skypilot-org/skypilot

[GCP] Move from `gsutil cp/rsync` to `gcloud storage cp/rsync`

Open

#8,457 建立於 2025年12月30日

在 GitHub 查看
 (3 留言) (0 反應) (0 負責人)Python (4,859 star) (311 fork)batch import
good first issuegood starter issues

描述

According to https://docs.cloud.google.com/storage/docs/working-with-big-data, gcloud storage cp is much faster due to optimizations in automatic parallelism. We can migrate the usage gsutil cp in https://github.com/skypilot-org/skypilot/blob/e6a41b121ae27c33e1815ecf316858752dcc9982/sky/data/storage.py#L2370.

I manually tried a few file downloads with gcloud storage cp, it easily saturates the 300MB/s throughput, i.e. it worth a try.

貢獻者指南

技術棧
pythonshellgcp
領域
cloudperformance
議題類型
performance
難度面向新貢獻者的預計實作難度，1 表示很小改動，5 表示專家級工作。
3
預計時間有經驗貢獻者完成調查、實作、測試並準備 pull request 的粗略時間範圍。
1-3 hours
活動狀態議題目前的可參與程度：新鮮、活躍、陳舊、阻塞或等待維護者輸入。
fresh
清晰度議題是否清楚說明預期改動、驗收標準和下一步。
clear
前置要求
Basic PythonGCP storage concepts
新手友善度1-100 的估計分數，表示該議題對首次貢獻者的友善程度。
70
研究方向
Examine the file sky/data/storage.py at line 2370 to understand how gsutil cp/rsync is currently used. Then, replace those commands with gcloud storage cp/rsync following the reference mentioned in the issue (https://docs.cloud.google.com/storage/docs/working with big data). After making the change, test the performance improvement by downloading files with gcloud storage cp and compare throughput. No additional changes to the logic are required.