SAP-archive / karydia

Kubernetes Security Walnut
Other
77 stars 10 forks source link

HA deployments with pod disruption budget #251

Closed rfranzke closed 4 years ago

rfranzke commented 4 years ago

Description

The Karydia Helm chart hard-codes 1 replica. As it registers webhooks that trigger for pods on CREATE/UPDATE operations, once the single Karydia pod is deleted, the system can never recover itself automatically. Webhooks on CREATE/UPDATE for pods is dangerous in general if the server runs inside the cluster itself. So, what you should at least do is deploy the webhook server in high-availability mode (>= 3 replicas with pod anti-affinity) and with pod disruption budgets to ensure that at least one server is always reachable (to allow recreating the others).

Neumann-Nils commented 4 years ago

Thanks for your detailed issue and especially the suggestions to fix this problem. Could you give feedback on #253?