fix: do not require pod readiness when switching desired service sele…

…ctor on abort (#3338)

* do not switch service selectors back when using alb due to race between two controllers with pod readiness gates

Signed-off-by: Zach Aller <zachaller@users.noreply.github.com>

* update tests for alb

Signed-off-by: Zach Aller <zachaller@users.noreply.github.com>

* lets not check for readiness instead

Signed-off-by: Zach Aller <zachaller@users.noreply.github.com>

* clean up notes

Signed-off-by: Zach Aller <zachaller@users.noreply.github.com>

* fix /

Signed-off-by: Zach Aller <zachaller@users.noreply.github.com>

---------

Signed-off-by: Zach Aller <zachaller@users.noreply.github.com>

rollout/trafficrouting.go

@@ -179,13 +179,15 @@ func (c *rolloutContext) reconcileTrafficRouting() error {
 			desiredWeight = c.calculateDesiredWeightOnAbortOrStableRollback()
 			if (c.rollout.Spec.Strategy.Canary.DynamicStableScale && desiredWeight == 0) || !c.rollout.Spec.Strategy.Canary.DynamicStableScale {
 				// If we are using dynamic stable scale we need to also make sure that desiredWeight=0 aka we are completely
-				// done with aborting before resetting the canary service selectors back to stable
-				err = c.ensureSVCTargets(c.rollout.Spec.Strategy.Canary.CanaryService, c.stableRS, true)
+				// done with aborting before resetting the canary service selectors back to stable. For non-dynamic scale we do not check for availability because we are
+				// fully aborted and stable pods will be there, if we check for availability it causes issues with ALB readiness gates if all stable pods
+				// have the desired readiness gate on them during an abort we get stuck in a loop because all the stable go unready and rollouts won't be able
+				// to switch the desired services because there is no ready pods which causes pods to get stuck progressing forever waiting for readiness.
+				err = c.ensureSVCTargets(c.rollout.Spec.Strategy.Canary.CanaryService, c.stableRS, false)
 				if err != nil {
 					return err
 				}
 			}
-
 			err := reconciler.RemoveManagedRoutes()
 			if err != nil {
 				return err