Hacktoberfest 2026：维护者为十月标记出来的 issue，仍然开放、适合新手。 浏览 Hacktoberfest issue

maple start --in-memory: generate chDB settings that suit a throwaway store

未关闭 适合新手
  • 哪里不对？
#1,888 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
2/5
预计耗时
1-3 小时
新手友好度
74/100
Issue 类型
功能
描述清晰度
描述清楚
活跃度
活跃
技术栈
clickhouse, typescript
领域
cli

调研方向

从 apps/cli/src/commands/server.ts#L1535-L1538 开始，这里的内存模式分支跳过了配置生成；参考 L689-L705 的 resolveChdbConfigFile 作为要效仿的模型。添加一条内存模式路径，在存储旁边写入生成的配置文件，其中包含表格中的四个 <merge_tree> 值，并保持用户提供的 --chdb-config-file 不被修改。完成标准是 maple start --in-memory 能生成该文件，并且 system.merge_tree_settings 中的设置显示为已更改；运行 CLI server 命令的现有测试，确认 data-dir 模式不受影响。

由索引模型根据 Issue 内容生成。

描述

Written by an AI agent (Claude) on behalf of @jasonkuhrt.

maple start --in-memory keeps one run's telemetry and deletes it on exit, but its chDB runs with chDB's defaults, which suit a long-lived store. Each OTLP export is a synchronous INSERT per datasource that fans out through the materialized views into 16 to 20 new parts, so even a small dev workload writes about 315 files per insert. To get settings that suit a throwaway store, each caller has to write a chDB config and pass it in:

maple start --in-memory --offline --chdb-config-file ./throwaway-store.chdb.xml

<!-- throwaway-store.chdb.xml -->
<clickhouse>
  <merge_tree>
    <min_bytes_for_wide_part>10737418240</min_bytes_for_wide_part>
    <min_parts_to_merge_at_once>20</min_parts_to_merge_at_once>
    <cleanup_delay_period>1</cleanup_delay_period>
    <max_cleanup_delay_period>5</max_cleanup_delay_period>
  </merge_tree>
</clickhouse>

Cause

Data-dir mode generates a chDB config (resolveChdbConfigFile, for checkpoints), while in-memory mode passes on only a user-supplied file (apps/cli/src/commands/server.ts#L1535-L1538):

const chdbConfigFile =
	inMemory === undefined
		? yield* resolveChdbConfigFile(dataDir, Option.getOrUndefined(a.chdbConfigFile))
		: Option.getOrUndefined(a.chdbConfigFile)

Proposal

When --in-memory is set and no --chdb-config-file is given, generate a config beside the in-memory store holding the four <merge_tree> settings above, the way resolveChdbConfigFile generates one for data-dir mode. A user-supplied file stays untouched.

Setting Value What it does for a throwaway store
min_bytes_for_wide_part 10 GiB (chDB default 10 MiB) Parts stay compact, one data file for all columns. A merged wide traces part is about 200 files and is rewritten on nearly every merge; a new compact one is 29.
min_parts_to_merge_at_once 20 A merge waits for 20 parts instead of rewriting on every insert.
cleanup_delay_period / max_cleanup_delay_period 1 s / 5 s (chDB default 30 s / 300 s) Merged-away parts were over 90% of what sat on disk; they now go within seconds.

Costs: up to about 20 active parts per partition per table, and compact parts read single columns more slowly. Neither shows at the size of one run's telemetry.

Measurements

A harness ran Maple v0.3.0's schema (local schema v41) on libchdb 26.7 directly, 15k rows preloaded, 5 rows inserted every 2 s, 150 s per variant, on an M-series Mac under heavy unrelated load. File counts were stable between repeats; CPU was noisy.

Variant Parts per insert Files per insert CPU-s per row
chDB defaults 20.0 314 0.08
chDB defaults, repeat 20.1 316 0.11
compact parts 20.1 284 0.07
min_parts_to_merge_at_once 20 17.5 251 0.06
compact + min_parts_to_merge_at_once 20 17.6 246 0.07
same + cleanup delay 1 s 17.4 244 0.11
  • Files per insert: 315 → 244, about 22% fewer.
  • Files standing at the end of the run: 11,096 → 2,492 with the 1 s cleanup delay.
  • The harness under-counts merges (0.19 merges per insert on traces, against about 0.9 on a live store), so the compact-parts gain on a live store is likely larger.
  • maple start --in-memory --chdb-config-file <file> on v0.2.0 shows all four overrides as changed in system.merge_tree_settings.

We run Maple this way, with the file above, for every dev stack and every test.

Related: #1852 (per-export cost), #1764 (added --in-memory).

主要语言
TypeScript
星标
1.8k
派生
120
平均合并
8 小时 11 分钟
30 天内合并 PR
667

环境准备

这个项目没有提供开发容器、Dockerfile 或贡献指南，环境需要你自己搭建：先看它的 README，通用步骤见我们的新手贡献指南。

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

MapleTechLabs/maple 的其他 Issue

查看 MapleTechLabs/maple 的全部 Issue

相似的 Issue

更多 TypeScript Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。