Merge pull request #880 from leonidc/limit_manual_ns_change_lb

To refuse change load balancing group of the namespace if ana-group owner is another gw

Author: leonidc

control/cephutils.py

@@ -12,6 +12,7 @@
 import rbd
 import rados
 import time
+import json
 from .utils import GatewayLogger
 
 class CephUtils:
@@ -31,6 +32,23 @@ def execute_ceph_monitor_command(self, cmd):
             rply = cluster.mon_command(cmd, b'')
             self.logger.debug(f"Monitor reply: {rply}")
             return rply
+    def get_gw_id_owner_ana_group(self, pool, group, anagrp):
+        str = '{' + f'"prefix":"nvme-gw show", "pool":"{pool}", "group":"{group}"' + '}'
+        self.logger.debug(f"nvme-show string: {str}")
+        rply = self.execute_ceph_monitor_command(str)
+        self.logger.debug(f"reply \"{rply}\"")
+        conv_str = rply[1].decode()
+        data = json.loads(conv_str)
+
+        # Find the gw-id that contains "2: ACTIVE" in "ana states"
+        gw_id = None
+        comp_str = f"{anagrp}: ACTIVE"
+        for gateway in data["Created Gateways:"]:
+            if comp_str in gateway["ana states"]:
+              gw_id = gateway["gw-id"]
+              self.logger.debug(f"found gw owner of anagrp {anagrp}: gw {gw_id}")
+              break
+        return gw_id
 
     def get_number_created_gateways(self, pool, group):
         now = time.time()

control/grpc.py

@@ -1225,6 +1225,13 @@ def namespace_change_load_balancing_group_safe(self, request, context):
                     errmsg = f"{change_lb_group_failure_prefix}: Can't find entry for namespace {request.nsid} in {request.subsystem_nqn}"
                     self.logger.error(errmsg)
                     return pb2.req_status(status=errno.ENOENT, error_message=errmsg)
+                anagrp = ns_entry["anagrpid"]
+                gw_id = self.ceph_utils.get_gw_id_owner_ana_group(self.gateway_pool, self.gateway_group, anagrp)
+                self.logger.debug(f"ANA group of ns#{request.nsid} - {anagrp} is owned by gateway {gw_id}, self.name is {self.gateway_name}")
+                if self.gateway_name != gw_id:
+                    errmsg = f"ANA group of ns#{request.nsid} - {anagrp} is owned by gateway {gw_id} so try this command from it, this gateway name is {self.gateway_name}"
+                    self.logger.error(errmsg)
+                    return pb2.req_status(status=errno.ENODEV, error_message=errmsg)
 
             try:
                 ret = rpc_nvmf.nvmf_subsystem_set_ns_ana_group(

tests/ha/ns_lb_change.sh

@@ -60,11 +60,9 @@ make exec SVC=bdevperf OPTS=-T CMD="$bdevperf -v -t $timeout -s $BDEVPERF_SOCKET
 
 (
 sleep 8;   
-
 lb_group=1;
-docker compose run -T --rm nvmeof-cli --server-address $ip --server-port 5500  namespace change_load_balancing_group -n nqn.2016-06.io.spdk:cnode1 --nsid 1 --load-balancing-group $lb_group ;
-
 
+docker compose run -T --rm nvmeof-cli --server-address $ip --server-port 5500  namespace change_load_balancing_group -n nqn.2016-06.io.spdk:cnode1 --nsid 1 --load-balancing-group $lb_group;
 priv_res1=$(calc_written_bytes_in_sec $ip) ; 
 
 echo "ℹ️  written bytes through $ip   $priv_res1 ";
@@ -77,14 +75,16 @@ echo "ℹ️  written bytes through $ip2  $priv_res2 ";
 for i in $(seq 6); do
    if [ $lb_group -eq 1 ]; then
        lb_group=2
+       IP=$ip
    else 
        lb_group=1
+       IP=$ip2
    fi;
 
    echo "ℹ️ ℹ️ Change lb group of ns 1 to $lb_group :" ;
-   docker compose run -T --rm nvmeof-cli --server-address $ip --server-port 5500  namespace change_load_balancing_group -n nqn.2016-06.io.spdk:cnode1 --nsid 1 --load-balancing-group $lb_group ;
+   docker compose run -T --rm nvmeof-cli --server-address $IP --server-port 5500  namespace change_load_balancing_group -n nqn.2016-06.io.spdk:cnode1 --nsid 1 --load-balancing-group $lb_group;
    sleep 4;
-
+ 
    res1=$(calc_written_bytes_in_sec $ip) ;
 
    echo "ℹ️  written bytes through $ip ?:  $res1";

tests/test_cli_change_lb.py

@@ -109,8 +109,13 @@ def create_namespaces(caplog, ns_count, subsys):
 
 def change_one_namespace_lb_group(caplog, subsys, nsid_to_change, new_group):
     caplog.clear()
-    cli(["--server-port", "5501", "namespace", "change_load_balancing_group", "--subsystem", subsys, "--nsid", nsid_to_change, "--load-balancing-group", new_group])
-    time.sleep(10)
+    cli(["--server-port", "5502", "namespace", "change_load_balancing_group", "--subsystem", subsys, "--nsid", nsid_to_change, "--load-balancing-group", new_group])
+    time.sleep(8)
+    if "so try this command from it" in caplog.text:
+        caplog.clear()
+        cli(["--server-port", "5501", "namespace", "change_load_balancing_group", "--subsystem", subsys, "--nsid", nsid_to_change, "--load-balancing-group", new_group])
+        time.sleep(8)
+
     assert f"Changing load balancing group of namespace {nsid_to_change} in {subsys} to {new_group}: Successful" in caplog.text
     assert f"Received request to change load balancing group for namespace with NSID {nsid_to_change} in {subsys} to {new_group}, context: <grpc._server" in caplog.text
     assert f"Received request to delete namespace" not in caplog.text