fix(tests/cpu-monitor): handle errors in case process or thread dies

The logic I've added to monitor the CPU wasn't handling the error when
the process or thread died.
This change adds the required error handling. In case a process
terminates, the get_cpu_times will return an empty dict and the
get_cpu_utilization will also return an empty dict. In case a thread
dies, the cpu usage of that thread will be omitted from get_cpu_times
and get_cpu_utilization (due to the set intersection). In all cases,
track_cpu_utilization will only add to the list of values the threads
that are returned, if any.

Signed-off-by: Riccardo Mancini <[email protected]>

Author: Manciukic

tests/framework/utils.py

@@ -58,18 +58,35 @@ def set_cpu_affinity(pid: int, cpulist: list) -> list:
 
 def get_thread_name(pid: int, tid: int) -> str:
     """Return thread name from pid and tid pair."""
-    return Path("/proc", str(pid), "task", str(tid), "comm").read_text("utf-8").strip()
+    try:
+        return (
+            Path("/proc", str(pid), "task", str(tid), "comm").read_text("utf-8").strip()
+        )
+    except FileNotFoundError as exc:
+        raise psutil.NoSuchProcess(tid) from exc
 
 
 CpuTimes = namedtuple("CpuTimes", ["user", "system"])
 
 
 def get_cpu_times(pid: int) -> Dict[str, CpuTimes]:
     """Return a dict mapping thread name to CPU usage (in seconds) since start."""
+    threads = []
+    try:
+        threads = psutil.Process(pid).threads()
+    except psutil.NoSuchProcess as exc:
+        logging.warning("Process %d does not exist", pid, exc_info=exc)
+        return {}
+
     cpu_times = {}
-    for thread in psutil.Process(pid).threads():
-        thread_name = get_thread_name(pid, thread.id)
-        cpu_times[thread_name] = CpuTimes(thread.user_time, thread.system_time)
+    for thread in threads:
+        try:
+            thread_name = get_thread_name(pid, thread.id)
+            cpu_times[thread_name] = CpuTimes(thread.user_time, thread.system_time)
+        except psutil.NoSuchProcess as exc:
+            logging.warning("Thread %d no longer exists", thread.id, exc_info=exc)
+            continue
+
     return cpu_times