- Adding cinder-volume on infra-hosts once ceph is enabled
- Fix missing osa_no_containers default value of false
Change-Id: Id7898c9dff84cbb981c9c0c916a65abcc5256dfd
This patch updates various playbooks within the MNAIO to use ansible
collections, which is required for the latest Ansible versions.
Change-Id: Ia7694e6b182dc8898b56730537cc99759a722b20
This patch removes legacy support for 14.04/16.04/18.04
on the deploy node and moves the default deploy to
Xena on 20.04 LTS. Root disk size has been bumped to support
upgrades (8 GB -> 12 GB).
Change-Id: I81a13464b9daa90090cb380e2b0d89e5eb8fe89a
The path for get-pip.py script in versions 3.2 and 3.4 have been changed
with this commit [1].
[1] 2360f025eb
Change-Id: Iedef41b53c4e5fa28dac8206c5ba39dab5033815
This patch implements support for deploying an MNAIO with
Open vSwitch and DVR.
Change-Id: I0fb03e2eb0ead198c64019eb0cdd06451e1e7c94
Implements: openvswitch+dvr
This patch updates the bootstrap script for the MNAIO to use python3
rather than python2.7. This only impacts the deploy node and not
deployed VMs.
Change-Id: I41b5d746a8f31018e15bd282b7a358d50610eabb
Appears we ran into this bug on 16.04:
https://bugs.launchpad.net/ubuntu/+source/debian-installer/+bug/1816876
and the fix is in xenial-updates. Switching all the ubuntu installer
kernels to the updated ones so that we can make sure we're using
latest and greatest.
Change-Id: I68cccbb142c8b4c657a60929ad21e8164d07a1ef
This commit adds the ability for users to enable Designate (DNSaaS)
in a MNAIO build. This service is disabled by default.
Change-Id: I36e10922b6fe8e5cba3cc929e2b91b59507c210d
This patch adds support for deploying the ML2/Open Virtual Network (OVN)
plugin for Neutron to an MNAIO deployment. A new var,
osa_enable_networking_ovn, can be set to lay down the appropriate bits.
Change-Id: Ib5bd4e0c20be62ddbf0bff13c91d1918907bf230
NVMe partitions generally show a `p` before the partition number,
causing the existing data disk partitioning script to fail. The
new var support changing the suffix to support alternative naming
schemes. The default remains '1' for backwards compatibility.
Change-Id: I7c0039ee13ba301c152424fc0b327195ec6a4367
vmx flag was not passed in to the guest, so test VMs failed to boot on
compute2. With host-passthrough the cpu flags are unfiltered.
Change-Id: I1a4607ba5bc96c92512a2f4aa43df51f7eef6b27
On occasion, a set of saved images may not have a gvwstate.dat file.
Given that the MNAIO tooling is for test purposes only, we can take
the risk of just making one up in order to ensure that the cluster
starts on boot. DO NOT DO THIS IN PRODUCTION!
Change-Id: I1b055ef0d02d77afe03d1a90baa2e6890e986c32
In order to allow testing within OpenStack-CI, we need to cater
for the ability to run the tooling without a secondary data disk.
This patch changes the tooling to try the detection, and if there
is no secondary disk it will just use the root disk for /data/images.
Change-Id: I56d76e68b6914b3e12948b022e0e702557684c50
This patch resolves the warning about using jinja braces
the conditional for the guestfish workaround task. While
doing so, the task is made idempotent and we also ensure
that the task that follows exits immediately on any errors.
Change-Id: I7ec254a96312e07229b0a5c799d4a98b5d9c8957
This commit enables overriding the default storage options of
cinder/swift to a Ceph backed storage for the multinode all in one
deployments.
We also correct the README and build.sh script to show/use the
current defaults correctly for VM_DISK_SIZE and INFRA_VM_SERVER_RAM.
Change-Id: I9e1f1b09d1bcf224f4afa765c585baf28e6cafa8
With the switch to systemd-networkd the veth pairs for flat networking
and lbaas aren't created. This creates veth pairs if the server has the
appropriate server_networks defined.
* Create the veth network files
With this simple Match directive networkd will bring the interface up.
Change-Id: I0ae4375be128594a861c18aa33a98378859dd9d3
Due to [1] the task 'Create data cinder-volumes group' is currently
failing when run on a restored image due to the idempotence failure.
In this patch we enable pinning the requirements file and plugins
repo to a different OSA branch, and set it to stable/rocky which is
currently pinned to Ansible 2.5.10 and does not exhibit this issue.
[1] https://github.com/ansible/ansible/issues/47301
Change-Id: Id0f98f3ea0f53173dae800d45f1a4554c8a0c38a
In https://review.openstack.org/611582 we removed
the legacy group as it has been deprecated since
Newton, however it appears to still be used by some
downstream tests, so we add it back, but make it
only get implemented if the associated services
are enabled.
Change-Id: I477a46d606d75d44a1ecd5bcfcb29c8308c65245
The default of 1 GiB isn't sufficient to handle the logging load
generated on a MNAIO. This is a workable value and can be adjusted
down.
Change-Id: I451af0babf9ae7bd8ae58eccef9c908bb7f91f52
1. If osa_enable_block_storage=false, then there should
be no cinder hosts deployed, nor any presence of the
vars related to it in openstack_user_config.
2. If osa_enable_compute=false, then there should be no
compute hosts deployed, nor any presence of the
vars related to it in openstack_user_config..
3. If osa_enable_object_storage=false, then there should
be no swift hosts deployed, nor any presence of the
vars related to it in openstack_user_config..
Change-Id: Id6858d277c80095024af5d8e04dfc97cc3e3b253
The comment is a bit confusing without the rest of the content,
so we move it inside the conditional so that it only appears
if the ELK deployment is enabled.
Change-Id: I9da35b848d6c52a31162361f98c8514d8e1103ea
The default lxc_container_backing_store for newton onwards
is 'dir' anyway, so this override is pointless.
Change-Id: I931a8dca2186dc51b468ec4fc1fc67859dadb171
The glance_default_store variable is automatically set
by OSA's group_vars to 'swift' if any swift hosts are
present, and it falls back to 'file' if no swift is
present. This override therefore interferes with tests
which are not using swift, and it is totally unnecessary.
Change-Id: I5f938788f7998f0ca763d7395dc11c84ba85d2ee
The current default of 8GB simply does not work for a default
deployment of Rocky/Stein without doing a severe amount of
worker tuning.
While we do some work to improve the default number of workers,
we should rather increase the default value for these VM's.
Change-Id: Ic31bd9145e7feb67fa5eacc602c6152ecde1a47c
The legacy group 'os-infra_hosts' is not actually part
of the infrastructure - it includes all the openstack
infrastructure groups (keystone, nova, neutron, glance,
heat, swift). This group's use is unnecessary because
all the other groups included are represented with
individual options.
Having this in the osa_enable_infra conditional also
means that if you mean to disable swift/heat, it does
not work and you end up with a broken deployment.
Change-Id: Icd80fd96aad713372b1fe21752799d56ada3dac4
This commit enables the openstack Octavia load balancing service
if the option is enabled. The Octavia service replaced the legacy
Neutron LBAAS service.
Change-Id: Ib820ec3c4a7f6c9116608140b59332d03cf4c451
When starting an imaged environment, the SSH keys on the host do
not match the SSH keys in the containers. In this patch we ensure
that if the 'all_containers' group is present, then the deploy-vms
playbook will wait for them to come up and copy the new keys into
them. It also adds the new public key into the authorized_keys.
This ensures that the whole environment is ready to be used just
as it was before imaging.
We also remove the 'port' argument given to the wait_for_connection
module because the argument is invalid.
Change-Id: Iff0a3327c2031e0dd977e8e403b417e495fee14f
This play is duplicated, so we remove the second execution
and ensure that the first uses a local connection.
Change-Id: I6162e93c999a792cbcdc42bf605271b999327e08
There are times when it's useful to run playbooks against the
OSA inventory, but having to do so via infra1 is a bit of a
pain. Rather than do that we add a script which fetches the
OSA inventory file from infra1, and we also bootstrap the OSA
plugins repo into the default ansible plugins path so that it
is possible to run playbooks against the containers in the VM's
from the host without needing to do anything else.
Change-Id: I751af19841ea825d21dd67f4343674b8522b689f
The MNAIO tooling is a test system, and with that target
we can afford to be opinionated in the implementation
rather than try to cater to every possibility.
Now that the file-backed VM implementation has matured,
we can switch to using it exclusively. This cuts down on
code complexity and allows us to mature the implementation
further with more options without having to cater to two
options.
Change-Id: Ibe04b5676a392301cd79a5d290b77df4c7d9f79a
* Set the logstash_queue_type as it detection is not fullproof on the
MNAIO.
* Add the elastic.co repositires to the DontCache stanza for ACNG.
Change-Id: I1a826f9949e3f4b83b2defff6b715518b45ce78c
If the log path doesn't exist, we should create it - otherwise
the download command will fail because it cannot create the log.
Change-Id: I8f3d7604da11aae89c83ec3bd5c28b05f9853324