override pre-existing anti-affinity

commit f7a37e6a, "Removing default injector anti-affinity rules",
disables anti-affinity for the injector pod.  This is sufficient for
future application updates.  However, during application-update an old
pod that still has anti-affinity will still prevent scheduling of a new
pod.  This is observed on AIO-SX when testing application-update in
preparation for USM.

Using injector.strategy.rollingUpdate.maxUnavailable
(DeploymentStrategy) set to 100% the old pod will immediately terminate
while the new pod waits for its termination.

This is the workaround described in the original in starlingx bug:
https://bugs.launchpad.net/starlingx/+bug/2030901.

Test Plan:
PASS  AIO-SX vault sanity
PASS  application-update

Partial-Bug: 2030901
Story: 2011073
Task: 50484

Change-Id: I66fe336ece7f1ccd68caa665aabc693f1b9a5c18
Signed-off-by: Michel Thebeau <Michel.Thebeau@windriver.com>
This commit is contained in:
Michel Thebeau 2024-01-04 20:04:01 +00:00
parent 72d7b82fdf
commit f994829fcb

View File

@ -14,6 +14,9 @@ injector:
nodeSelector: |
node-role.kubernetes.io/control-plane: ""
affinity: null
strategy:
rollingUpdate:
maxUnavailable: 100%
image:
repository: hashicorp/vault-k8s
tag: 1.2.1