Metallb be is a L2/L3 load balancer that we
use for exposing the Openstack services
outside the cluster.
Before that we used to deploy the ingress-nginx
contoller in the host network namespace.
Change-Id: I9fdb5f1b2f9403ce04f9d34b1792a0f29f55d879
Recently we refactored the deploy-env role which
among other things can deploy Openstack provider
network gateway.
Depends-On: I41f0353b286f817cb562b3bd59992e4baa473568
Change-Id: Iece2cc83c68cc282389f8380ceebeebf17f788fb
- Openvswitch agent init script skips attaching
interface if it does not exist. And the compute-kit.sh
deploys neutron with
auto_bridge_add: {"br-ex": "provider1"}
where "provider1" is a tap interface that is going
to be created while deploying the test env.
- Heat test script checks only public endpoints
- Add 1+2 nodes nodeset. The primary node is used
as a client node and not a member of K8s cluster.
Change-Id: If7c8763dd619dec31f9d141f21399d159395049a
Running dpdk test job on a sinlge node env
does not require a real hardware dpdk interfaces
to be attached to the tunnel bridge. Let's
make the list of dpdk interfaces empty which
will allow us to test the Openvswitch in user space
but not touch real hardware.
Change-Id: I2f9d954258451f64eb87d03affc079b71b00f7bd
Co-Authored-By: Arina Stebenkova <astebenkova@mirantis.com>
We recently re-worked all the deployment jobs
so they use `deploy-env` Ansible role which works
for both multi-node and single-node environments.
This means there is no need to have diffrent sets
of scripts for these two cases.
Also when we deploy Openstack components it is better
to have values overrides for different scenarios but
not different sets of scripts. Here we remove unused
deployment scripts which in many cases duplicated
the code base.
We will be cleaning up the code base even further to
provide excelent user experience.
Change-Id: Iacda03964a4dd0e60873593df9f590ce20504f2f
In Bobcat by default the ssh keypair is generated
with ed25519 algorithm which is incompatible with
Cirros 0.3.5. The Cirros 0.6.2 also requires minimum
128Mi of RAM
Change-Id: I0135c09a9ae5bc3171891576b29a450f7000f180
The issue was that we used to add buildset registry
name to the /etc/hosts file only on the primary node.
This PR fixes this.
Change-Id: I48a861c7c6661e3fe1e7384f4b987f1646d652d7
We recently merged https://review.opendev.org/c/openstack/openstack-helm-infra/+/891720
which updates ingress controller deployment and assumes
we now use IngressClass objects instead of annotations.
We have to update the ingress deployment script
so it creates IngressClass objects with correct names.
Change-Id: I547f27250ba21bb068bc83c62954092111df8b50
The change updates all deployment jobs so they use
deploy-env role which leverages kubeadm to deploy k8s.
This role works for both single-node/multi-node
inventories.
Also all jobs are reorganized to improve job
maintenance. Check pipeline runs tests for 3
most recent releases: Yoga, Zed, 2023.1
We are focusing on 2023.1 for which we run both
Focal and Jammy jobs.
Change-Id: Ibba9b72876b11484fd7cc2e4710e92f964f15cc3
At this point it requires nfs provisioner that provides
ReadWriteMany volumes for vnfpackages, csar files and
also the same storage class is used for logs.
Also this patch adds a job that only deploys Tacker but
does not tests it in any way. This job is put to the experimental
pipeline.
Co-authored-by: Vladimir Kozhukalov <kozhukalov@gmail.com>
Story: 2010682
Task: 47771
Change-Id: I56d7ba489746ab4f818086440a7783f4b1ecb292
The issue is that playbook that configures containerd
defines buildset registry namespace w/o skip_verify
parameter which leads to the error in the template.
This change add additional check whether skip_verify
parameter is defined or not.
Change-Id: I7b6f19d4fe881e33fef2370bac887e9cd66c7b35
The compute-kit jobs are used to test new images
which are published to buildset registry. We have
to configure containerd which is used for multinode
compute-kit jobs to use this buildset registry.
The role use-buildset-registry that we used before
does not properly configure containerd. So we
extended deploy-docker playbook to configure
both buildset registry and registry mirror
if they are defined.
Change-Id: Idb892a3fcaf51385998d466dbdff8de36d9dd338
For recent releases we use 32GB nodes for compute-kit
jobs. The number of such nodes is extremely limited.
So we'd better use multinode nodesets for compute-kit
jobs.
We deploy K8s using kubeadm and then we set labels to the
K8s nodes so charts can use these labels for node selectors.
We deploy L3 agent only on the node where we run test scripts.
This is because we want test virual router to be always created
on this node. Otherwise the L2 overlay needs to be created
to emulate provider network (will be implemented later).
Glance is deployed w/o backend storage (will be fixed later).
Change-Id: Id2eb639fb67d41006940a7d7b45a865b2f1124f7
Cgroup name of Kubernetes is determined by the
cgroup driver of Docker. If cgroup driver of Docker
is systemd, there will be an error in deploying HugePages.
https: //github.com/kubernetes/kubernetes/blob/release-1.26/pkg/kubelet/cm/cgroup_manager_linux.go#L77-L81
Change-Id: If709a968dde24d01e17fef63f09ebf9d7a23fc62
This change updates all Ceph image references to use Focal images
for all charts in openstack-helm.
Change-Id: I67cd294e2aabf3c3af404da42204f9b6157b06f7
Beginning with the Pacific release, Ceph pools are not allowed to
use 1x replication by default. This is problematic for the
openstack-helm gate scripts, which frequently use 1x replication
for automated testing. This change adds Ceph configuration and
command overrides to allow those gate scripts to continue to use
1x replication for testing.
Change-Id: I21ed3e43f3773d5ea830959f1b66b35f38185ca7
When deploy ceph on loop devices we need lvm2
to be installed on the host to create necessary
device links like /dev/<vgname>/<lvname>
Change-Id: Ice7031da38009e913cd43b95398218d25dc97930
The ceph-rbd-pool job is being changed to use the "final_osd"
target instead of the "osd" target to determine if more OSDs than
expected are running. The deployment test scripts therefore must
be updated to include this value.
Change-Id: I63e4a0b3ba58d5ee79cda75bd9e9650d5a49aff5
Occasionally rabbitmq replicas fail to come up when running in
zuul due to resource constraints. This causes random failures
and needless rechecks when it comes up.
This change sets the number of replicas when deploying the compute
kit script to 1 from the default 2.
Change-Id: Icdcc075a88d05f87a7f787d306d6b430485964c9
With the placement db migrate job removed, we can now deploy
placement ahead of nova. This fixes issues of several nova pods
failing due to placement not being up before nova is.
This change moves the placement service to deploy before nova
in the compute-kit script.
Depends-on: https://review.opendev.org/c/openstack/openstack-helm/+/860975
Change-Id: I62f82224e898761d5b388867c63a259967601d45
Before if we tried to add an address that was already added to the
br-ex device the script would error out. Now, use replace which is
idempotent.
Change-Id: I34bbb361c4fa2f347431629f79753f75b458926f
Vladimir Kozhukalov