The change will limit the hosts the ceph client role is executed against
by checking the ceph conditionals before simply executing the role.
This will help general run times especially on large scale clouds.
Change-Id: Ic7130368c5a2bb231c35bb459a2bebaa28a5a968
Signed-off-by: Kevin Carter <kevin.carter@rackspace.com>
This change introduces a set of functions for dynamically populating the
lxc_hosts function on each inventory run. Any previous membership is
overwritten each time. The output is written to the
openstack_inventory.json still in order to provide parity with the
values provided to the ansible executable on stdout.
Previous code was fairly naive, inserting hosts into lxc_hosts when they
were marked as `is_metal`. This would add hosts, such as Ceph, that
were on metal but had no LXC containers.
A previous attempt at fixing this changed the _build_container_hosts
function to provide more information about the container build process,
then used that data in the _append_host_containers function to populate
the lxc_hosts group. However, this approach failed due to limited
information in each pass of the loop - if a node was an AIO, it might be
erroneously removed from the lxc_hosts group because a container wasn't
built on a given pass, and due to ordering, that pass may be the last
one of the loop.
To get around such problems, this code instead processes the inventory
in whole, after all containers have been made. Population into the group
is determined according to whether or not a given host's `physical_host`
hostvar matches the host name.
Change-Id: I9f3336f77cd0ef05fe1c7edeaf7defc6d93c3111
Closes-Bug: #1660996
In https://review.openstack.org/432134 we added support
for more specific reactions to the db sync check, but
the current SHA we have for keystone does not include
fixes for a new deployment.
This updates the keystone SHA to include that support.
Change-Id: Ib85ae24ee50513c02cd32f09b0bf48b95aab20dc
If not forcing to install urllib3, ansible on the deploy node
will reuse the urrlib3 from site-packages, which could be older
than the one in global-requirements, because ansible doesn't
list it in its dependencies, so we don't explicit force a
certain version to be installed.
On my machine, I had an urllib3 installed with a version of
1.7.1, and all the lookups with https had SSL issues. Moving to
urllib3 fixed the thing.
This only cares about the ansible side, to ensure ansible venv
has the proper version. Hosts targetted by ansible (for example
when using get_url), also need a fix. This will be done in
different patches.
Change-Id: Ia07969ac9aee90724eecc6657affd1dbe16045a7
With the current play we cannot over-ride the
```/openstack/{{ inventory_hostname }}``` host directory for bind
mounting database directory. Updated it so that user can specify
a particular host direcotry of his choice.
Change-Id: I29115b048cc7c65f55217fbe2abf7f4e954a0e5e
Removed hardcoded gnocchi database bind mount dirs in play and
added to group_vars so users can over-ride if needed.
Change-Id: Icf2c1abf661302b54d870028df98c06e89a32dde
This change removes ceilometer from the integrated gate. Sadly the
ceilomteter roles and playbooks have been a little neglected which has
resulted in a slow and sometimes unstable system. To improve gate times
ceilometer has been pulled out of the integrated gate. Once we're able
to get the ceilometer roles and playbooks updated to the latest
release and we're able to resolve all of the deprecations we should
put these rolse back in rotation for the integrated gate.
Change-Id: I116928a1db96e242ed87177578581fe7bf16a001
Signed-off-by: Kevin Carter <kevin.carter@rackspace.com>
The playbooks were running a bunch of common tasks that could be
simplified and unified into a single common_task include. This PR moves
the interface look into a single place and cleans up the duplicate code
within the cinder, neutron, and nova playbooks.
Change-Id: I46f2812670ea17b9ff28ef6f054d82fd6dccc999
Signed-off-by: Kevin Carter <kevin.carter@rackspace.com>
When haproxy endpoint management has a wait condition for the
connection drain to occur, it causes the module to fail if a backend
is not specified. Since our current implementation relies on backend
auto discovery (since the same host may be present in multiple backends),
this causes the module to fail when it attempts to disable the endpoints.
Change-Id: Ia292e4cb4d8ba73a72af8305cc2006aea9c43e32
Use load_inventory instead for load_from_json. Rename
load_from_json to _load_from_json.
Move most of the --clear-ip and --remove calls to their
corresponding functions. Modified other calls to account
for the change
Add a test to test for inventory item removal.
fixes bug: 1663328
Change-Id: I2dbecb085383b3d02e298cc09b02566d1e52a064
Per https://review.openstack.org/#/c/413920/, the ceilometer-collector
service is now deprecated.
This commit removes the inventory group mappings to the
ceilometer-collector service/container. It also removes
the service from the bootstrap and testing inventory.
Relates-To: I396b154d106c0afba44d57792ae6dad39b33a6f5
Closes-Bug: 1666238
Change-Id: Ied14c2a0eda928197c187ad28781568f2d58f51c
on CentOS:
In [2]: platform.linux_distribution()
Out[2]: ('CentOS Linux', '7.2.1511', 'Core')
Also don't add repo_build_pip_extra_indexes if the mirror list is empty
Change-Id: Ia17915a03e7460689a8e403f93bcc87485cde13d
The os_nova role is referring to the ironic_service_user_name variable
but it is not defined anywhere. This patch adds it back into the
default set of variables.
Change-Id: I4b9f49db9b101be84d4792fb2d409c489161c953
Closes-Bug: #1647253
