MySQL Group Replication connects database servers that must agree on membership and transaction order. The group communication channel carries more than query traffic. It carries replication state, failure detection, and membership changes. Encrypting that channel prevents passive disclosure, but a secure group also needs a trustworthy answer to which nodes are allowed to join.

MySQL 8.4 documentation makes the default easy to miss: group_replication_ssl_mode defaults to DISABLED. REQUIRED encrypts the connection when the server supports it. VERIFY_CA also validates the peer certificate against configured certificate authorities. The setting must be consistent across members, or some members may fail to join. That consistency is an operational control, not a cosmetic TLS preference.

The documentation also covers distributed recovery connections. A node that has fallen behind may need a full state transfer from a donor. Protecting only the steady-state group traffic while leaving recovery connections weak creates an uneven security posture. Configure the server TLS material and the recovery path together. Then test node addition, donor selection, certificate renewal, and recovery after a long outage.

A certificate check without a useful identity check is still incomplete. If every database server trusts a broad CA and accepts any certificate from it, an unrelated workload with a certificate from the same PKI may be able to reach the group port. Use a dedicated CA or an intermediate with a narrow purpose. Put the expected DNS names or other identities in the server certificates. Restrict the group communication port to database hosts. Test that a certificate with the right CA but the wrong subject is rejected.

Network filtering remains relevant. The group port is commonly documented as TCP 33061, but changing the default does not create security by itself. An allowlist should name the database hosts and management jump paths, not an entire application subnet. Keep the MySQL client port, the group communication port, and the distributed recovery path in the same inventory even when they use different policies.

Certificate rotation deserves a real runbook. Group membership can become unstable if one node drops its old certificate before peers trust the replacement. Stage the new CA, distribute trust, replace leaf certificates in a controlled order, and verify group health after each step. Keep the old trust anchor until every member has completed the transition. Remove it only after the rollback window closes.

The same discipline applies to automation. A provisioning system that can request a certificate for any hostname in the organization can create a valid node identity. Limit that permission to the database platform role. Record who requested the certificate and which group it was intended to join.

Group Replication is designed for closely coupled database servers. That makes a small identity mistake expensive. Treat membership as a privileged operation, verify both encryption and peer identity, and rehearse the recovery path before an outage forces the issue.

References