2. Motivation

As the domain name system continues to adapt to new uses and changes in deployment, polling has the potential to burden DNS servers at many levels throughout the network. Other network protocols have successfully deployed a publish/subscribe model following the Observer design pattern [OBS]. Extensible Messaging and Presence Protocol (XMPP) Publish-Subscribe [XEP0060] and Atom [RFC4287] are examples. While DNS servers are generally highly tuned and capable of a high rate of query/response traffic, adding a publish/subscribe model for tracking changes to DNS records can deliver more timely notifications of changes with reduced CPU usage and lower network traffic.

The guiding design principle of DNS Push Notifications is that clients that choose to use DNS Push Notifications, instead of repeated polling with DNS queries, will receive the same results as they could via sufficiently rapid polling, except more efficiently. This means that the rules for which records match a given DNS Push Notification subscription are the same as the already established rules used to determine which records match a given DNS query [RFC1034]. For example, name comparisons are done in a case-insensitive manner, and a record of type CNAME in a zone matches any DNS TYPE in a query or subscription.

Multicast DNS [RFC6762] implementations always listen on a well-known link-local IP multicast group address, and changes are sent to that multicast group address for all group members to receive. Therefore, Multicast DNS already has asynchronous change notification capability. When DNS-based Service Discovery [RFC6763] is used across a wide area network using Unicast DNS (possibly facilitated via a Discovery Proxy [RFC8766]), it would be beneficial to have an equivalent capability for Unicast DNS in order to allow clients to learn about DNS record changes in a timely manner without polling.

The DNS Long-Lived Queries (LLQ) mechanism [RFC8764] is an existing deployed solution to provide asynchronous change notifications; it was used by Apple's Back to My Mac [RFC6281] service introduced in Mac OS X 10.5 Leopard in 2007. Back to My Mac was designed in an era when the data center operations staff asserted that it was impossible for a server to handle large numbers of TCP connections, even if those connections carried very little traffic and spent most of their time idle. Consequently, LLQ was defined as a UDP-based protocol, effectively replicating much of TCP's connection state management logic in user space and creating its own imitation of existing TCP features like flow control, reliability, and the three-way handshake.

This document builds on experience gained with the LLQ protocol, with an improved design. Instead of using UDP, this specification uses DNS Stateful Operations (DSO) [RFC8490] running over TLS over TCP, and therefore doesn't need to reinvent existing TCP functionality. Using TCP also gives long-lived low-traffic connections better longevity through NAT gateways without depending on the gateway to support NAT Port Mapping Protocol (NAT-PMP) [RFC6886] or Port Control Protocol (PCP) [RFC6887], or resorting to excessive keepalive traffic.