chore(make): clean make up / make down output

Three fixes for the noisy shutdown surfaced after wiring web into make up:

- start.sh: replace `sh -c 'cd web && pnpm dev'` with `pnpm -C web dev`.
  Now WEB_PID is pnpm's PID directly; pnpm forwards SIGTERM to its child
  `next dev` on shutdown, so the PID-file kill is a complete shutdown
  for the happy path (no orphaned children for pkill to catch).
- stop.sh: sleep 0.5s after the parent kill before running pkill, so
  the parent has time to propagate SIGTERM to children. Quiet shutdowns
  now stay quiet — the "(also killed leftover process...)" lines
  only print when the fallback was genuinely needed.
- .env.example: NEXT_TELEMETRY_DISABLED=1. Next.js was spawning a
  detached-flush.js helper after `make down` to phone home telemetry;
  this prevents that.

Verified: `make up` → all 3 services 200; `make down` → 3 "stopped"
lines + docker compose down + zero leftover processes (no telemetry
helper, no next workers).

Author: mayur-tolexo

.env.example

@@ -19,3 +19,8 @@ SHOPLIT_API_ADDR=:8080
 
 # ─── shoplit-redirect ──────────────────────────────────────────────────
 SHOPLIT_REDIRECT_ADDR=:8081
+
+# ─── shoplit-web (Next.js) ─────────────────────────────────────────────
+# Disable Next.js usage telemetry so it doesn't spawn a detached-flush.js
+# helper after `make down`. https://nextjs.org/telemetry
+NEXT_TELEMETRY_DISABLED=1

scripts/start.sh

@@ -32,6 +32,7 @@ done
 # the PID file went missing (shell crashed before stop.sh ran, etc.).
 pkill -f "bin/shoplit-api" 2>/dev/null || true
 pkill -f "bin/shoplit-redirect" 2>/dev/null || true
+pkill -f "pnpm.*-C.*web.*dev" 2>/dev/null || true
 pkill -f "node.*next/dist/bin/next" 2>/dev/null || true
 pkill -f "next-server" 2>/dev/null || true
 sleep 0.3
@@ -51,8 +52,10 @@ nohup ./bin/shoplit-redirect >"${LOGS_DIR}/redirect.log" 2>&1 &
 REDIRECT_PID=$!
 echo "$REDIRECT_PID" > "${PIDS_DIR}/redirect.pid"
 
-# Next.js dev runs from web/. Subshell so the cd doesn't leak.
-nohup sh -c 'cd web && pnpm dev' >"${LOGS_DIR}/web.log" 2>&1 &
+# Use pnpm's -C flag instead of a sh wrapper so the PID we track is pnpm's
+# (which forwards signals to its child `next dev` on SIGTERM), not a transient
+# sh process that would be orphaned by `kill $WEB_PID`.
+nohup pnpm -C web dev >"${LOGS_DIR}/web.log" 2>&1 &
 WEB_PID=$!
 echo "$WEB_PID" > "${PIDS_DIR}/web.pid"
 

scripts/stop.sh

@@ -7,6 +7,9 @@ set -euo pipefail
 PIDS_DIR=.pids
 stopped_any=0
 
+# Phase 1: kill the tracked parent PIDs from .pids/. pnpm forwards SIGTERM to
+# its child `next dev`, which kills its own workers — so this should be a
+# complete shutdown for the happy path.
 for svc in api redirect web; do
   pidfile="${PIDS_DIR}/${svc}.pid"
   if [[ -f "$pidfile" ]]; then
@@ -20,23 +23,19 @@ for svc in api redirect web; do
   fi
 done
 
-# Fallback: kill anything matching binary/process names
-if pkill -f "bin/shoplit-api" 2>/dev/null; then
-  echo "stopped stray shoplit-api process(es) via pkill"
-  stopped_any=1
-fi
-if pkill -f "bin/shoplit-redirect" 2>/dev/null; then
-  echo "stopped stray shoplit-redirect process(es) via pkill"
-  stopped_any=1
-fi
-if pkill -f "node.*next/dist/bin/next" 2>/dev/null; then
-  echo "stopped stray next dev process(es) via pkill"
-  stopped_any=1
-fi
-if pkill -f "next-server" 2>/dev/null; then
-  echo "stopped stray next-server process(es) via pkill"
-  stopped_any=1
-fi
+# Give the parent SIGTERMs a moment to propagate to children before we go
+# looking for leftovers — avoids noisy "stray" messages on a clean shutdown.
+sleep 0.5
+
+# Phase 2: defensive fallback. Anything matching these names that's still
+# alive is a real leftover (parent died without cleaning up, PID file lost,
+# etc.). Only print when we actually kill something.
+for pattern in "bin/shoplit-api" "bin/shoplit-redirect" "pnpm.*-C.*web.*dev" "node.*next/dist/bin/next" "next-server"; do
+  if pkill -f "$pattern" 2>/dev/null; then
+    echo "  (also killed leftover process matching: $pattern)"
+    stopped_any=1
+  fi
+done
 
 if [[ $stopped_any -eq 0 ]]; then
   echo "(no shoplit services were running)"