A single WAL file has a problem: it grows without bound. Once it reaches gigabytes, seeks become slow and cleanup is impossible without rewriting the entire file.

The solution: segment files. When a WAL segment exceeds a size threshold, seal it (make it immutable) and open a new active segment. An offset index enables O(1) lookups by mapping each log offset to the correct segment file and byte position.

This is how Kafka, etcd, and most production systems organize their logs:

1. Segments are named by their starting offset (e.g. 00000000.log, 00001000.log)
2. Each segment has a companion .index file mapping offset -> byte position
3. Old sealed segments can be deleted, compressed, or archived independently
4. The active segment is the only one receiving new writes

Request:  {"type": "wal_segment_config", "msg_id": 1, "max_segment_bytes": 67108864}
Response: {"type": "wal_segment_config_ok", "in_reply_to": 1, "max_segment_bytes": 67108864}

Request:  {"type": "wal_segment_info", "msg_id": 2}
Response: {"type": "wal_segment_info_ok", "in_reply_to": 2, "segments": [
    {"file": "00000000.log", "start_offset": 0, "end_offset": 999, "size_bytes": 67108000, "sealed": true},
    {"file": "00001000.log", "start_offset": 1000, "end_offset": 1050, "size_bytes": 5120, "sealed": false}
], "active_segment": "00001000.log"}