cursus

Cluster transport security

Distributed mode now fails configuration validation unless both of these are configured:

allow_insecure_cluster_transport: true is an explicit escape hatch for isolated test fixtures only. It must not be used for production.

Upgrade constraint

TLS Raft framing is not wire-compatible with the previous plaintext Raft transport. Do not perform a mixed plaintext/TLS rolling upgrade. Validate all certificates and advertised DNS names first, stop the cluster cleanly, update all members to the secured build and configuration, then start the complete quorum. Verify discovery membership, Raft leadership, in-sync replicas, group registration, committed offsets, and readiness before restoring client traffic.

Each broker certificate must be signed by the configured internal CA and valid for the advertised hostname used by peers. The configured internal_tls_server_name must match the certificate SAN. A discovery bind failure, TLS validation failure, Raft initialization error, or missing security configuration prevents readiness instead of falling back to plaintext.

The Kubernetes production chart treats the public broker listener as a separate security boundary. It requires a client-listener TLS Secret and a SASL user Secret, enables both controls on port 9000, and rejects an insecure render unless the test-only clientSecurity.allowInsecure=true escape hatch is explicitly set. Internal mTLS credentials do not grant client API access.