fix(tests/startup): measure test time from spawn instead of start

The process startup time is included in spawn, not start. Measuring the
time it takes to start is incorrect. However, let's keep the API calls
in the measurement as this is used as an upper bound to validate the
reported startup time is meaningful.

Also, remove the 400ms sleep as start already waits for the vm to be up.

Signed-off-by: Riccardo Mancini <mancio@amazon.com>

Author: Manciukic

tests/integration_tests/performance/test_process_startup_time.py

@@ -44,18 +44,17 @@ def test_startup_time_custom_seccomp(
 
 
 def _test_startup_time(microvm, metrics, test_suffix: str):
+    test_start_time = time.time()
     microvm.spawn()
     microvm.basic_config(vcpu_count=2, mem_size_mib=1024)
     metrics.set_dimensions(
         {**microvm.dimensions, "performance_test": f"test_startup_time_{test_suffix}"}
     )
-    test_start_time = time.time()
     microvm.start()
-    time.sleep(0.4)
+    datapoints = microvm.get_all_metrics()
 
-    # The metrics should be at index 1.
+    # The metrics should be at index 0.
     # Since metrics are flushed at InstanceStart, the first line will suffice.
-    datapoints = microvm.get_all_metrics()
     test_end_time = time.time()
     fc_metrics = datapoints[0]
     startup_time_us = fc_metrics["api_server"]["process_startup_time_us"]
@@ -68,7 +67,7 @@ def _test_startup_time(microvm, metrics, test_suffix: str):
     )
 
     assert cpu_startup_time_us > 0
-    # Check that startup time is not a huge value
+    # Check that startup time is not a huge value (overflow)
     # This is to catch issues like the ones introduced in PR
     # https://github.com/firecracker-microvm/firecracker/pull/4305
     test_time_delta_us = (test_end_time - test_start_time) * 1000 * 1000