The health check URL was hardcoded to http://127.0.0.1:3000 but the
production .env sets PORT=3002. Read the PORT from .env dynamically
so the health check matches the actual server port.
Changing ownership to www-data at end of deploy breaks permission
handling when pm2 runs as a different user (e.g., root or the deploy
user). Keep ownership as the deploy user throughout.
- pm2 stop now uses --kill-timeout 10000 for graceful shutdown
- Wait 10s after stopping (was 3s) for MariaDB to reclaim connections
- Migration retries increased from 5 to 10 with 5s wait (was 3s)
- Total retry window: ~50s instead of ~15s
Each deploy creates a fresh worktree and deletes .next, losing the
Turbopack build cache. Now restore .next/cache from the live site
into the stage before pnpm build, so Next.js can do incremental
compilation. After cutover the cache persists in the live .next/ for
the next deploy.
- Allow ARGON2_MEMORY_SIZE, ARGON2_ITERATIONS, ARGON2_PARALLELISM env overrides
- Use m=1024,t=1 in tests (was m=65536,t=4 → ~1s per hash)
- Set fast params in CI and deploy workflows
Cut Auth.js DB load with cached jwtVersion checks, surface Redis in /api/health and deploy warnings, add admin help-center ticket reply UI, rate-limit API tickets/reactions/referral claims, and revoke PATs on sign-out-everywhere.
Co-authored-by: Cursor <[email protected]>
Stage migrate hit ER_CON_COUNT_ERROR while live still held the pool. Build in stage with DATABASE_POOL_SIZE=5, run db:migrate only after stopping the service (with retries), and lower the default pool from 40 to 10.
Co-authored-by: Cursor <[email protected]>
Build install/test/migrate in a detached worktree while the live site keeps serving, then swap .next and node_modules during a brief stop. Drops nuclear rm -rf src and rolls back .next.prev on cutover failure.
Co-authored-by: Cursor <[email protected]>
Align nodemailer with Auth.js peers, bump patch deps, validate env on deploy builds, add admin error boundary, and warn when Redis is missing in production.
Co-authored-by: Cursor <[email protected]>
Only clear paths that already have skip-worktree/assume-unchanged; keep nuclear src replace and content verify.
Co-authored-by: Cursor <[email protected]>
Host built a different event-form than HEAD while git looked clean; delete src, restore from git objects, and hash-verify every tracked blob.
Co-authored-by: Cursor <[email protected]>
Host next build still saw Json on nitro while tsc passed; use any helpers, verify blob hash, and delete .next entirely before build.
Co-authored-by: Cursor <[email protected]>
next build failed on host-local rooms.ts using Prisma without import while tsc incremental passed; force non-incremental typecheck and verify src matches HEAD.
Co-authored-by: Cursor <[email protected]>
Stale host sources survived reset when files were owned by www-data, leaving an old nitro editor with a removed Json type that failed typecheck.
Co-authored-by: Cursor <[email protected]>
Rewrite getNested/updateNested without fragile any/Json inference, clear stale .next types before build, and skip env refine during compile.
Co-authored-by: Cursor <[email protected]>
Align onlyBuiltDependencies with the workspace, fail fast without AUTH_SECRET in production, and delete catalog_items via VARCHAR-safe SQL so page deletes do not leave orphans.
Co-authored-by: Cursor <[email protected]>
openhands
SimoandCursor
remco