On Fri, 28 Jun 2019 at 09:03, <
josh.f...@gmail.com> wrote:
>
> For myself the only requirements have been to apply default deny for egress traffic across all namespaces, log those denies(which isn’t a function of NetworkPolicies supports now). We have additional global policies that then allow egress to the Kubernetes API service or the kube-system namespace.
We've had similar discussions with users. In the case of CIlium, we
chose to implement the global default deny with an agent flag to avoid
solving the race condition that the policy must be loaded before the
first workload is scheduled onto the cluster.
The main discussions we had with users regarding the potential
introduction of global policy is visibility. While RBAC may provide a
user with access to a particular namespace which has NetworkPolicy
loaded, it can lead to confusion if additional policies are in scope
which are not visible to that user for any communication within that
namespace.