Skip to content

Server Pulse v2.1.1

Choose a tag to compare

View all tags
@LuSeptem LuSeptem released this 29 Aug 11:25
· 10 commits to main since this release
v2.1.1
3ce62b6

v2.1.1 — 2026-08-29

中文

变更

  • 冻结「按用户磁盘归因」功能（基于 find 的扫描在远端服务器上非常占用内存和磁盘，计划后续正式下架）：
    • 远端 agent 不再调度或部署每日 find 扫描：所有代理配置恒以 scan_enabled=0 写入；以旧配置部署的已有代理在执行一次「重启」「注入」或保存配置后停止调度。
    • 卡片「立即扫描」入口、按用户磁盘归属弹窗、History 页按用户磁盘曲线全部隐藏；各窗口 5 分钟归属自动刷新关闭。
    • 记录合并不再拉取/合并归属记录，历史查询不再返回归属数据。
    • 实时磁盘容量监控（DISK 行、挂载点列表、每挂载点使用率曲线）不受影响。
    • 本地与远端已有的归属历史原样保留，不清理、不删除。
    • 实现方式：代码级功能开关——Rust 层 serverpulse_core::DISK_ATTRIBUTION_FROZEN + 前端 DISK_ATTRIBUTION_FROZEN，代码完整保留、翻转开关即可恢复；正式下架时整体删除相关代码路径。

修复

  • 修复历史查询内存占用过高的问题（快速采样下历史查询可冲到 3–4 GB）：日文件改为按时间戳预过滤的流式读取，不再整文件载入并逐行构建值树；去重改用结构化记录哈希，不再为每条记录保留再序列化副本。
  • 本地历史记录现在每台服务器每个 UTC 分钟最多落盘一行（实时卡片仍按所选采样间隔刷新），历史日文件体积缩小约 40 倍，也降低了对远端与本地的磁盘写入压力。

English

Changed

  • Per-user disk attribution frozen (the find-based scan is heavy on memory and disk on the remote server; decommissioning is planned):
    • Remote agents no longer schedule or deploy the daily find scan: every agent config is written with scan_enabled=0; agents deployed with an older config stop scheduling after one Restart/Inject/config save.
    • The card "Scan now" entry, the per-user disk attribution popup, and the per-user disk curves in History are hidden; the 5-minute attribution auto-refresh in every window is off.
    • Record merges no longer pull/merge attribution records, and history queries no longer return attribution data.
    • Real-time disk capacity monitoring (DISK row, mount list, per-mount usage curves) is unaffected.
    • Existing local and remote attribution history is preserved untouched — not cleaned up, not deleted.
    • Implementation: code-level feature flags — Rust serverpulse_core::DISK_ATTRIBUTION_FROZEN plus frontend DISK_ATTRIBUTION_FROZEN; all code is retained and flipping the flags re-enables the feature; decommissioning will remove the code paths entirely.

Fixed

  • Fixed excessive memory usage in history queries (3–4 GB spikes at fast sampling): day files are now read through a streaming reader with a timestamp prefilter instead of loading each file in full and building a value tree per line; deduplication uses a structural record hash instead of a re-serialized copy of every record.
  • Local history now writes at most one line per server per UTC minute (live cards still refresh at the chosen sampling interval), shrinking day files roughly 40x and reducing disk-write pressure on both the local machine and the remote servers.
Assets 3
Loading