In cf0c25c37 [0] a play was added to run the common role against all
hosts. This ends up being redundant since every role includes the common
role as a dependancy. The reasoning behind this change as pointed out by
the author in the review comments [1] was so that an operator could run
with '--tags common' and just have the common role applied.
To avoid redundancy, the common play has been removed and tags have been
added to the common role. This allows for just the common role to run
when another role is including it while reducing redundancy.
A side affect of removing the common playbook which runs against all
hosts is that not all facts on all hosts are gathered at the beginning
of the site.yml. This breaks the haproxy role since it relies heavily on
facts to build out the haproxy.cfg file.
Previously, the haproxy role would include several hosts purely for fact
gathering purposes as pointed out in c68c9d95 [2] and a guard was put in
place so that the tasks would only run against the 'haproxy' group. In
423e3f3f [3] these hosts were removed. After reading the review [4],
this seems to have been done without fully understanding why the hosts
were there in the first place.
This change did not break anything however since the common role that
ran on all hosts mentioned previously would gather all of the facts
necessary.
To fix this fact gathering issue replace the common role play with a
play that will simply gather facts with an 'always' tag to ensure it is
run regardless of what might be passed in the '--tags' argument by the
operator.
Kudos to Paul Bourke for helping identify many of these issues.
[0] cf0c25c37d
[1] https://review.openstack.org/#/c/369212/
[2] c68c9d95fc
[3] 423e3f3fdf
[4] https://review.openstack.org/#/c/355861
TrivialFix
Closes-Bug: #1628472
Change-Id: Ia94146579e743935501f1ff4b4c1bf6cb7c43aa3
By default, if the same role is called in ansible it will only run
once per node. Due to how we have the inventory setup ansible views
every service as a different node and will try to run the common role
tasks again. This causes slowdown in all cases but is particularly
noticable when pulling images. A small change will ensure these tasks
only run once per node per run as originally intended.
TrivialFix
Change-Id: I20b9c46991d10176c8f8645a335eb7a9ed750ee3
Throughout the project overtime some of these file permissions have
changed to have an executable bit. They should not have this bit set.
TrivialFix
Change-Id: I1748b5bde813a0fcac36aeecdfd83245b8ee5be3
This implements all the openstack pieces needed to make ceph work.
DocImpact
Change-Id: I1d24476a966602cf955e5ef872b0efb01319894a
Partially-Implements: blueprint ceph-container
Implements: blueprint kolla-live-migration
Currently we require a slew of deps on each destination node, this
includes a gcc compiler and installing things via pip. We can remove
these dependencies by containerizing them and running and Ansible
inside the container itself. The container would then report back
facts about idempotency.
DocImpact
Closes-Bug: #1481495
Implements: blueprint containerize-dependencies
Change-Id: I3dfccbf9fafc06ffc36e78f3006fe5d3367891df
