Andreas Jaeger [Thu, 22 May 2014 16:48:04 +0000 (18:48 +0200)]
Remove all mostly untranslated PO files
We now only import PO files that are at least 75 % translated,
so we can delete now all other PO files. The patch only
removes the mostly untranslated files. Once a file becomes mostly
translated, the bot will import it again.
Ryan Moe [Fri, 2 May 2014 20:08:45 +0000 (13:08 -0700)]
Make pid file locking non-blocking
fcntl.flock will block indefinitely if another process holds an exclusive
lock. A non-blocking flock operation will raise an error when a lock already
exists so we can fail immediately.
Rajesh Mohan [Wed, 21 May 2014 02:41:26 +0000 (19:41 -0700)]
Do not defer IPTables apply in firewall path
By default, iptables apply is deferred in L3 agent. For
external gateways, iptables is applied immediately (to
enable NAT for floating IP). Similarly, when firewall
is created/updated/deleted, iptable rules are applies
immediately.
Steven Gonzales [Mon, 12 May 2014 14:51:13 +0000 (07:51 -0700)]
Metaclass Python 3.x Compatibility
According to the OpenStack style guide
(http://docs.openstack.org/developer/hacking/),
for Python 3.x compatibility use "six.add_metaclass"
instead of " __metaclass__."
Terry Wilson [Tue, 20 May 2014 15:58:32 +0000 (10:58 -0500)]
Remove hard dependency on novaclient
The nova notification patch introduces a hard dependency on
novaclient when it is a runtime-configurable dependency. The
import from novaclient should be conditional on the
appropriate nova notification options being enabled in the
config.
Ed Bak [Fri, 7 Mar 2014 17:16:15 +0000 (17:16 +0000)]
Provide way to reserve dhcp port during failovers
This change provides a way to save the dhcp port when failing
over a network from one dhcp agent to another. When a
dhcp-agent-network-remove is issued, the dhcp port device_id is
marked as reserved which causes it to not be deleted. When a
subsequent dhcp-agent-network-add is issued, the reserved port
is used and the device_id is corrected. This is desirable
in order to maintain the dhcp port ip address so that dns doesn't
get impacted. Unit test added.
With ML2 plugin and l2-pop mechanism driver, it's possible to locally
answer to the ARP request of the VM and avoid ARP broadcasting emulation
on overlay which is costly.
When this functionality is enabled, the OVS flows logic evolves to [1].
This functionality was introduce in 2.1 OVS branch [2].
A README is added to describe l2-pop mechanism driver and the agents
particularities.
Sam Hague [Fri, 2 May 2014 18:08:09 +0000 (14:08 -0400)]
Allow vlan type usage for OpenDaylight ml2
The initial OpenDaylight integration with Openstack did not support
vlan isolation so it was not included as a valid type. This change
adds the vlan type as allowed.
Also modified tests to include vlan as supported in the
check_segment().
DocImpact:
Use of VLANs with ML2 and the OpenDaylight mechanism driver
requires OpenDaylight Helium or newer to be installed.
NSX: fix error when creating VM ports on subnets without dhcp
The error was caused by a missing LSN port (which is created when
the subnet is setup with dhcp enabled). Therefore, check the port
is in place before proceeding with the setup.
Gary Kotton [Thu, 1 May 2014 13:10:57 +0000 (06:10 -0700)]
Add missing translation support
Update a number of files to add missing translation support.
The patch adds a new hacking check - N320. This ensures that
all log messages, except debug ones, have translations.
A '# noqa' indicates that the validation will not be done on
the specific log message. This should be used in cases where
the translations do not need to be done, for example, the log
message is logging raw data.
NSX: Do not extend fault map for network gateway ext
Most of the exceptions defined for this extension already extend
an exception-mapped class. This patch ensures this is true for
every exception and then removes the fault map extension code.
Roman Podoliaka [Wed, 16 Apr 2014 10:59:33 +0000 (13:59 +0300)]
Fix issubclass() hook behavior in PluginInterface
Currently, PluginInterface provides an issubclass() hook that returns
True for issubclass(A, B) call, if all abstract methods of B (stored
in B.__abstractmethods__) can be found in the A.__mro__ tuple of
classes. But there is an edge case, when B doesn't have any abstract
methods, which leads to issubclass(A, B) call returning True even if
A and B are not related all.
E.g. issubclass(NeutronPluginPLUMgridV2, NsxPlugin) returns True,
while these two are different core plugins. And it gets even more
trickier when superclasses are involved: e.g. SecurityGroupDbMixin is
a superclass of NsxPlugin, so depending on the fact whether the
python module with NsxPlugin class is imported or not,
issubclass(NeutronPluginPLUMgridV2, SecurityGroupDbMixin) will return
either False or True accordingly.
The OVS lib deferred apply methods use a dict to save flows to add,
modify or delete when deffered apply is switched off.
If another thread adds, modifies or deletes flows on that dict during
another process called deffered_apply_off, its flows could be ignored.
This fix stash reference flows list and point the flows list to a new
cleared flows list. Then, it applies flows from the stashed flows list.
Maru Newby [Fri, 9 May 2014 22:24:45 +0000 (22:24 +0000)]
Make plugin deallocation check optional
The fix for memory leakage in the related bug added a check for plugin
deallocation that performed a call to gc.collect() after every test.
This had the side-effect of increasing test execution time by ~50%, so
this patch makes the check optional via an environment variable
(OS_CHECK_PLUGIN_DEALLOCATION).
It may make sense to create a periodic job that runs with the check
enabled, but otherwise the check can be used by developers for
debugging purposes.
Restore the default value of the send_arp_for_ha configuration
parameter to 3. This parameter was set to 0 by default to
prevent kernel crashes due to bug 1273386.
Maru Newby [Wed, 7 May 2014 22:41:40 +0000 (22:41 +0000)]
Ensure core plugin deallocation after every test
The unit tests were previously consuming an excessive amount of memory
(4GB+) due to plugin instances persisting in memory. Deallocation was
not possible where a combination of circular references and mocking
was involved. This patch ensures that only NeutronManager holds a
plugin reference and that all other references are instances of
weakref.proxy. Residual memory footprint for tox executed on a
12-core machine has been reduced to ~1.3GB. Plugin deallocation is
validated at the end of each test to prevent regressions.
This change also includes fixes to unit tests that depended on plugin
instances persisting across tests.
Kevin Benton [Thu, 8 May 2014 23:33:06 +0000 (16:33 -0700)]
Big Switch: Check source_address attribute exists
Check that the source_address attribute exists on
HTTPSConnection objects before referencing it since
it's not present on python 2.6 deployments. If it
does not exist, create a socket connection without
specifying the source.
ML2 VxlanTypeDriver: Synchronize of VxlanAllocation table
At neutron startup VxlanTypeDriver syncs tunnel range from conf
to DB. In case multiple servers deployment restarting several
servers at the same time could lead to DB exceptions being thrown.
Need to synchronize between neutron servers by locking VxlanAllocation
table.
NSX: ensure that no LSN is created on external networks
Logical Service Nodes (LSN) are those that provide dhcp
and metadata services to networks when the NSX plugin is
used without the openstack dhcp and metadata proxy
agents. External networks do not work with Neutron's
dhcp and metadata services, therefore there is no need
to provision a LSN in this case.
In case IpSecPolicy or IKEPolicy is updated while VPN connection that uses it
is already established, IPsecPolicyInUse or IKEPolicuInUse is raised.
Need to clarify their messages to emphasize that policies can't be updated
because they are used by established connection.
Another patch (commit-id: df96a03e6c74f1fceb4525900bca885307907b79) to
fix the same issue with less intrusive consequences was already merged
into master. There's still benefit to have debug logs if this does not
make gate jobs fail, so revert to previous setting value.
Ihar Hrachyshka [Tue, 6 May 2014 13:52:35 +0000 (15:52 +0200)]
Disable debug messages when running unit tests
Some runs started to fail due to large sub_unit.log files. We should
reduce their size. The easy way to achieve this is to disable INFO
messages that are filling our logs.
The policy engine is currently being called for every attribute
of every resource to be returned by a list response. This is
harming the API performance; moreover such a high number of checks
is also unnecessary.
This patch therefore slightly changes the API logic so that list
response first determine the list of attributes which should be
returned querying the policy engine and then use this list for
all resource items to be returned.
To this aim a few methods in base.py needed to be refactored.
This patch also removes the routine check_if_exists from policy.py
and the related PolicyNotFound exception.
Finally, this patch also removes unnecessary admin_or_owner rules
when applied to attributes. This kind of rule indeed has no effect
anyway because of Neutron's ownership checks. The rules were removed
because this change won't allow anymore for having attribute-level
policies whose evaluation result depends on the resource value.
Jakub Libosvar [Tue, 6 May 2014 08:56:31 +0000 (10:56 +0200)]
Fix importing module in test_netscaler_driver
During H302 was introduced new bug importing package instead of module.
This patch is reverting back change on this file from
https://review.openstack.org/#/c/89628/5
Attila Fazekas [Sun, 4 May 2014 17:54:37 +0000 (19:54 +0200)]
L3 RPC loop could delete a router on concurrent update
routers_updated does not acquire any lock just updates
a set for future rpc loop processing.
The self.updated_routers can be changed by concurrent update
notification. If this change happens at the time around the
self.plugin_rpc.get_routers call, the additional routers
- by mistake - is considered as admin_state_up=false routers, which
are safe to delete.
Creating a local copy of the updated_routers and preserve
the fresh updated_routers entries for the next _rpc_loop
operations.
Code Review / openstack-build / neutron-build.git / log