Which strategy minimizes log noise for non-fatal TLS alerts in OpenSSL?

OpenSSL utilizes TLS alerts to signal warnings or fatal errors during the record layer process. While fatal alerts must trigger immediate connection termination to maintain security, non-fatal warnings can generate significant notification noise in high-traffic environments, potentially leading to log exhaustion.

The library provides SSL_set_msg_callback to monitor alert traffic, but there is a design tension between maintaining full protocol visibility for debugging and suppressing ignorable warnings to protect system resources. This is particularly relevant when managing asynchronous socket shutdowns and close_notify alerts.

What is the recommended approach to filter non-fatal alerts without masking critical protocol failures? Does OpenSSL provide a native configuration to categorize alert severity before they reach the application logging layer?