test: Modernize test_send_ctrl_alt_del

Instead of sleeping for 2 seconds to wait for booting, add a network
device and wait for ssh availability (happens automatically in
.start()). Use mark_killed() to assert that Firecracker exited instead
of "waitpid". Using `waitpid(2)` here is wrong, because it only works on
child processes, and Firecracker is no child of the pytest process (it
is daemonized, for starters). The reason we never caught this being
broken is because the test also simply ignores all exceptions raised.

But there's more issues here: The test never caused _killed to be set to
`True`, so if Firecracker really did exit, then we would have been
seeing intermittent failures during microvm teardown (as Microvm.kill()
asserts that the process is still alive if _killed is False). In fact,
our guest kernels do not have the required drivers compiled in to
support CTRL+ALT+DEL (thanks Riccardo for figuring this part out).

Signed-off-by: Patrick Roy <[email protected]>

Author: roypat

tests/integration_tests/functional/test_api.py

@@ -782,24 +782,14 @@ def test_send_ctrl_alt_del(uvm_plain):
     test_microvm.spawn()
 
     test_microvm.basic_config()
+    test_microvm.add_net_iface()
     test_microvm.start()
 
-    # Wait around for the guest to boot up and initialize the user space
-    time.sleep(2)
-
     test_microvm.api.actions.put(action_type="SendCtrlAltDel")
 
-    firecracker_pid = test_microvm.firecracker_pid
-
     # If everything goes as expected, the guest OS will issue a reboot,
     # causing Firecracker to exit.
-    # waitpid should block until the Firecracker process has exited. If
-    # it has already exited by the time we call waitpid, WNOHANG causes
-    # waitpid to raise a ChildProcessError exception.
-    try:
-        os.waitpid(firecracker_pid, os.WNOHANG)
-    except ChildProcessError:
-        pass
+    test_microvm.mark_killed()
 
 
 def _drive_patch(test_microvm):