Now that we are python3 only, we should move to using the built
in version of mock that supports all of our testing needs and
remove the dependency on the "mock" package.
This completes removal of all references to "import mock",
changing to "from unittest import mock" in fullstack and
functional tests.
Added a hacking check to enforce it in future patches.
Change-Id: Ifcaf1c21bea0ec3c35278e49cecc90a101a82113
All of the externally consumed variables from neutron.common.constants
now live in neutron-lib. This patch removes neutron.common.constants
and switches all uses over to lib.
NeutronLibImpact
Depends-On: https://review.openstack.org/#/c/647836/
Change-Id: I3c2f28ecd18996a1cee1ae3af399166defe9da87
There is no need to translate help message for config option
which is added and used only in tests.
Change-Id: I8b8bdd4131e6183245bcd2556b2107d769fd6427
Multiple SIGHUPs in quick succession might cause the master keepalived
to forfeit its mastership (which will cause keepalived to remove IPs of
its external devices, severing connectivity). This can happen when, for
example, associating or disassociating multiple floatingips.
The patch makes the agent throttle SIGHUP sent to keepalived: the very first
SIGHUP is always sent; as for subsequent signals, they are delayed till
agent threshold is reached. (It's 3 seconds by default.)
As an example, when three consequent router updates trigger keepalived
respawn then:
* the very first signal is sent as usual;
* the second signal is deferred and sent in up to 3 seconds since the
first signal;
* the third signal is ignored, though the change that triggered it will
be correctly applied by the second signal handler when it is triggered
after threshold delay.
If the last time a spawn request occurred is older than current-time
minus threshold then there is no delay.
Co-Authored-By: Jakub Libosvar <libosvar@redhat.com>
Co-Authored-By: Cedric Brandily <zzelle@gmail.com>
Co-Authored-By: Ihar Hrachyshka <ihrachys@redhat.com>
Closes-Bug: 1647432
Change-Id: I2955e0de835458a2eea4dd088addf33b656f8670
This reverts commit eb14f66d39.
Turned out that fullstack uses those files to start services, instead of
relying on console scripts generated by pbr. So without those files
having +x permissions, fullstack tests using those agents completely
fail.
Change-Id: I14f98fdd244ecfedfa97fb31bff9763a49e5cfb0
Creates classes representing the 3 types of namespaces handled by the L3 agent:
router, snat and fip.
The scope of this change is:
- Creation and deletion methods are provided for each namespace class
- Creation and deletion of router and snat namespaces are moved to the router
classes. These namespaces are now members of the corresponding router class
- Invocation of Fip namespace creation and deletion is left in the agent, since
the agent owns it
- A context manager is provided to move the namespaces (router and snat)
cleanup code out of the agent
A follow up patchset will add methods to create and delete interfaces in the
namespaces. These methods are intended to be used by the router classes
Change-Id: I54b14e593ded6b2990d57a3ae9d598a699ae133e
Partially-Implements: bp restructure-l3-agent
The idea behind the AdvancedServices (Metadata, *aaS) services
to be singletons was to (1) offer a way to get an instance of such
a service in a convinient manner, and to (2) ensure that only one
service registered per service type.
(1) Since the AdvancedService.instance method required an instance
of a L3 agent, this point was kind of missed. If you have a L3 agent
instance in your hand, just use it to access the service you're looking
for.
(2) This is now fulfilled by asserting that only one driver is registered
(per type) in the .add method.
The motivation to make them non-singletons is that:
a) They don't need to be
b) The code is simplified this way
c) I've been facing crazy issues during functional testing where we're
constantly instantiating L3 agents, and the (singleton) metadata
service was referencing the wrong configuration object. The service
was re-initialized during a test, screwing up the configuration of
other mid-run tests.
Closes-Bug: #1425759
Depends-On: I155aece9b7028f1bdd3d9709facef392b38aec27
Depends-On: I199f6a967c64e921624b397fa5b7dfc56572d63a
Change-Id: I192e4d12380346c53c8ae0246371ccc41a836c4c
L3 agent drivers are singletons. They're created once, and hold
self.l3_agent. During testing, the agent is tossed away and
re-built, but the drivers singletons are pointing at the old
agent, and its old configuration.
Change-Id: Ie8a15318e71ea47cccad3b788751d914d51cbf18
Closes-Bug: #1404662
The file l3_agent.py has become too large. This patch is a simple
pure refactor to move some of the functionality in to other files
where things aren't too tangled up. There is no functional change
with this patch and I avoided gratuitous other fixups in this patch in
order to make it easier to review.
I plan to follow up on the new l3_dvr and l3_agent_router modules with
more restructuring in the near future.
Partially-Implements: bp restructure-l3-agent
Change-Id: I3529fe4146c50c940f41eb26d0b5efc5870b3af9
HA failover testing requires the ability for two L3 agents to
co-exist on the same machine. The following changes were made
to support this:
* Add l3 agent ability to simulate nested namespaces by suffixing
each router namespace with '@host'. For example, if a router
with id '1' is created, its namespace will be:
'qrouter-1@host'
* Added test.common submodule which will hold common code which is
shared between the different kind of tests (unit and funtional).
* Added tests.common.agents submodule which holds test-only agents, and
added a test L3NAT agent to be used in the HA functional testings.
Co-Authored-By: Assaf Muller <amuller@redhat.com>
Closes-bug: #1374946
Change-Id: I2185fe276d9d32b34c407396e7ed5d6e2801340c
Brian Haley