Fix the logical port created twice

Logical port creation is a POST request. Sometimes it will trigger
ConnectionResetError which is a IOError. request_with_retry_on_ssl_error
will retry it.
If request has parameter retry_confirm, exception will be raised so ncp
could query if port has been created to avoid creating port twice.

Change-Id: Ic97b39c7a3736f02a79ab891970c1ad67b123156
(cherry picked from commit ac224a85a8)

vmware_nsxlib/tests/unit/v3/policy/test_resources.py

@@ -2613,7 +2613,7 @@ class TestPolicyEnforcementPoint(NsxPolicyLibTestCase):
                                                          tenant=TEST_TENANT)
             api_post.assert_called_once_with(
                 expected_def.get_resource_path() + '?action=reload',
-                None, expected_results=None, headers=None)
+                None, expected_results=None, headers=None, retry_confirm=False)
 
 
 class TestPolicyDeploymentMap(NsxPolicyLibTestCase):

vmware_nsxlib/v3/client.py

@@ -22,6 +22,7 @@ from oslo_serialization import jsonutils
 import requests
 
 from vmware_nsxlib._i18n import _
+from vmware_nsxlib.v3 import constants
 from vmware_nsxlib.v3 import exceptions
 from vmware_nsxlib.v3 import utils
 
@@ -173,9 +174,10 @@ class RESTClient(object):
                             expected_results=expected_results)
 
     def create(self, resource='', body=None, headers=None,
-               expected_results=None):
+               expected_results=None, retry_confirm=False):
         return self.url_post(resource, body, headers=headers,
-                             expected_results=expected_results)
+                             expected_results=expected_results,
+                             retry_confirm=retry_confirm)
 
     def patch(self, resource='', body=None, headers=None):
         return self.url_patch(resource, body, headers=headers)
@@ -205,9 +207,11 @@ class RESTClient(object):
         return self._rest_call(url, method='PUT', body=body, headers=headers,
                                expected_results=expected_results)
 
-    def url_post(self, url, body, headers=None, expected_results=None):
+    def url_post(self, url, body, headers=None, expected_results=None,
+                 retry_confirm=False):
         return self._rest_call(url, method='POST', body=body, headers=headers,
-                               expected_results=expected_results)
+                               expected_results=expected_results,
+                               retry_confirm=retry_confirm)
 
     def url_patch(self, url, body, headers=None):
         return self._rest_call(url, method='PATCH', body=body, headers=headers)
@@ -268,6 +272,7 @@ class RESTClient(object):
                    silent=False, expected_results=None, **kwargs):
         request_headers = headers.copy() if headers else {}
         request_headers.update(self._default_headers)
+        retry_confirm = kwargs.pop(constants.API_RETRY_CONFIRM, False)
 
         if utils.INJECT_HEADERS_CALLBACK:
             inject_headers = utils.INJECT_HEADERS_CALLBACK()
@@ -281,7 +286,8 @@ class RESTClient(object):
         result = do_request(
             request_url,
             data=body,
-            headers=request_headers)
+            headers=request_headers,
+            retry_confirm=retry_confirm)
         te = time.time()
         if silent:
             self._conn.set_silent(False)

vmware_nsxlib/v3/cluster.py

@@ -104,15 +104,23 @@ class TimeoutSession(requests.Session):
     # wrapper timeouts at the session level
     # see: https://goo.gl/xNk7aM
     def request(self, *args, **kwargs):
+        retry_confirm = kwargs.pop(constants.API_RETRY_CONFIRM, False)
+
         def request_with_retry_on_ssl_error(*args, **kwargs):
             try:
                 return super(TimeoutSession, self).request(*args, **kwargs)
-            except (IOError, OpenSSL.SSL.Error):
+            except (IOError, OpenSSL.SSL.Error) as err:
                 # This can happen when connection tries to access certificate
                 # file it was opened with (renegotiation?)
                 # Proper way to solve this would be to pass in-memory cert
                 # to ssl C code.
                 # Retrying here works around the problem
+
+                # IOError covered too much Exception, if resources created
+                # successfully on nsxt side but ConnectResetError happened,
+                # Retrying here will create resource twice
+                if isinstance(err, IOError) and retry_confirm:
+                    raise exceptions.RetryConfirm(msg=err)
                 return super(TimeoutSession, self).request(*args, **kwargs)
 
         def get_cert_provider():

vmware_nsxlib/v3/constants.py

@@ -127,3 +127,6 @@ API_DEFAULT_MAX_RATE = 100
 # API Call Log Related const
 API_CALL_LOG_PER_CLUSTER = 'API_LOG_PER_CLUSTER'
 API_CALL_LOG_PER_ENDPOINT = 'API_LOG_PER_ENDPOINT'
+
+# API Retry confirm parameter
+API_RETRY_CONFIRM = 'retry_confirm'

vmware_nsxlib/v3/exceptions.py

@@ -238,6 +238,10 @@ class CannotConnectToServer(ManagerError):
     message = _("Cannot connect to server")
 
 
+class RetryConfirm(NsxLibException):
+    message = _("Retry will be handled by upper layer: %(msg)s")
+
+
 class ResourceInUse(ManagerError):
     message = _("The object cannot be deleted as either it has children or it "
                 "is being referenced by other objects")

vmware_nsxlib/v3/resources.py

@@ -164,7 +164,7 @@ class LogicalPort(utils.NsxLibApiBase):
                switch_profile_ids=None, vif_type=None, app_id=None,
                allocate_addresses=nsx_constants.ALLOCATE_ADDRESS_NONE,
                description=None, tn_uuid=None,
-               extra_configs=None):
+               extra_configs=None, retry_confirm=False):
         tags = tags or []
         body = {'logical_switch_id': lswitch_id}
         # NOTE(arosen): If parent_vif_id is specified we need to use
@@ -181,7 +181,8 @@ class LogicalPort(utils.NsxLibApiBase):
             attachment=attachment,
             description=description,
             extra_configs=extra_configs))
-        return self.client.create(self.get_path(), body=body)
+        return self.client.create(self.get_path(), body=body,
+                                  retry_confirm=retry_confirm)
 
     def delete(self, lport_id):
         self._delete_with_retry('%s?detach=true' % lport_id)