x
/
vmware-nsx
Code Issues Proposed changes

NSX|P Change order of actions while deleting a port 

The NSX port should be deleted after the neutron DB entry, to make sure
a concurent update-port command do not recreate it after it was deleted
from the DB.
Thereason for this is that PATCH used for update can also recreate a
deleted resource.

Change-Id: I73f363127f79510fe58dc88fa15af0673c2f8562
This commit is contained in:
asarfaty 2019-12-10 09:56:01 +02:00 committed by Adit Sarfaty
parent 968a10ecbe
commit 9293081128
1 changed files with 23 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
vmware_nsx/plugins/nsx_p/plugin.py
View File

@ -1281,27 +1281,43 @@ class NsxPolicyPlugin(nsx_plugin_common.NsxPluginV3Base):
# a l3 router. If so, we should prevent deletion here
if l3_port_check:
self.prevent_l3_port_deletion(context, port_id)
port = self.get_port(context, port_id)
# Prevent DHCP port deletion if native support is enabled
if (cfg.CONF.nsx_p.allow_passthrough and
not force_delete_dhcp and
port['device_owner'] in [const.DEVICE_OWNER_DHCP]):
port_data['device_owner'] in [const.DEVICE_OWNER_DHCP]):
msg = (_('Can not delete DHCP port %s') % port_id)
raise n_exc.BadRequest(resource='port', msg=msg)
if not force_delete_vpn:
self._assert_on_vpn_port_change(port)
if self._is_backend_port(context, port_data):
self._delete_port_on_backend(context, net_id, port_id)
self._assert_on_vpn_port_change(port_data)
self.disassociate_floatingips(context, port_id)
# Remove Mac/IP binding from native DHCP server and neutron DB.
if cfg.CONF.nsx_p.allow_passthrough:
self._delete_dhcp_binding(context, port)
self._delete_dhcp_binding(context, port_data)
super(NsxPolicyPlugin, self).delete_port(context, port_id)
# Delete the backend port last to prevent recreation by another process
if self._is_backend_port(context, port_data):
try:
self._delete_port_on_backend(context, net_id, port_id)
except nsx_lib_exc.ResourceNotFound:
# If the resource was not found on the backend do not worry
# about it. The conditions has already been logged, so there
# is no need to do further logging
pass
except nsx_lib_exc.ManagerError as e:
# If there is a failure in deleting the resource, fail the
# neutron operation even though the neutron object was already
# deleted. This way the user will be aware of zombie resources
# that may fail future actions.
msg = (_("Backend segment port deletion for neutron port "
"%(id)s failed. The object was however removed from "
"the Neutron database: %(e)s") %
{'id': port_id, 'e': e})
raise nsx_exc.NsxPluginException(err_msg=msg)
def _update_port_on_backend(self, context, lport_id,
original_port, updated_port,
is_psec_on, qos_policy_id):