Directory behavior

The directory consumes every Kafka partition independently on each DNS instance, replaying from the beginning on process restart. It does not join a consumer group for load balancing: each replica must see the whole directory. Partitions added later are discovered within 60 seconds. Expired and older updates are discarded. Kafka remains the deployment's durable replay source; this service adds no database.

The only accepted registration message is rpt_protobuf.dns_update_pb2.dns_update, with the existing shared nested node fields. Legacy messages have no decoder here. An entry expires at timestamp + node.ttl seconds. DNS responses cap TTL at 60 seconds and the entry's remaining validity. The bounded directory defaults to 10000 entries; choose an adequate capacity for your network. Invalid addresses, IDs, ports, timestamps, lifetimes, or missing metadata are rejected.

Configuration: RPT_DOMAIN, RPT_PORT (default 5054 TCP and UDP), RPT_CACHE_SIZE, RPT_MAX_LIFETIME (86400 seconds), RPT_CLOCK_SKEW (30 seconds), RPT_WORKERS (32), and nested Kafka settings. Only IPv4 registrations and A/TXT/SRV responses are implemented. Domain names must match the configured suffix exactly. TCP frames are read incrementally with timeouts. Worker admission is bounded; configure shared network limits at ingress for public deployments.