internal/xdscache: only update ClusterLoadAssignment cache on changes (…

…projectcontour#3092)

Only updates the ClusterLoadAssignment xDS cache and notifies via the
condition variable if the contents have changed during recalculation.

Closes projectcontour#3053.

Signed-off-by: Steve Kriss <krisss@vmware.com>

internal/xdscache: only update ClusterLoadAssignment cache on changes (…

…projectcontour#3092)

Only updates the ClusterLoadAssignment xDS cache and notifies via the
condition variable if the contents have changed during recalculation.

Closes projectcontour#3053.

Signed-off-by: Steve Kriss <krisss@vmware.com>

Add conflict retries.

Signed-off-by: James Peach <jpeach@vmware.com>

internal/k8s: consistently use the controller-runtime informer cache

Prior to this change, the Kubernetes client API exposed methods to
create a informer factories as well as namespaced informer factories.
This encouraged other systems to use multiple factories and multiple
informers for each factory. The result of this is that objects could be
cached multiple times in the different informer factory caches, and that
the informer states could be subtly out of sync.

This change removes the internal APIs that expose informer factories.
There is now just one API to create an informer, and it is backed by the
controller-runtime cache API. This ensures that all the informers used
across Contour share a common cache and therefore a consistent view of
Kubernetes state.

This updates projectcontour#2683.
This fixes projectcontour#2904.

Signed-off-by: James Peach <jpeach@vmware.com>

contour 0.15.1

Contour 1.0.0 beta 1

contour 0.15.0

contour 0.14.2

contour 0.14.1

contour 0.14.0