test(install-e2e): trace early updater exits and marker owner

tests/install/e2e-assets/handoff-trace/sitecustomize.py

@@ -7,9 +7,31 @@ import os
 
 if os.environ.get("GITHUB_ACTIONS") == "true" and os.environ.get("HERMES_E2E_HANDOFF_TRACE"):
     import faulthandler
+    import sys
     from pathlib import Path
 
     _directory = Path(os.environ["HERMES_E2E_HANDOFF_TRACE"])
     _directory.mkdir(parents=True, exist_ok=True)
     _stream = (_directory / f"python-stacks-{os.getpid()}.log").open("a", encoding="utf-8")
+    _stream.write(f"started pid={os.getpid()} parent={os.getppid()} executable={sys.executable}\n")
+    _stream.flush()
     faulthandler.dump_traceback_later(90, repeat=True, file=_stream)
+
+    # Call boundaries locate an early exit before the first timed stack dump.
+    # Never record arguments, frame locals, or return values.
+    _watched = {
+        "cmd_update", "_cmd_update_impl", "_run_pre_update_backup",
+        "_pause_windows_gateways_for_update", "find_gateway_pids",
+        "_scan_gateway_pids", "_get_service_pids", "_install_hangup_protection",
+        "_finalize_update_output", "load_config", "is_installed",
+    }
+
+    def _trace_calls(frame, event, arg):
+        if event in ("call", "return") and frame.f_code.co_name in _watched:
+            _stream.write(f"{event} {frame.f_code.co_filename}:{frame.f_lineno} {frame.f_code.co_name}\n")
+            _stream.flush()
+        elif event == "c_call" and getattr(arg, "__name__", "") in {"kill", "_exit", "abort"}:
+            _stream.write(f"c_call {frame.f_code.co_filename}:{frame.f_lineno} {arg.__name__}\n")
+            _stream.flush()
+
+    sys.setprofile(_trace_calls)

tests/install/e2e-assets/july-handoff-diagnostics.ps1

@@ -59,15 +59,19 @@ Write-Section "meta"
 Write-Output ("utc={0} workroot={1}" -f (Get-Date).ToUniversalTime().ToString("o"), $WorkRoot)
 
 Write-Section "processes"
-$procs = @(Get-CimInstance Win32_Process -ErrorAction SilentlyContinue)
-if ($procs.Count -eq 0) {
-    Write-Output "Get-CimInstance returned nothing"
-}
+$procs = @(Get-CimInstance Win32_Process -ErrorAction Continue)
+Write-Output "CIM process count=$($procs.Count)"
+# The updater PID is authoritative even when its argv uses unnormalized paths.
+$ownerPath = Join-Path $WorkRoot "hermes-home\.hermes-update-in-progress"
+$ownerPid = if (Test-Path $ownerPath) { (Get-Content $ownerPath -First 1).Trim() } else { "" }
+Get-Process -ErrorAction Continue | Where-Object { $_.Id -eq $ownerPid -or $_.Path -like '*hermes-desktop-gui-e2e*' } |
+    Select-Object Id, ProcessName, Path, CPU, WorkingSet64 | Format-List | Out-String | Write-Output
 $rootPids = @{}
 foreach ($p in $procs) {
     $exe = [string]$p.ExecutablePath
     $cmd = [string]$p.CommandLine
-    $ref = ($exe -like "$WorkRoot\*") -or ($cmd -like "*$WorkRoot\*")
+    if ($exe) { $exe = [System.IO.Path]::GetFullPath($exe) }
+    $ref = ($exe -like "$WorkRoot\*") -or ($p.ProcessId -eq $ownerPid)
     if ($ref) { $rootPids[[uint32]$p.ProcessId] = $true }
 }
 # Expand descendants transitively (both directions of interest: children of