During cold start we pick leader node by seqno. When node is running
of finished non gracefully seqno may stay as -1 unless periodic task
update its based on local grastate.dat or will detect latest seqno via
wsrep_recover. This patch adds an unfinite waiter to leader election
function to wait unless all nodes report seqno different that -1 to make
sure we detect leader based on correct data.
Change-Id: Id042f6f4c915b21b905bde4d57d40e159d924772
Sometimes the pod fails to terminate correctly,
leaving zombie processes. Add option to use tini
to handle processes correctly. Additionally update
log-tail script to handle sigterm and sigint.
Change-Id: I96af2f3bef5f6c48858f1248ba85abdf7740279c
Recently we switched from Deployment to Statefulset
to make it possible to work with memcached instances
directly w/o load balancer. The strategy field is not
valid for statefulsets, so here we remove it.
Change-Id: I52db7dd4563639a55c12850147cf256cec8b1ee4
This commit adds recommended kubernetes name label to pods definition.
This label is used by FluxCD operators to correctly look for the
status of every pod.
Change-Id: I866f1dfdb3ca8379682e090aca4c889d81579e5a
Signed-off-by: Johnny Chia <johnny.chialung@windriver.com>
This will help other services to access to OVN DB.
So services like Octavia can use OVN Octavia provider agent.
Change-Id: Iddaa6214ece63a5f1e692fe019bcba1b41fdb18f
* Allow to pass custom helm chart version during build like
make all version=1.2.3+custom123
* add get-version target that allows to get version based on
number of git commits in format <git-tag>+<commits number>
Change-Id: I1f04aeaa8dd49dfa2ed1d76aabd54a0d5bf8f573
This change add the ability to add fqdn alias to namespace and cluster ingress resources. This change is specifically required for keystone so HA of backup solution can be implemented.This change allows user to specify alias_fqdn in the endpoints section, and user can have alias configued. This change is backward compatible, so without specifying this option in charts gives one fqdn ingress rule without cname alias as default behaviour.
Change-Id: Ib1c60524e2f247bb057318b1143bfbc3bde5b73a
Rabbitmqcluster does not work with short node names, as
there is unresolvable dependency in dns resolution, it is
not possible to resolve only pod name svc must be added.
This reverts commit bb7580944a5268a1e5f7fcd195b156f53dc668c5.
Change-Id: I42b25ba4f569bae94bbc2939a1022bd14e66e527
Recently we fixed the libvirt.sh script
and removed the conditionals cgroup commands
which were introduced for smooth transition
to Jammy and cgroups v2
https://review.opendev.org/c/openstack/openstack-helm-infra/+/929401
But because we didn't have overrides for 2023.1
we used to run 2023.1 with the default libvirt image
openstackhelm/libvirt:latest-ubuntu_focal
which does not work with cgroups v2 on the host
system with this recent fix (see above).
So the 2023.1 Ubuntu Jammy compute-kit test jobs fails.
This PR fixes this job by means of introducing
explicit image overrides for 2023.1.
Change-Id: Ie81f8fb412362388274ea92ad7fa5d3d176c0441
Some applications require perisitant volumes to be stored
on the hosts where they running, usually its done via
kubernetes PV. One of PV implementations is local-volume-provisioner [0]
This patch adds helm chart to deploy LVP. Since LVP creates a volumes for
each mountpoint, helm chart provides a script to create mountpoints
in the directory, which later exposed to kubernetes as individual volumes.
Change-Id: I3f61088ddcbd0a83a729eb940cbf9b2bf1e65894
For effective cache use all endpoints should be specified
explicitly as memcache client use specific algorithm to
identify on which cache server key is stored based on
servers availability and key name.
If memcached deployed behind the service unless same key is
stored on all memcached instances clients will always got
cache misses and will require to use heavy calls to database.
So in the end all keys will be stored on all memcached instances.
Furthermore delete operations such as revoke token or remove
keystone group call logic in service to remove data from cache
if Loadbalancer is used this functionality can't work as we
can't remove keys from all backends behind LB with single call.
Change-Id: I253cfa2740fed5e1c70ced7308a489568e0f10b9
Add job that waits when initial bootstrapping of cluster is completed
which is required to pause db creation and initialization when cluster
is not fully bootstrapped.
Change-Id: I705df1a1b1a34f464dc36a36dd7964f8a7bf72d9
Ingress deployment is not used for a while and there are
more elegant ways to provide same functionality based on
controller to pick up master service.
Remove ingress deployment completely.
Change-Id: Ica5d778f5122f8a4f0713353aa5e0ef4e21c77f8
Use virsh connect instead of list which is heavy and may
stuck for a while when libvirt creating domains.
Change-Id: I515c70b0b3a050599726ca2548eeeb7fd3f3e6ea
