Security¶
Pod security¶
Unless a node set, DatabaseServer or log sender sets them otherwise, the operator gives the pods it creates:
allowPrivilegeEscalation: falseon each container, except where a setting such asprivilegedrequires it.- The container runtime's default seccomp profile (
seccompProfile.type: RuntimeDefault). - No service account token (
automountServiceAccountToken: false), as none of them use the Kubernetes API.
Capabilities and the user aren't changed. The Stroom and MySQL images start as root to set up their volumes, then
switch to an unprivileged user, so dropping every capability or requiring runAsNonRoot stops them starting. If
your cluster enforces the Kubernetes restricted Pod Security Standard, set the security contexts yourself, with
images that support running as non-root throughout:
- name: data
podSecurityContext:
runAsNonRoot: true
runAsUser: 1000
fsGroup: 1000
securityContext:
capabilities:
drop: [ALL]
Operator permissions¶
The operator runs with a ClusterRole, as it manages resources in any namespace. It can get and create
Secrets, but not list, watch, change or delete them: it reads only the database password Secrets it uses, and
creates the generated DatabaseServer password Secret.
Database connections¶
The operator and Stroom connect to MySQL with TLS when the server supports it, which MySQL 8 does by default. Neither verifies the server's certificate. See External database.
DatabaseServer.spec.additionalConfig can't set the options the operator sets, or options that give access to
the server's files or bypass authentication. See Operator-managed DatabaseServer.
Network policies¶
The operator doesn't create NetworkPolicies, as which pods need to reach Stroom depends on your cluster: where your
Ingress controller runs, and whether data senders run inside it.
samples/operator-database/network-policies.yaml
is a starting point. It denies connections to pods in the Stroom namespace except:
- between the nodes of a
StroomCluster, - from the Ingress controller to the nodes' application ports (8080 and 8443),
- to the database server from the cluster's nodes, its backup jobs and the operator.
The nodes' admin port (8081), which serves health checks, is left reachable only from within the node's own pod and by the kubelet's probes, which most network plugins allow regardless of NetworkPolicies; check yours does. NetworkPolicies need a network plugin that enforces them.