add option to remove iperf servers

Signed-off-by: vezio <tyler.rimaldi@ibm.com>

autopilot-daemon/network/iperf3_stop_servers.py

@@ -0,0 +1,70 @@
+from iperf3_utils import *
+
+import subprocess
+import os
+import sys
+import signal
+
+
+def kill_all_iperf_servers():
+    try:
+        result = subprocess.run(
+            ["ps", "aux"], text=True, capture_output=True, check=True
+        )
+    except subprocess.CalledProcessError as e:
+        print(f"Error occurred while listing processes: {e}")
+        sys.exit(1)
+
+    processes = result.stdout.splitlines()
+
+    for process in processes:
+        try:
+            if "iperf3 -s" in process and "-s" in process:
+                parts = process.split()
+                if len(parts) > 1:
+                    pid = int(parts[1])
+                    # Not killing default iperf server spun up on entrypoint...
+                    if pid > 1:
+                        log.info(
+                            f"Killing iperf3 server process (PID: {pid}) in {CURR_POD_NAME} on {CURR_WORKER_NODE_NAME}"
+                        )
+                        try:
+                            os.kill(pid, signal.SIGTERM)
+                        except PermissionError:
+                            log.error(
+                                f"Permission denied: Could not kill process with PID {pid} in {CURR_POD_NAME} on {CURR_WORKER_NODE_NAME}."
+                            )
+                            sys.exit(1)
+                        except ProcessLookupError:
+                            log.error(
+                                f"Process with PID {pid} does not exist in {CURR_POD_NAME} on {CURR_WORKER_NODE_NAME}."
+                            )
+                            sys.exit(1)
+                        except Exception as e:
+                            log.error(
+                                f"Failed to kill process with PID {pid}: {e} in {CURR_POD_NAME} on {CURR_WORKER_NODE_NAME}"
+                            )
+                            sys.exit(1)
+                    else:
+                        log.info(
+                            f"Nothing left to kill in {CURR_POD_NAME} on {CURR_WORKER_NODE_NAME} (Not killing default entrypoint iperf3 server)."
+                        )
+                else:
+                    log.error(
+                        f"Unexpected format in process line: {process} in {CURR_POD_NAME} on {CURR_WORKER_NODE_NAME}"
+                    )
+                    sys.exit(1)
+        except ValueError:
+            log.error(
+                f"Could not convert PID to an integer: {process} in {CURR_POD_NAME} on {CURR_WORKER_NODE_NAME}"
+            )
+            sys.exit(1)
+        except Exception as e:
+            log.error(
+                f"An unexpected error occurred: {e} in {CURR_POD_NAME} on {CURR_WORKER_NODE_NAME}"
+            )
+            sys.exit(1)
+
+
+if __name__ == "__main__":
+    kill_all_iperf_servers()

autopilot-daemon/network/network_workload.py

@@ -23,6 +23,10 @@ def __init__(self, namespace=None, workload_name="Ring Topology"):
             exit(1)
 
     def gen_autopilot_node_map_json(self):
+        #
+        # TODO: This is bad because it gets all endpoints, but what happens if
+        # we have a failing worker that doesn't have any pods? Well we skip it...this bad...
+        #
         try:
             endpoints = self.v1.list_namespaced_endpoints(
                 self.namespace,