diff --git a/frontend/server/src/main/java/com/amazonaws/ml/mms/wlm/WorkLoadManager.java b/frontend/server/src/main/java/com/amazonaws/ml/mms/wlm/WorkLoadManager.java
index 4bcfafe38..ba445bc9e 100644
--- a/frontend/server/src/main/java/com/amazonaws/ml/mms/wlm/WorkLoadManager.java
+++ b/frontend/server/src/main/java/com/amazonaws/ml/mms/wlm/WorkLoadManager.java
@@ -147,7 +147,7 @@ private CompletableFuture<HttpResponseStatus> shutdownServerThread(
         Process workerProcess = lifecycle.getProcess();
         if (workerProcess.isAlive()) {
             boolean workerDestroyed = false;
-            workerProcess.destroyForcibly();
+            workerProcess.destroy(); // sends SIGTERM to workerProcess
             try {
                 workerDestroyed =
                         workerProcess.waitFor(
diff --git a/mms/model_service_worker.py b/mms/model_service_worker.py
index 164d463b1..0017fa41d 100644
--- a/mms/model_service_worker.py
+++ b/mms/model_service_worker.py
@@ -186,6 +186,25 @@ def start_worker(self, cl_socket):
                 cl_socket.close()
                 sys.exit(0)
 
+    def server_worker_sigterm_handler(self, signum, frame):
+        # Frontend Process.destroy() sends sigterm
+        logging.info("PID=%s, received signal  %s server worker", os.getpid(), signum)
+        for p in multiprocessing.active_children():
+            logging.info("PID=%s, Killing child %s", os.getpid(), p.pid)
+            p.terminate()
+
+        if self.sock_type == 'unix' and os.path.exists(self.sock_name):
+            os.remove(self.sock_name)
+
+        logging.info("PID=%s, Sending self kill signal", os.getpid())
+        os.kill(os.getpid(), 9)
+
+    def sigchld_handler(self, signum, frame):
+        # This is to handle zombie processes.
+        # Calling `active_children` has the side effect of `joining` any processes which have already finished
+        val = len(multiprocessing.active_children())
+
+
     def run_server(self):
         """
         Run the backend worker process and listen on a socket
@@ -200,6 +219,8 @@ def run_server(self):
         logging.info("[PID] %d", os.getpid())
         logging.info("MMS worker started.")
         logging.info("Python runtime: %s", platform.python_version())
+        signal.signal(signal.SIGCHLD, self.sigchld_handler)
+        signal.signal(signal.SIGTERM, self.server_worker_sigterm_handler)
         while True:
             if self.service is None and self.preload is True:
                 # Lazy loading the models
diff --git a/tests/performance/agents/metrics/__init__.py b/tests/performance/agents/metrics/__init__.py
index 9d7bf7eb2..756051a9c 100644
--- a/tests/performance/agents/metrics/__init__.py
+++ b/tests/performance/agents/metrics/__init__.py
@@ -92,7 +92,7 @@ def get_metrics(server_process, child_processes, logger):
     """
     result = {}
     children.update(child_processes)
-    logger.debug("children : {0}".format(",".join([str(c.pid) for c in children])))
+    logger.info("children : {0}".format(",".join([str(c.pid) for c in children])))
 
     def update_metric(metric_name, proc_type, stats):
         stats = list(filter(lambda x: isinstance(x, (float, int)), stats))
@@ -125,10 +125,12 @@ def update_metric(metric_name, proc_type, stats):
                 processes_stats.append({'type': ProcessType.WORKER, 'stats': child.as_dict()})
             else:
                 reclaimed_pids.append(child)
-                logger.debug('child {0} no longer available'.format(child.pid))
+                logger.info('child {0} no longer available'.format(child.pid))
         except (NoSuchProcess, ZombieProcess):
             reclaimed_pids.append(child)
-            logger.debug('child {0} no longer available'.format(child.pid))
+            logger.info('child {0} no longer available'.format(child.pid))
+        except psutil.AccessDenied as e:
+            logger.info('child {0} threw access denied {1}'.format(child.pid, str(e)))
 
     for p in reclaimed_pids:
         children.remove(p)