fix(e2e): run binary service tests with sudo -E pv

[munezaclovis]

Fixes #67

Previous E2E phases use sudo -E pv start, which causes cleanSitesDirs() to recreate ~/.pv/config/sites/ as root-owned (0755). When s3-binary.sh and mail-binary.sh then ran pv start without sudo, the non-root daemon called os.RemoveAll on the root-owned sites dir — which failed with permission denied — causing GenerateAllConfigs to abort, Start() to exit, and the PID file to be cleaned up. server.IsRunning() then returned false, service:add s3/mail printed "daemon not running", and the stale daemon-status.json from Phase 19 (empty supervised) was what the test inspected.

Use sudo -E pv for all daemon and service commands in both binary-service scripts, consistent with every other E2E phase that needs a live daemon. Also increase the post-start sleep from 3 s to 8 s to match the pattern used in start-curl.sh / verify-final.sh.

Generated with Claude Code

[munezaclovis]

@claude CI still failing. plz check

[claude]

Claude finished @munezaclovis's task in 4m 17s —— View job

---

CI Fix Pushed

• Check CI status and identify failing jobs
• Read failing logs — E2E — S3 binary service lifecycle fails at step 24
• Diagnose root cause
• Fix and push

Root Cause

service:add s3 calls server.SignalDaemon() which sends SIGHUP asynchronously — it returns immediately, before the daemon has run Reconcile() and written the updated daemon-status.json. The test then checked the file immediately and found supervised: {} (the initial snapshot from pv start):

FAIL: daemon-status.json does not contain rustfs entry
{ "pid": 7451, "supervised": {} }

Fix (commit 66cc096)

Replaced the immediate grep check with a 20-iteration / 1 s retry loop in both s3-binary.sh and mail-binary.sh — the same pattern already used for the port-reachability checks in those scripts:

for i in $(seq 1 20); do
    if grep -q '"rustfs"' "$HOME/.pv/daemon-status.json" 2>/dev/null; then break; fi
    sleep 1
done
grep -q '"rustfs"' "$HOME/.pv/daemon-status.json" 2>/dev/null || { ... }

This gives the daemon up to 20 s to finish its SIGHUP-triggered reconcile and flush the updated snapshot before the assertion fires.
| Branch

[munezaclovis]

@claude progress! we have a new error

Run scripts/e2e/start-curl.sh
          
   ERROR  
          
  Pv is already running (PID file exists and process is alive).                                                       

==> pv status

  pv vdev

  ● Stopped

  TLD  .test
  DNS  127.0.0.1:10053  HTTPS  :443
  PHP  8.3 · 8.4 (default)

  Sites  5 linked

  ╭──────────────────────────┬────────────────┬─────╮
  │Site                      │Type            │PHP  │
  ├──────────────────────────┼────────────────┼─────┤
  │ https://e2e-static.test/  │ static         │ 8.4 │
  │ https://e2e-php.test/     │ php            │ 8.4 │
  │ https://e2e-laravel.test/ │ laravel        │ 8.4 │
  │ https://e2e-octane.test/  │ laravel-octane │ 8.4 │
  │ https://e2e-php83.test/   │ php            │ 8.3 │
  ╰──────────────────────────┴────────────────┴─────╯
FAIL: server not running
Error: Process completed with exit code 1.

[claude]

Claude finished @munezaclovis's task in 3m 59s —— View job

---

Fix Pushed

• Check CI status and identify failing run
• Download and analyze failure logs
• Diagnose root cause
• Fix and push (commit 348275e)

Root Cause

pv install auto-starts the daemon immediately (log line 373: ✓ Daemon installed (starts automatically on login) at 20:05:27). The daemon runs as the runner user and writes a PID file.

start-curl.sh called pv daemon:disable 2>/dev/null || true — this unregisters the launchd service but the daemon process doesn't exit synchronously. FrankenPHP + DNS need time to shut down. The sleep 1 wasn't enough; sudo -E pv start found a live PID and returned "already running".

Fix

Added sudo -E pv stop >/dev/null 2>&1 || true between daemon:disable and pv start in start-curl.sh. The stop command sends SIGTERM and polls for up to 5 s (SIGKILL fallback), so by the time it returns the process is gone and the PID file is cleaned up.

 # Disable daemon if auto-enabled during install — e2e tests use foreground mode with sudo.
 pv daemon:disable 2>/dev/null || true
+# pv install starts the daemon immediately; stop it so the PID file is gone before we start fresh.
+sudo -E pv stop >/dev/null 2>&1 || true
 sleep 1
 
 sudo -E pv start &

| Branch

[munezaclovis]

@claude Next:

Run scripts/e2e/mail-binary.sh
==> Phase: Mail binary service (Mailpit) lifecycle
FAIL: pv link for env test
Error: Process completed with exit code 1.

[claude]

Claude finished @munezaclovis's task in 1m 4s —— View job

---

Fix Pushed

• Check CI failure — E2E — Mail binary service lifecycle fails at step 25
• Read mail-binary.sh to diagnose
• Fix and push (commit d6d0b75)

Root Cause

mail-binary.sh starts the daemon with sudo -E pv start, which creates root-owned config dirs (~/.pv/config/sites/). The pv link call on line 26 ran without sudo, so the non-root process couldn't write to the root-owned registry — the command failed silently and triggered FAIL: pv link for env test.

The cleanup trap already correctly used sudo -E pv unlink, but the forward link call was missing sudo -E.

Fix

-pv link "$ENVTEST_DIR" --name e2e-mail-env >/dev/null 2>&1 || { echo "FAIL: pv link for env test"; exit 1; }
+sudo -E pv link "$ENVTEST_DIR" --name e2e-mail-env >/dev/null 2>&1 || { echo "FAIL: pv link for env test"; exit 1; }

| Branch