test-driver.py: delete VM state directory after test run

Keeping the VM state test across several run sometimes lead to subtle
and hard to spot errors in practice. We delete the VM state which
contains (among other things) the qcow volume.

We also introduce a -K (--keep-vm-state) flag making VM state to
persist after the test run. This flag makes test-driver.py to match
its previous behaviour.

nixos/lib/test-driver/test-driver.py

@@ -4,6 +4,7 @@ from queue import Queue, Empty
 from typing import Tuple, Any, Callable, Dict, Iterator, Optional, List
 from xml.sax.saxutils import XMLGenerator
 import _thread
+import argparse
 import atexit
 import base64
 import codecs
@@ -751,6 +752,11 @@ class Machine:
 
         self.log("QEMU running (pid {})".format(self.pid))
 
+    def cleanup_statedir(self) -> None:
+        self.log("delete the VM state directory")
+        if os.path.isfile(self.state_dir):
+            shutil.rmtree(self.state_dir)
+
     def shutdown(self) -> None:
         if not self.booted:
             return
@@ -889,6 +895,15 @@ def subtest(name: str) -> Iterator[None]:
 
 
 if __name__ == "__main__":
+    arg_parser = argparse.ArgumentParser()
+    arg_parser.add_argument(
+        "-K",
+        "--keep-vm-state",
+        help="re-use a VM state coming from a previous run",
+        action="store_true",
+    )
+    (cli_args, vm_scripts) = arg_parser.parse_known_args()
+
     log = Logger()
 
     vlan_nrs = list(dict.fromkeys(os.environ.get("VLANS", "").split()))
@@ -896,8 +911,10 @@ if __name__ == "__main__":
     for nr, vde_socket, _, _ in vde_sockets:
         os.environ["QEMU_VDE_SOCKET_{}".format(nr)] = vde_socket
 
-    vm_scripts = sys.argv[1:]
     machines = [create_machine({"startCommand": s}) for s in vm_scripts]
+    for machine in machines:
+        if not cli_args.keep_vm_state:
+            machine.cleanup_statedir()
     machine_eval = [
         "{0} = machines[{1}]".format(m.name, idx) for idx, m in enumerate(machines)
     ]
@@ -911,7 +928,6 @@ if __name__ == "__main__":
                     continue
                 log.log("killing {} (pid {})".format(machine.name, machine.pid))
                 machine.process.kill()
-
             for _, _, process, _ in vde_sockets:
                 process.terminate()
         log.close()