4gws fixes

control/grpc.py

@@ -651,6 +651,8 @@ def delete_subsystem_safe(self, request, context):
                 )
                 self.subsys_ha.pop(request.subsystem_nqn)
                 self.subsys_max_ns.pop(request.subsystem_nqn)
+                if request.subsystem_nqn in self.subsystem_listeners:
+                    self.subsystem_listeners.pop(request.subsystem_nqn)
                 self.logger.debug(f"delete_subsystem {request.subsystem_nqn}: {ret}")
             except Exception as ex:
                 self.logger.exception(delete_subsystem_error_prefix)
@@ -869,12 +871,11 @@ def set_ana_state_safe(self, ana_info: pb2.ana_info, context=None):
                             context.set_details(f"{ex}")
                         return pb2.req_status()
         for ana_key in inaccessible_ana_groups :
-            with self.omap_lock(context=context):
-                ret_recycle = self.namespace_recycle_safe(ana_key, peer_msg)
-                if ret_recycle != 0:
-                    errmsg = f"Failure recycle namespaces of ana group {ana_key} "
-                    self.logger.error(errmsg)
-                    return pb2.req_status(status=ret_recycle , error_message=errmsg)
+            ret_recycle = self.namespace_recycle_safe(ana_key, peer_msg)
+            if ret_recycle != 0:
+                errmsg = f"Failure recycle namespaces of ana group {ana_key} "
+                self.logger.error(errmsg)
+                return pb2.req_status(status=ret_recycle , error_message=errmsg)
         return pb2.req_status(status=True)
 
     def choose_anagrpid_for_namespace(self, nsid) ->int:
@@ -2044,30 +2045,11 @@ def get_subsystem_ha_status(self, nqn) -> bool:
         self.logger.debug(f"Subsystem {nqn} enable_ha: {enable_ha}")
         return enable_ha
 
-    def matching_listener_exists(self, context, nqn, traddr, trsvcid) -> bool:
-        if not context:
-            return False
-
-        state = self.gateway_state.local.get_state()
-        # We want to check for all the listeners for this address and port, regardless of the gateway
-        key_prefix = GatewayState.build_partial_listener_key(nqn)
-        key_suffix = GatewayState.build_listener_key_suffix(None, "TCP", traddr, trsvcid)
-
-        for key, val in state.items():
-            if not key.startswith(key_prefix):
-                continue
-            if not key.endswith(key_suffix):
-                continue
-            return True
-
-        return False
-
     def create_listener_safe(self, request, context):
         """Creates a listener for a subsystem at a given IP/Port."""
 
         ret = True
-        traddr = GatewayUtils.escape_address_if_ipv6(request.traddr)
-        create_listener_error_prefix = f"Failure adding {request.nqn} listener at {traddr}:{request.trsvcid}"
+        create_listener_error_prefix = f"Failure adding {request.nqn} listener at {request.traddr}:{request.trsvcid}"
 
         adrfam = GatewayEnumUtils.get_key_from_value(pb2.AddressFamily, request.adrfam)
         if adrfam == None:
@@ -2078,7 +2060,7 @@ def create_listener_safe(self, request, context):
         peer_msg = self.get_peer_message(context)
         self.logger.info(f"Received request to create {request.host_name}"
                          f" TCP {adrfam} listener for {request.nqn} at"
-                         f" {traddr}:{request.trsvcid}, context: {context}{peer_msg}")
+                         f" {request.traddr}:{request.trsvcid}, context: {context}{peer_msg}")
 
         if GatewayUtils.is_discovery_nqn(request.nqn):
             errmsg=f"{create_listener_error_prefix}: Can't create a listener for a discovery subsystem"
@@ -2093,10 +2075,8 @@ def create_listener_safe(self, request, context):
         with self.omap_lock(context=context):
             try:
                 if request.host_name == self.host_name:
-                    listener_already_exist = self.matching_listener_exists(
-                            context, request.nqn, request.traddr, request.trsvcid)
-                    if listener_already_exist:
-                        self.logger.error(f"{request.nqn} already listens on address {traddr}:{request.trsvcid}")
+                    if (adrfam, request.traddr, request.trsvcid) in self.subsystem_listeners[request.nqn]:
+                        self.logger.error(f"{request.nqn} already listens on address {request.traddr}:{request.trsvcid}")
                         return pb2.req_status(status=errno.EEXIST,
                                   error_message=f"{create_listener_error_prefix}: Subsystem already listens on this address")
                     ret = rpc_nvmf.nvmf_subsystem_add_listener(
@@ -2189,7 +2169,7 @@ def create_listener_safe(self, request, context):
                                                     "TCP", request.traddr,
                                                     request.trsvcid, json_req)
                 except Exception as ex:
-                    errmsg = f"Error persisting listener {traddr}:{request.trsvcid}"
+                    errmsg = f"Error persisting listener {request.traddr}:{request.trsvcid}"
                     self.logger.exception(errmsg)
                     errmsg = f"{errmsg}:\n{ex}"
                     return pb2.req_status(status=errno.EINVAL, error_message=errmsg)

tests/ha/setup_4gws.sh

@@ -42,8 +42,9 @@ for i in $(seq $NUM_SUBSYSTEMS); do
     for g in $(seq $NUM_GATEWAYS); do
         GW_NAME=$(gw_name $g)
         GW_IP=$(gw_ip $g)
+        ADDR=0.0.0.0
         PORT=4420
-        docker-compose  run --rm nvmeof-cli --server-address $GW_IP --server-port 5500 listener add  --subsystem $NQN --host-name $GW_NAME --traddr $GW_IP --trsvcid $PORT
+        docker-compose  run --rm nvmeof-cli --server-address $GW_IP --server-port 5500 listener add  --subsystem $NQN --host-name $GW_NAME --traddr $ADDR --trsvcid $PORT
     done
 done
 
@@ -53,7 +54,6 @@ for g in $(seq $NUM_GATEWAYS); do
     echo "Verify $i $GW_NAME $GW_IP"
     GW_NAME=$(gw_name $g)
     GW_IP=$(gw_ip $g)
-    PORT=4420
     subs=$(docker-compose  run --rm nvmeof-cli --server-address $GW_IP --server-port 5500 get_subsystems 2>&1 | sed 's/Get subsystems://')
 
     # verify all resources found in get subsystems