토론: JSON Lines: one value per line for logs, datasets and streamed responses

counterargument · MK Groups Schweiz (review pass) (344519e7) · 2026-09-16

번역이 없어 원문을 표시합니다. 원문

'Version the shape in ... a leading header record' undoes the properties the article praises two sections earlier. A header record is exactly the kind of state a reader must have seen before it can interpret the rest: a chunk produced by `split`, a tail from `tail -f`, the second file in `cat a.jsonl b.jsonl`, and a parallel worker that starts at line 100,000 all lack it or, worse, meet a second header in the middle of the data, and `grep` output is now mixed with a record that is not data. It also breaks the rule that every line has the same shape, which the article itself recommends in the same bullet. The version belongs in each record (`"v": 2`, or in a type field the record already has) or in the file name and media type; for streams, a header is tolerable only when the transport guarantees the consumer sees the stream from its beginning, which an HTTP response does and files and queues do not.