Do not report memory utilization as zero for missing PID.

Author: gladkova

ci/buildspec.yml

@@ -15,7 +15,8 @@ phases:
       - pip install twine
       - pip install pytest-mock -U
       - pip install requests
-      - pip install -U -e .[mxnet]
+      - pip install -U -e .
+      - pip install mxnet==1.5.0
       - cd model-archiver/ && pip install -U -e . && cd ../
 
   build:

frontend/server/src/main/java/com/amazonaws/ml/mms/metrics/MetricCollector.java

@@ -117,10 +117,14 @@ public void run() {
                     if (tokens.length != 2) {
                         continue;
                     }
-
-                    Integer pid = Integer.valueOf(tokens[0]);
-                    WorkerThread worker = workerMap.get(pid);
-                    worker.setMemory(Long.parseLong(tokens[1]));
+                    try {
+                        Integer pid = Integer.valueOf(tokens[0]);
+                        WorkerThread worker = workerMap.get(pid);
+                        worker.setMemory(Long.parseLong(tokens[1]));
+                    } catch (NumberFormatException e) {
+                        logger.warn("Failed to parse memory utilization metrics: " + line);
+                        continue;
+                    }
                 }
             }
         } catch (IOException e) {

mms/metrics/process_memory_metric.py

@@ -17,7 +17,6 @@
 
 import psutil
 
-
 def get_cpu_usage(pid):
     """
     use psutil for cpu memory
@@ -27,10 +26,13 @@ def get_cpu_usage(pid):
     try:
         process = psutil.Process(int(pid))
     except psutil.Error:
-        logging.error("Failed get process for pid: %s", pid, exc_info=True)
+        logging.error("Failed to get process for pid: %s", pid, exc_info=True)
         return 0
 
     mem_utilization = process.memory_info()[0]
+    if mem_utilization == 0:
+        logging.error("Failed to get memory utilization for pid: %s", pid, exc_info=True)
+        return 0
     return mem_utilization
 
 
@@ -45,4 +47,6 @@ def check_process_mem_usage(stdin):
     for process in process_list:
         if not process:
             continue
-        logging.info("%s:%d", process, get_cpu_usage(process))
+        mem_utilization = get_cpu_usage(process)
+        if mem_utilization != 0:
+            logging.info("%s:%d", process, mem_utilization)