Agent Skills › timeplus-io/proton › alloc-profile

alloc-profile

GitHub

分析 jemalloc 或 async-profiler 的内存分配配置文件，定位分配热点与代码归属。

.claude/skills/alloc-profile/SKILL.md timeplus-io/proton

Trigger Scenarios

用户提到内存分析、堆分析、分配热点涉及 jemalloc、async-profiler 或折叠栈格式文件

Install

npx skills add timeplus-io/proton --skill alloc-profile -g -y

More Options

Non-standard path

npx skills add https://github.com/timeplus-io/proton/tree/develop/.claude/skills/alloc-profile -g -y

Use without installing

npx skills use timeplus-io/proton@alloc-profile

指定 Agent (Claude Code)

npx skills add timeplus-io/proton --skill alloc-profile -a claude-code -g -y

安装 repo 全部 skill

npx skills add timeplus-io/proton --all -g -y

预览 repo 内 skill

npx skills add timeplus-io/proton --list

SKILL.md

Frontmatter

{
    "name": "alloc-profile",
    "description": "Analyze jemalloc or async-profiler allocation profiles in collapsed stack format. Make sure to use this skill whenever the user mentions memory profiling, heap analysis, allocation hotspots, jemalloc, async-profiler, collapsed stacks, .folded files, or asks about memory usage patterns."
}

Allocation Profile

Inputs

$0 (optional): path to a .collapsed or .folded profile

If no path is given, search near the current directory:

find . -maxdepth 4 \( -name "*.collapsed" -o -name "*.folded" \) | sort

Tooling

Use the bundled analyzer script:

python3 .claude/skills/alloc-profile/scripts/analyze_alloc_profile.py <profile>

What the analyzer reports

Total samples/bytes and unique stack traces
Top stack traces
Top outermost meaningful frames
Top leaf allocation frames

This gives both:

"why did we allocate?" via outermost frames
"what code actually allocated?" via leaf frames

Interpretation guidance

Large outermost buckets often point to the higher-level operation to optimize first.
Large leaf buckets inside allocators or containers are usually symptoms; look one or two frames above for product code.
For streaming paths, pay special attention to joins, windows, aggregation state, and block materialization.
For historical paths, focus on scans, merges, deserialization, and sort/aggregation buffers.

Related repo context

Jemalloc profiling notes live in tests/instructions/jemalloc_memory_profile.txt.
Generic heap-profiler notes live in tests/instructions/heap-profiler.txt.

Output expectations

Summarize:

biggest operations by memory share
likely owning subsystem
likely next inspection target in code
whether the profile looks like steady-state memory, burst allocation, or leak-like retention