discoverNodes function uses db.node.create() which crashes with Unique constraint failed on the fields: (address) when a discovered node has an address already in the database.
Quick fix applied on VPS: Changed to db.node.upsert() with where: { address }, create: {...}, update: {...}.
Needs: Commit the fix to the repo and redeploy via CI.
2. Collector not in Docker image
The collector runs as a bind-mount dev container (node:20 + npm install at runtime from /home/pnodepulse/pnode-pulse-src). This is fragile — the original node:20-alpine image lacked Python for usb native module (node-gyp).
Needs: Add collector as a build target in the Dockerfile (or separate Dockerfile) so it deploys via GHCR like the web app. The docker-compose.prod.yml already has the correct config (command: ["node", "-e", "require('./scripts/start-collector.js')"]) but the GHCR image doesn't include the compiled collector script.
3. Frontend shows "No Data Available" despite 23M+ metrics
- DB has 23,028,956 metrics and 499 active nodes
- Collector runs successfully (669 nodes, ~101 responding per cycle)
- Frontend Overview page shows "No Data Available"
Needs: Investigation — likely an API query issue, Redis cache, or frontend state problem.
4. Nginx blue/green port mismatch
Production nginx was pointing to port 7000 (blue) but active container was on 7001 (green). Fixed manually. Deploy workflow should update nginx to match active slot.