Use correct interface for LNet in SSI testing (whamcloud#864)

* Use correct interface for LNet in SSI testing

Fixes #719

SSI testing is currently using the host interface for LNet. Instead it should be using the LNet interface (.20) as intended.

Signed-off-by: Joe Grund <jgrund@whamcloud.io>

* fix formatting

Signed-off-by: Joe Grund <jgrund@whamcloud.io>

* send config in expected format

Signed-off-by: Joe Grund <jgrund@whamcloud.io>

* add lnd_type

Signed-off-by: Joe Grund <jgrund@whamcloud.io>

* try to avoid server restart

Signed-off-by: Joe Grund <jgrund@whamcloud.io>

provisioner_output-vagrant-ssi.json.in

@@ -68,6 +68,7 @@
       "lustre_client_version": "",
       "address": "@HOSTNAME@vm@CLUSTER@2@DOMAINNAME@",
       "ip_address": "@VM2_IPADDRESS@",
+      "lnet_address": "@VM2_LNETADDRESS@",
       "distro": "el7.4"
     }
   ],
@@ -132,6 +133,7 @@
       "root_password": "vagrant",
       "address": "@HOSTNAME@vm@CLUSTER@5@DOMAINNAME@",
       "ip_address": "@VM5_IPADDRESS@",
+      "lnet_address": "@VM5_LNETADDRESS@",
       "distro": "el7.4"
     },
     {
@@ -160,6 +162,7 @@
       "root_password": "vagrant",
       "address": "@HOSTNAME@vm@CLUSTER@6@DOMAINNAME@",
       "ip_address": "@VM6_IPADDRESS@",
+      "lnet_address": "@VM6_LNETADDRESS@",
       "distro": "el7.4"
     },
     {
@@ -188,6 +191,7 @@
       "root_password": "vagrant",
       "address": "@HOSTNAME@vm@CLUSTER@7@DOMAINNAME@",
       "ip_address": "@VM7_IPADDRESS@",
+      "lnet_address": "@VM7_LNETADDRESS@",
       "distro": "el7.4"
     },
     {
@@ -216,6 +220,7 @@
       "root_password": "vagrant",
       "address": "@HOSTNAME@vm@CLUSTER@8@DOMAINNAME@",
       "ip_address": "@VM8_IPADDRESS@",
+      "lnet_address": "@VM8_LNETADDRESS@",
       "distro": "el7.4"
     },
     {
@@ -239,6 +244,7 @@
       "root_password": "vagrant",
       "address": "@HOSTNAME@vm@CLUSTER@9@DOMAINNAME@",
       "ip_address": "@VM9_IPADDRESS@",
+      "lnet_address": "@VM9_LNETADDRESS@",
       "distro": "el7.4"
     }
   ],

pyproject.toml

@@ -6,9 +6,11 @@ exclude = '''
     \.git
   | \.tox
   | \.venv
+  | \.cargo
   | _build
   | build
   | dist
   | south_migrations
+  | target
 )/
 '''

tests/framework/utils/provisioner_interface/provision_cluster

@@ -111,11 +111,12 @@ else
             -e "s/@VM7_IPADDRESS@/10.73.10.23/g"     \
             -e "s/@VM8_IPADDRESS@/10.73.10.24/g"     \
             -e "s/@VM9_IPADDRESS@/10.73.10.39/g"     \
-            -e "s/@VM2_IPADDRESS@/10.73.20.32/g"     \
+            -e "s/@VM2_LNETADDRESS@/10.73.20.32/g"   \
             -e "s/@VM5_LNETADDRESS@/10.73.20.21/g"   \
             -e "s/@VM6_LNETADDRESS@/10.73.20.22/g"   \
             -e "s/@VM7_LNETADDRESS@/10.73.20.23/g"   \
             -e "s/@VM8_LNETADDRESS@/10.73.20.24/g"   \
+            -e "s/@VM9_LNETADDRESS@/10.73.20.39/g"   \
             < "$CHROMA_DIR"/provisioner_output-vagrant-"${MAKE_TARGET%_tests}".json.in \
             > provisioner_output.json
     fi

tests/integration/core/chroma_integration_testcase.py

@@ -239,6 +239,31 @@ def check_for_HYD_2849_4050():
 
         self.wait_for_commands(self.chroma_manager, command_ids, timeout=INSTALL_TIMEOUT)
 
+    def configure_lnet(self, host):
+        host_config = self.get_host_config(host["nodename"])
+
+        if host_config is None:
+            logger.warning("No host config for '{}'".format(host["nodename"]))
+            return
+
+        info = self._get_lnet_info(host)
+
+        def convert(ni):
+            lnd_network = 0 if ni.get("inet4_address") == host_config["lnet_address"] else -1
+
+            return {"lnd_network": lnd_network, "lnd_type": "tcp", "network_interface": ni["resource_uri"]}
+
+        objects = map(convert, info.network_interfaces)
+
+        logger.info("updating LNet to: {}".format(objects))
+
+        response = self.chroma_manager.post("/api/nid/", body={"objects": objects})
+
+        self.assertTrue(response.successful, response.text)
+        command_id = response.json["command"]["id"]
+
+        self.wait_for_command(self.chroma_manager, command_id, timeout=LONG_TEST_TIMEOUT)
+
     def _add_hosts(self, addresses, auth_type):
         """Add a list of lustre server addresses to chroma and ensure lnet ends in the correct state."""
         self.validate_hosts(addresses, auth_type)
@@ -260,6 +285,12 @@ def _add_hosts(self, addresses, auth_type):
             expected_return_code=None,
         )
 
+        for host in new_hosts:
+            profile = self.get_current_host_profile(host)
+
+            if profile.get("managed") or profile.get("worker"):
+                self.configure_lnet(host)
+
         for host in new_hosts:
             # Deal with pre-3.0 versions.
             if host["state"] in ["lnet_up", "lnet_down", "lnet_unloaded"]:

tests/integration/core/remote_operations.py

@@ -951,27 +951,19 @@ def clear_ha(self, server_list):
     def clear_lnet_config(self, server_list):
         """
         Removes the lnet configuration file for the server list passed.
-         """
+        """
 
-        # This isn't really that bad because the file will not be recreated if I delete it so probably ammounts
-        # to at most an extra reboot cycle per test session.
-        # Note the sleep ensures the reboot really happens otherwise it might look alive in the await_server_boot
         for server in server_list:
             self._ssh_address(
                 server["address"],
                 """
                               [ -f /etc/modprobe.d/iml_lnet_module_parameters.conf ] &&
                               rm -f /etc/modprobe.d/iml_lnet_module_parameters.conf &&
-                              reboot &&
-                              sleep 20
+                              lustre_rmmod
                               """,
                 expected_return_code=None,
             )  # Keep going if it failed - may be none there.
 
-        # Now ensure they have all comeback to life
-        for server in server_list:
-            self.await_server_boot(server["fqdn"], restart=True)
-
     def install_upgrades(self):
         raise NotImplementedError("Automated test of upgrades is HYD-1739")