Try subprocess

Signed-off-by: Alexander Indenbaum <[email protected]>

control/server.py

@@ -16,7 +16,7 @@
 import json
 import logging
 import signal
-import threading
+import time
 from concurrent import futures
 from google.protobuf import json_format
 
@@ -87,6 +87,12 @@ def __exit__(self, exc_type, exc_value, traceback):
             self.logger.info("Stopping the server...")
             self.server.stop(None)
 
+        if self.discovery_pid:
+            self.logger.info("Terminating discovery service...")
+            os.kill(self.discovery_pid, signal.SIGTERM)
+            time.sleep(0.1) # a bit of grace period before final blow
+            os.kill(self.discovery_pid, signal.SIGKILL)
+
         self.logger.info("Exiting the gateway process.")
 
     def serve(self):
@@ -115,25 +121,28 @@ def serve(self):
         # Start server
         self.server.start()
 
+        # Start discovery service
+        self._start_discovery_service()
+
+    def _start_discovery_service(self):
+        """Runs either SPDK on CEPH NVMEOF Discovery Service."""
         enable_spdk_discovery_controller = self.config.getboolean_with_default("gateway", "enable_spdk_discpovery_controller", False)
-        if not enable_spdk_discovery_controller:
-            try:
-                rpc_nvmf.nvmf_delete_subsystem(self.spdk_rpc_ping_client, "nqn.2014-08.org.nvmexpress.discovery")
-            except Exception as ex:
-                self.logger.error(f"  Delete Discovery subsystem returned with error: \n {ex}")
-                raise
+        if enable_spdk_discovery_controller:
+            self.logger.info("Using SPDK discovery service")
+            return
+
+        try:
+            rpc_nvmf.nvmf_delete_subsystem(self.spdk_rpc_ping_client, "nqn.2014-08.org.nvmexpress.discovery")
+        except Exception as ex:
+            self.logger.error(f"  Delete Discovery subsystem returned with error: \n {ex}")
+            raise
 
-            def discovery_thread():
-                while True:
-                    try:
-                        DiscoveryService(self.config).start_service()
-                    except:
-                        self.logger.exception("Discovery service exception:")
-                    self.logger.error("Discovery service restarting")
-
-            # Start discovery service thread
-            self.logger.info("Starting ceph nvmeof discovery service thread")
-            threading.Thread(target=discovery_thread).start()
+        # run ceph nvmeof discovery service in sub-process
+        self.discovery_pid = os.fork()
+        if self.discovery_pid == 0:
+            self.logger.info("Starting ceph nvmeof discovery service")
+            DiscoveryService(self.config).start_service()
+            raise SystemExit
 
     def _add_server_listener(self):
         """Adds listener port to server."""