发布

  • [OPIK-7487] [BE] fix: land multi-batch dataset file upload in a single version (#7613)

    frostbyte_neo 发布于 2026-07-27 15:39:21 +00:00

    File-based dataset upload (CSV/JSON) flushed each 1000-row batch through
    saveBatch, which called saveItemsWithVersion(..., null) and minted a new
    delta version per batch — a 10k-row file produced ~10 versions with O(n^2)
    carry-forward write amplification. Route the upload path through
    mutateLatestVersionWithInsert instead, so batch 1 creates the version and
    subsequent batches append into it, matching the null-batch_group_id
    "mutate latest" contract the array API already honors.

    Implements OPIK-7487.

    Co-authored-by: Claude Opus 4.8 (1M context) noreply@anthropic.com

    下载附件