feat: add guest IP retrieval functions and enhance VM preparation logic

src/ci_orchestrator/commands/template.py

@@ -827,9 +827,34 @@ def _build_seed_iso(staging_dir: Path, iso_path: Path) -> None:
     )
 
 
+def _read_guestinfo_ip(vmrun: str, vmx: str) -> str | None:
+    """Read guestinfo.ci-ip written by ci-report-ip service inside the guest."""
+    try:
+        result = subprocess.run(
+            [vmrun, "readVariable", vmx, "runtimeConfig", "guestinfo.ci-ip"],
+            capture_output=True,
+            text=True,
+            timeout=10,
+            check=False,
+        )
+        ip = result.stdout.strip()
+        if re.match(r"^\d{1,3}\.\d{1,3}\.\d{1,3}\.\d{1,3}$", ip):
+            return ip
+    except subprocess.TimeoutExpired:
+        pass
+    return None
+
+
 def _poll_guest_ip_deploy(vmrun: str, vmx: str, timeout: float) -> str | None:
     deadline = time.monotonic() + timeout
     while time.monotonic() < deadline:
+        # Prefer guestinfo.ci-ip written by ci-report-ip service (faster, no
+        # VMware Tools DHCP heuristic); fall back to getGuestIPAddress for
+        # first-boot VMs that don't have ci-report-ip installed yet.
+        ip = _read_guestinfo_ip(vmrun, vmx)
+        if ip:
+            click.echo(f"  ... guest IP from guestinfo.ci-ip: {ip}")
+            return ip
         try:
             result = subprocess.run(
                 [vmrun, "getGuestIPAddress", vmx],
@@ -1218,7 +1243,7 @@ def template_deploy_linux(
             f'numvcpus = "{vcpu_count}"',
             f'cpuid.coresPerSocket = "{cores_per_socket}"',
             'scsi0.present = "TRUE"',
-            'scsi0.virtualDev = "lsilogic"',
+            'scsi0.virtualDev = "pvscsi"',
             'scsi0:0.present = "TRUE"',
             'scsi0:0.fileName = "LinuxBuild2404.vmdk"',
             'scsi0:0.deviceType = "scsi-hardDisk"',
@@ -1247,8 +1272,14 @@ def template_deploy_linux(
             'floppy0.present = "FALSE"',
             'usb.present = "FALSE"',
             'sound.present = "FALSE"',
+            'mks.enable3d = "FALSE"',
+            'svga.vramSize = "4194304"',
             'tools.syncTime = "TRUE"',
             'uuid.action = "create"',
+            'sched.mem.pshare.enable = "FALSE"',
+            'MemAllowAutoScaleDown = "FALSE"',
+            'mainMem.useNamedFile = "FALSE"',
+            'isolation.tools.hgfs.disable = "TRUE"',
         ]
         if guest_user_data_b64 and guest_meta_data_b64:
             vmx_lines += [
@@ -1572,6 +1603,18 @@ def template_prepare_linux(
             backend.start(handle, headless=True)
         else:
             click.echo(f"[prepare-linux] VM already running: {vmx_path.name}")
+            # VM process alive but guest may be powered off (e.g. after a
+            # previous prepare run that shut down the guest).  Do a quick IP
+            # probe; if nothing comes back, stop and cold-boot the VM.
+            if not guest_ip:
+                click.echo("[prepare-linux] Quick IP probe (30s) ...")
+                quick_ip = _poll_guest_ip_deploy(vmrun, str(vmx_path), timeout=30.0)
+                if not quick_ip:
+                    click.echo(
+                        "[prepare-linux] Guest not responding — stopping and restarting VM ..."
+                    )
+                    backend.stop(handle)
+                    backend.start(handle, headless=True)
 
         if not guest_ip:
             click.echo("[prepare-linux] Detecting guest IP (vmrun getGuestIPAddress) ...")

tests/python/test_commands_template.py

@@ -1148,7 +1148,7 @@ def test_prepare_win_cleanup_only_fails(
 # template deploy-linux — helper unit tests
 # ═══════════════════════════════════════════════════════════════════════════
 
-from ci_orchestrator.config import BackendConfig, Config, IpPoolConfig, Paths  # noqa: E402
+from ci_orchestrator.config import BackendConfig, Config, Paths  # noqa: E402
 
 
 def _deploy_config(tmp_path: Path) -> Config:
@@ -1292,9 +1292,60 @@ def test_build_seed_iso_no_tool_raises(
         tmpl_mod._build_seed_iso(tmp_path, tmp_path / "seed.iso")
 
 
+# ── _read_guestinfo_ip ────────────────────────────────────────────────────────
+
+
+def test_read_guestinfo_ip_valid(monkeypatch: pytest.MonkeyPatch) -> None:
+    import subprocess as _sub
+
+    monkeypatch.setattr(
+        tmpl_mod.subprocess, "run",
+        lambda cmd, **kw: _sub.CompletedProcess(cmd, 0, "10.0.0.5\n", ""),
+    )
+    assert tmpl_mod._read_guestinfo_ip("/vmrun", "/vm.vmx") == "10.0.0.5"
+
+
+def test_read_guestinfo_ip_empty(monkeypatch: pytest.MonkeyPatch) -> None:
+    import subprocess as _sub
+
+    monkeypatch.setattr(
+        tmpl_mod.subprocess, "run",
+        lambda cmd, **kw: _sub.CompletedProcess(cmd, 0, "", ""),
+    )
+    assert tmpl_mod._read_guestinfo_ip("/vmrun", "/vm.vmx") is None
+
+
+def test_read_guestinfo_ip_timeout(monkeypatch: pytest.MonkeyPatch) -> None:
+    import subprocess as _sub
+
+    def raise_timeout(cmd: list[str], **kw: object) -> _sub.CompletedProcess[str]:
+        raise _sub.TimeoutExpired(cmd, 10)
+
+    monkeypatch.setattr(tmpl_mod.subprocess, "run", raise_timeout)
+    assert tmpl_mod._read_guestinfo_ip("/vmrun", "/vm.vmx") is None
+
+
 # ── _poll_guest_ip_deploy ─────────────────────────────────────────────────────
 
 
+def test_poll_guest_ip_uses_guestinfo_first(monkeypatch: pytest.MonkeyPatch) -> None:
+    """If ci-report-ip has written guestinfo.ci-ip, return it without getGuestIPAddress."""
+    called: list[list[str]] = []
+    import subprocess as _sub
+
+    def fake_run(cmd: list[str], **kw: object) -> _sub.CompletedProcess[str]:
+        called.append(list(cmd))
+        if "readVariable" in cmd:
+            return _sub.CompletedProcess(cmd, 0, "10.0.1.50", "")
+        return _sub.CompletedProcess(cmd, 0, "", "")
+
+    monkeypatch.setattr(tmpl_mod.subprocess, "run", fake_run)
+    monkeypatch.setattr(tmpl_mod.time, "sleep", lambda _: None)
+    result = tmpl_mod._poll_guest_ip_deploy("/vmrun", "/vm.vmx", 60.0)
+    assert result == "10.0.1.50"
+    assert not any("getGuestIPAddress" in " ".join(c) for c in called)
+
+
 def test_poll_guest_ip_returns_valid(monkeypatch: pytest.MonkeyPatch) -> None:
     import subprocess as _sub
 
@@ -1492,7 +1543,7 @@ def test_deploy_linux_step5_vmx_content(
         lambda cmd, **kw: _sub.CompletedProcess(list(cmd), 0, "Total running VMs: 0\n", ""),
     )
 
-    result = CliRunner().invoke(
+    CliRunner().invoke(
         cli,
         [
             "template", "deploy-linux",
@@ -1522,7 +1573,7 @@ def test_deploy_linux_step5_guestinfo_skipped_when_step2_skipped(
         lambda cmd, **kw: _sub.CompletedProcess(list(cmd), 0, "Total running VMs: 0\n", ""),
     )
 
-    result = CliRunner().invoke(
+    CliRunner().invoke(
         cli,
         [
             "template", "deploy-linux",
@@ -1956,6 +2007,43 @@ def test_prepare_linux_happy_path_with_vmx(
     assert "192.168.1.200" in result.output
 
 
+def test_prepare_linux_zombie_vm_restarts(
+    prepare_env: dict, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """VM process alive but guest powered off → quick probe fails → stop+restart."""
+    _stub_prepare(monkeypatch, prepare_env)
+
+    # First _poll call (quick probe, 30s) returns None; second (full 300s) returns IP.
+    poll_calls: list[float] = []
+
+    def fake_poll(vmrun: str, vmx: str, timeout: float) -> str | None:
+        poll_calls.append(timeout)
+        return None if timeout == 30.0 else "192.168.1.200"
+
+    monkeypatch.setattr(tmpl_mod, "_poll_guest_ip_deploy", fake_poll)
+
+    backend_m = MagicMock()
+    backend_m.is_running.side_effect = [True, True, False]  # already running; post-stop checks
+    backend_m.list_snapshots.return_value = []
+    monkeypatch.setattr(tmpl_mod, "WorkstationVmrunBackend", lambda **kw: backend_m)
+
+    import subprocess as _sub
+    monkeypatch.setattr(
+        tmpl_mod.subprocess, "run",
+        lambda cmd, **kw: _sub.CompletedProcess(list(cmd), 0, "", ""),
+    )
+
+    result = CliRunner().invoke(
+        cli,
+        ["template", "prepare-linux", "--vmx", str(prepare_env["vmx"])],
+    )
+    assert result.exit_code == 0, result.output
+    assert "restarting" in result.output
+    backend_m.stop.assert_called_once()
+    backend_m.start.assert_called()
+    assert 30.0 in poll_calls
+
+
 def test_prepare_linux_happy_path_skip_flags(
     prepare_env: dict, monkeypatch: pytest.MonkeyPatch
 ) -> None: