Skip to main content

inspect-a-database-query-plan

Inspect a database query plan through operators, estimates, actual work, memory, I/O, and locking evidence. Use when a query is slow, unstable, or unexpectedly expensive.

来源信息

仓库
protosphinx/sphinxstack
最近来源活动
2026年7月29日 13:55
检测到的 SKILL.md 语言
英语
星标
3
Forks
0

安装方式

默认使用会先检查来源的 Prompt;你也可以切换为直接命令,或下载本地副本。

检查来源文件

决定是否安装前,请先阅读 SKILL.md,以及 SkillsMP 当前展示的配套文件。

文件资源管理器
2 个文件

正在显示 SKILL.md

SKILL.md
来源说明 · 只读预览
name
inspect-a-database-query-plan
category
data
description
Inspect a database query plan through operators, estimates, actual work, memory, I/O, and locking evidence. Use when a query is slow, unstable, or unexpectedly expensive.
# inspect-a-database-query-plan Explain the work the database performed, then test the smallest evidence-based change. ## Procedure 1. Capture the exact normalized query, parameters, database version, schema, and representative data scale. 2. Obtain a safe estimated or actual plan without exposing sensitive literals. 3. Read the operator tree in execution order and identify scans, joins, sorts, aggregates, and exchanges. 4. Compare estimated and actual rows, loops, time, buffers, spills, and memory. 5. Check predicates, statistics, indexes, data skew, parameter sensitivity, locks, and concurrency. 6. Locate the first large estimation or work amplification, not only the most expensive displayed node. 7. Test one candidate change against representative and adversarial parameter sets. 8. Record before-and-after plans, runtime, resource use, and regression risk. ## Guardrails - Actual-plan collection can execute the query and may be unsafe for destructive statements. - Do not add an index solely because a plan mentions one. - A faster isolated query can still harm write cost or concurrent workloads. ## Done - A redacted query-plan analysis report is saved - Estimates, actual work, and environmental assumptions are verified - Any proposed change is tested against representative parameters and concurrency
在 GitHub 查看