test: harden parallels windows smoke polling

2026-03-27 09:21:35 +07:00 · 2026-03-27 02:14:56 +00:00
parent 3e5d86384e
commit 1ff1679984
2 changed files with 175 additions and 9 deletions
--- a/scripts/e2e/parallels-npm-update-smoke.sh
+++ b/scripts/e2e/parallels-npm-update-smoke.sh
@@ -321,13 +321,59 @@ PY
  prlctl exec "$WINDOWS_VM" --current-user powershell.exe -NoProfile -ExecutionPolicy Bypass -EncodedCommand "$encoded"
 }

+host_timeout_exec() {
+  local timeout_s="$1"
+  shift
+  HOST_TIMEOUT_S="$timeout_s" python3 - "$@" <<'PY'
+import os
+import subprocess
+import sys
+
+timeout = int(os.environ["HOST_TIMEOUT_S"])
+args = sys.argv[1:]
+
+try:
+    completed = subprocess.run(args, stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=timeout)
+except subprocess.TimeoutExpired as exc:
+    if exc.stdout:
+        sys.stdout.buffer.write(exc.stdout)
+    if exc.stderr:
+        sys.stderr.buffer.write(exc.stderr)
+    sys.stderr.write(f"host timeout after {timeout}s\n")
+    raise SystemExit(124)
+
+if completed.stdout:
+    sys.stdout.buffer.write(completed.stdout)
+if completed.stderr:
+    sys.stderr.buffer.write(completed.stderr)
+raise SystemExit(completed.returncode)
+PY
+}
+
+guest_powershell_poll() {
+  local timeout_s="$1"
+  local script="$2"
+  local encoded
+  encoded="$(
+    SCRIPT_CONTENT="$script" python3 - <<'PY'
+import base64
+import os
+
+script = "$ProgressPreference = 'SilentlyContinue'\n" + os.environ["SCRIPT_CONTENT"]
+payload = script.encode("utf-16le")
+print(base64.b64encode(payload).decode("ascii"))
+PY
+  )"
+  host_timeout_exec "$timeout_s" prlctl exec "$WINDOWS_VM" --current-user powershell.exe -NoProfile -ExecutionPolicy Bypass -EncodedCommand "$encoded"
+}
+
 run_windows_script_via_log() {
  local script_url="$1"
  local tgz_url="$2"
  local head_short="$3"
  local session_id="$4"
  local runner_name log_name done_name done_status launcher_state
-  local start_seconds poll_deadline startup_checked
+  local start_seconds poll_deadline startup_checked poll_rc state_rc log_rc
  runner_name="openclaw-update-$RANDOM-$RANDOM.ps1"
  log_name="openclaw-update-$RANDOM-$RANDOM.log"
  done_name="openclaw-update-$RANDOM-$RANDOM.done"
@@ -355,28 +401,55 @@ EOF
 )"

  while :; do
+    set +e
    done_status="$(
-      guest_powershell "\$done = Join-Path \$env:TEMP '$done_name'; if (Test-Path \$done) { (Get-Content \$done -Raw).Trim() }"
+      guest_powershell_poll 20 "\$done = Join-Path \$env:TEMP '$done_name'; if (Test-Path \$done) { (Get-Content \$done -Raw).Trim() }"
    )"
+    poll_rc=$?
+    set -e
    done_status="${done_status//$'\r'/}"
+    if [[ $poll_rc -ne 0 ]]; then
+      warn "windows update helper poll failed; retrying"
+      if (( SECONDS >= poll_deadline )); then
+        warn "windows update helper timed out while polling done file"
+        return 1
+      fi
+      sleep 2
+      continue
+    fi
    if [[ -n "$done_status" ]]; then
-      guest_powershell "\$log = Join-Path \$env:TEMP '$log_name'; if (Test-Path \$log) { Get-Content \$log }"
+      set +e
+      guest_powershell_poll 20 "\$log = Join-Path \$env:TEMP '$log_name'; if (Test-Path \$log) { Get-Content \$log }"
+      log_rc=$?
+      set -e
+      if [[ $log_rc -ne 0 ]]; then
+        warn "windows update helper log drain failed after completion"
+      fi
      [[ "$done_status" == "0" ]]
      return $?
    fi
    if [[ "$startup_checked" -eq 0 && $((SECONDS - start_seconds)) -ge 20 ]]; then
+      set +e
      launcher_state="$(
-        guest_powershell "\$runner = Join-Path \$env:TEMP '$runner_name'; \$log = Join-Path \$env:TEMP '$log_name'; \$done = Join-Path \$env:TEMP '$done_name'; 'runner=' + (Test-Path \$runner) + ' log=' + (Test-Path \$log) + ' done=' + (Test-Path \$done)"
+        guest_powershell_poll 20 "\$runner = Join-Path \$env:TEMP '$runner_name'; \$log = Join-Path \$env:TEMP '$log_name'; \$done = Join-Path \$env:TEMP '$done_name'; 'runner=' + (Test-Path \$runner) + ' log=' + (Test-Path \$log) + ' done=' + (Test-Path \$done)"
      )"
+      state_rc=$?
+      set -e
      launcher_state="${launcher_state//$'\r'/}"
      startup_checked=1
-      if [[ "$launcher_state" == *"runner=False"* && "$launcher_state" == *"log=False"* && "$launcher_state" == *"done=False"* ]]; then
+      if [[ $state_rc -eq 0 && "$launcher_state" == *"runner=False"* && "$launcher_state" == *"log=False"* && "$launcher_state" == *"done=False"* ]]; then
        warn "windows update helper failed to materialize guest files"
        return 1
      fi
    fi
    if (( SECONDS >= poll_deadline )); then
-      guest_powershell "\$log = Join-Path \$env:TEMP '$log_name'; if (Test-Path \$log) { Get-Content \$log }"
+      set +e
+      guest_powershell_poll 20 "\$log = Join-Path \$env:TEMP '$log_name'; if (Test-Path \$log) { Get-Content \$log }"
+      log_rc=$?
+      set -e
+      if [[ $log_rc -ne 0 ]]; then
+        warn "windows update helper log drain failed after timeout"
+      fi
      warn "windows update helper timed out waiting for done file"
      return 1
    fi
--- a/scripts/e2e/parallels-windows-smoke.sh
+++ b/scripts/e2e/parallels-windows-smoke.sh
@@ -315,6 +315,35 @@ guest_exec() {
  prlctl exec "$VM_NAME" --current-user "$@"
 }

+host_timeout_exec() {
+  local timeout_s="$1"
+  shift
+  HOST_TIMEOUT_S="$timeout_s" python3 - "$@" <<'PY'
+import os
+import subprocess
+import sys
+
+timeout = int(os.environ["HOST_TIMEOUT_S"])
+args = sys.argv[1:]
+
+try:
+    completed = subprocess.run(args, stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=timeout)
+except subprocess.TimeoutExpired as exc:
+    if exc.stdout:
+        sys.stdout.buffer.write(exc.stdout)
+    if exc.stderr:
+        sys.stderr.buffer.write(exc.stderr)
+    sys.stderr.write(f"host timeout after {timeout}s\n")
+    raise SystemExit(124)
+
+if completed.stdout:
+    sys.stdout.buffer.write(completed.stdout)
+if completed.stderr:
+    sys.stderr.buffer.write(completed.stderr)
+raise SystemExit(completed.returncode)
+PY
+}
+
 guest_powershell() {
  local script="$1"
  local encoded
@@ -331,6 +360,23 @@ PY
  guest_exec powershell.exe -NoProfile -ExecutionPolicy Bypass -EncodedCommand "$encoded"
 }

+guest_powershell_poll() {
+  local timeout_s="$1"
+  local script="$2"
+  local encoded
+  encoded="$(
+    SCRIPT_CONTENT="$script" python3 - <<'PY'
+import base64
+import os
+
+script = "$ProgressPreference = 'SilentlyContinue'\n" + os.environ["SCRIPT_CONTENT"]
+payload = script.encode("utf-16le")
+print(base64.b64encode(payload).decode("ascii"))
+PY
+  )"
+  host_timeout_exec "$timeout_s" prlctl exec "$VM_NAME" --current-user powershell.exe -NoProfile -ExecutionPolicy Bypass -EncodedCommand "$encoded"
+}
+
 guest_run_openclaw() {
  local env_name="${1:-}"
  local env_value="${2:-}"
@@ -779,11 +825,15 @@ verify_version_contains() {
 }

 run_ref_onboard() {
-  local openai_key_q runner_name log_name done_name done_status
+  local openai_key_q runner_name log_name done_name done_status launcher_state
+  local poll_rc state_rc log_rc start_seconds poll_deadline startup_checked
  openai_key_q="$(ps_single_quote "$OPENAI_API_KEY_VALUE")"
  runner_name="openclaw-onboard-$RANDOM-$RANDOM.ps1"
  log_name="openclaw-onboard-$RANDOM-$RANDOM.log"
  done_name="openclaw-onboard-$RANDOM-$RANDOM.done"
+  start_seconds="$SECONDS"
+  poll_deadline=$((SECONDS + TIMEOUT_ONBOARD_S + 60))
+  startup_checked=0

  guest_powershell "$(cat <<EOF
 \$runner = Join-Path \$env:TEMP '$runner_name'
@@ -815,15 +865,58 @@ EOF
 )"

  while :; do
+    set +e
    done_status="$(
-      guest_powershell "\$done = Join-Path \$env:TEMP '$done_name'; if (Test-Path \$done) { (Get-Content \$done -Raw).Trim() }"
+      guest_powershell_poll 20 "\$done = Join-Path \$env:TEMP '$done_name'; if (Test-Path \$done) { (Get-Content \$done -Raw).Trim() }"
    )"
+    poll_rc=$?
+    set -e
    done_status="${done_status//$'\r'/}"
+    if [[ $poll_rc -ne 0 ]]; then
+      warn "windows onboard helper poll failed; retrying"
+      if (( SECONDS >= poll_deadline )); then
+        warn "windows onboard helper timed out while polling done file"
+        return 1
+      fi
+      sleep 2
+      continue
+    fi
    if [[ -n "$done_status" ]]; then
-      guest_powershell "\$log = Join-Path \$env:TEMP '$log_name'; if (Test-Path \$log) { Get-Content \$log }"
+      set +e
+      guest_powershell_poll 20 "\$log = Join-Path \$env:TEMP '$log_name'; if (Test-Path \$log) { Get-Content \$log }"
+      log_rc=$?
+      set -e
+      if [[ $log_rc -ne 0 ]]; then
+        warn "windows onboard helper log drain failed after completion"
+      fi
      [[ "$done_status" == "0" ]]
      return $?
    fi
+    if [[ "$startup_checked" -eq 0 && $((SECONDS - start_seconds)) -ge 20 ]]; then
+      set +e
+      launcher_state="$(
+        guest_powershell_poll 20 "\$runner = Join-Path \$env:TEMP '$runner_name'; \$log = Join-Path \$env:TEMP '$log_name'; \$done = Join-Path \$env:TEMP '$done_name'; 'runner=' + (Test-Path \$runner) + ' log=' + (Test-Path \$log) + ' done=' + (Test-Path \$done)"
+      )"
+      state_rc=$?
+      set -e
+      launcher_state="${launcher_state//$'\r'/}"
+      startup_checked=1
+      if [[ $state_rc -eq 0 && "$launcher_state" == *"runner=False"* && "$launcher_state" == *"log=False"* && "$launcher_state" == *"done=False"* ]]; then
+        warn "windows onboard helper failed to materialize guest files"
+        return 1
+      fi
+    fi
+    if (( SECONDS >= poll_deadline )); then
+      set +e
+      guest_powershell_poll 20 "\$log = Join-Path \$env:TEMP '$log_name'; if (Test-Path \$log) { Get-Content \$log }"
+      log_rc=$?
+      set -e
+      if [[ $log_rc -ne 0 ]]; then
+        warn "windows onboard helper log drain failed after timeout"
+      fi
+      warn "windows onboard helper timed out waiting for done file"
+      return 1
+    fi
    sleep 2
  done
 }