Retry an insertion round with fewer workers when it is refused for memory
维护者通常 1 天内回复
还没有人认领这个 Issue。
评估
- 难度
- 4/5
- 预计耗时
- 1-2 天
- 新手友好度
- 48/100
- Issue 类型
- 功能
- 描述清晰度
- 基本清楚
- 活跃度
- 活跃
- 技术栈
- c
- 领域
- backend, performance
调研方向
Start with wirelog/columnar/eval_tdd_insert.c and the wl_columnar_eval_tdd_insert_stratum round, then read wirelog/columnar/session.c to follow how ENOSPC reaches col_session_snapshot_impl. Find the existing insert-route tests and round hook; the issue does not name their files. Done when budgeted snapshots at 4 and 8 workers retry, succeed, and match fresh evaluation, while unbudgeted round and task counts stay unchanged.
由索引模型根据 Issue 内容生成。
描述
Summary
Since #2122 a snapshot after an incremental insert runs a stratum's insertion rounds on up to num_workers worker clones (wl_columnar_eval_tdd_insert_stratum, wirelog/columnar/eval_tdd_insert.c). Each clone holds its own arena, delta pool, arrangements and output bags, and the leading inputs of split instances are copied into ranges. On the CSPA overlap bench row the process's peak RSS at 16 workers rose from about 400 MB to about 710 MB (433 to 488 MB at 8 workers).
Under an enforcing memory budget, a round that fitted on one clone can now be refused. The refusal is returned from the snapshot as ENOSPC. session_snapshot_insert's caller (col_session_snapshot_impl, wirelog/columnar/session.c) falls back to full evaluation only on ENOTSUP/EINVAL, so that snapshot fails and the session re-evaluates in full at the next snapshot.
Proposed change
When a parallel round fails with ENOSPC, discard its bags (the heads are untouched until the merge), and retry the same round with fewer tasks and no range splitting, down to one clone, before failing the snapshot.
Acceptance
- A test with an enforcing budget that admits the serial round but not the parallel one: at 4 and 8 workers the insert-route snapshot succeeds, matches a fresh evaluation, and the round hook reports the retry.
- Without a budget, round and task counts are unchanged.
- 主要语言
- C
- 星标
- 18
- 派生
- 2
- 平均合并
- 9 小时 48 分钟
- 30 天内合并 PR
- 165
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 没有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
semantic-reasoning/wirelog 的其他 Issue
-
难度 2/5 1-3 小时 新手友好度 83/100
semantic-reasoning/wirelog#2229 ·
维护者通常 1 天内回复
-
难度 1/5 1 小时以内 新手友好度 83/100
semantic-reasoning/wirelog#2133 ·
维护者通常 1 天内回复
-
难度 1/5 1 小时以内 新手友好度 88/100
semantic-reasoning/wirelog#2042 ·
维护者通常 1 天内回复
-
p2-medium
难度 2/5 1-3 小时 新手友好度 86/100
semantic-reasoning/wirelog#1914 ·
维护者通常 1 天内回复
-
enhancement track/semantics
难度 5/5 一周以上 新手友好度 12/100
semantic-reasoning/wirelog#2305 ·
维护者通常 1 天内回复
查看 semantic-reasoning/wirelog 的全部 Issue
相似的 Issue
-
难度 2/5 1-3 小时 新手友好度 78/100
libsdl-org/SDL#16607 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 76/100
PowerShell/Win32-OpenSSH#2482 · 1 个 reaction ·
-
难度 2/5 1-3 小时 新手友好度 66/100
维护者通常 2 天内回复
-
难度 2/5 1-3 小时 新手友好度 78/100
-
难度 2/5 1-3 小时 新手友好度 78/100
zephyrproject-rtos/zephyr#123881 ·
维护者通常 2 天内回复