WIP: perf: add stream download/upload mem benchmark

.gitignore

@@ -123,3 +123,6 @@ run
 !test/ctx_register.js
 
 .egg/
+
+# Benchmark test files
+benchmark/stream_download/nginx/50mb_ones.txt

benchmark/stream_download/.gitignore

@@ -0,0 +1,4 @@
+tmp
+coredumps
+*.heapsnapshot
+core.*

benchmark/stream_download/COREDUMP_ANALYSIS.md

@@ -0,0 +1,188 @@
+# Node.js Coredump Analysis Guide
+
+This document describes how to analyze Node.js coredump files for memory leak detection.
+
+## Prerequisites
+
+- Docker container with `gdb` and `procps` installed
+- Core dump file generated with `ulimit -c unlimited`
+- Node.js built with debug symbols (optional but helpful)
+
+## Generating Coredump
+
+```bash
+# Run benchmark and generate coredump
+./run-benchmark-with-coredump.sh 60
+
+# Or manually:
+# 1. Start benchmark
+docker exec -d nginx-benchmark-server bash -c "cd /root/workspace && node --expose-gc --heapsnapshot-signal=SIGUSR2 benchmark.js"
+
+# 2. Get PID
+docker exec nginx-benchmark-server cat /tmp/benchmark.pid
+
+# 3. Generate heap snapshot (optional)
+docker exec nginx-benchmark-server kill -SIGUSR2 <PID>
+
+# 4. Generate coredump
+docker exec nginx-benchmark-server kill -SIGABRT <PID>
+
+# 5. Copy coredump
+./copy-coredump.sh
+```
+
+## Analysis Methods
+
+### Method 1: Benchmark Log Analysis
+
+Extract memory stats from benchmark log:
+
+```bash
+# Get memory trend
+grep -E "(rss:|heapUsed:|external:|arrayBuffers:)" benchmark.log | paste - - - - | awk '{
+  gsub(/,/,"",$0);
+  printf "RSS=%3.0fMB, heapUsed=%2.0fMB, external=%2.0fMB, arrayBuffers=%2.0fMB\n",
+    $2/1024/1024, $4/1024/1024, $6/1024/1024, $8/1024/1024;
+}'
+
+# Calculate statistics
+grep "rss:" benchmark.log | awk '{gsub(/,/,"",$2); print $2}' | sort -n | tail -5
+```
+
+### Method 2: String Extraction from Coredump
+
+```bash
+# Find error patterns
+strings core.58 | grep -E "(Error|ENOMEM|EMFILE|leak)" | head -50
+
+# Find memory stats captured in core
+strings core.58 | grep -E "(heapTotal|heapUsed|external|arrayBuffers|rss):" | tail -20
+
+# Count object references (e.g., temp files by UUID)
+strings core.58 | grep -oE "[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}" | wc -l
+
+# Find connection/socket patterns
+strings core.58 | grep -E "(Socket|Stream|Pool|Agent|keepAlive)" | sort | uniq -c | sort -rn | head -20
+
+# Find error codes
+strings core.58 | grep -E "(ECONNRESET|ETIMEDOUT|ENOTFOUND|ECONNREFUSED|EPIPE)" | head -20
+
+# Find loaded modules
+strings core.58 | grep -E "node_modules/.+\.js" | sort | uniq -c | sort -rn | head -30
+```
+
+### Method 3: LLDB Analysis (macOS)
+
+```bash
+# Load coredump
+lldb -c core.58
+
+# Commands in lldb:
+(lldb) bt all                    # Backtrace all threads
+(lldb) thread list               # List all threads
+(lldb) memory region --all       # Show memory regions
+(lldb) process status            # Process state
+```
+
+### Method 4: GDB Analysis (Linux/Docker)
+
+```bash
+# Copy coredump to container
+docker cp core.58 nginx-benchmark-server:/tmp/core.58
+
+# Analyze with GDB
+docker exec -it nginx-benchmark-server gdb /usr/local/bin/node /tmp/core.58
+
+# GDB commands:
+(gdb) bt                         # Backtrace
+(gdb) info threads               # List threads
+(gdb) thread apply all bt        # Backtrace all threads
+(gdb) info registers             # Register state
+```
+
+### Method 5: llnode (Node.js LLDB Plugin)
+
+```bash
+# Install llnode
+npm install -g llnode
+
+# Analyze V8 heap
+lldb -c core.58
+(lldb) plugin load /path/to/llnode.dylib
+(lldb) v8 bt                     # V8-aware backtrace
+(lldb) v8 findjsobjects          # Find JS objects by type
+(lldb) v8 findjsinstances Array  # Find Array instances
+```
+
+## Memory Metrics Reference
+
+| Metric         | Description                              | Normal Range                 |
+| -------------- | ---------------------------------------- | ---------------------------- |
+| `rss`          | Resident Set Size (total process memory) | Varies, should stabilize     |
+| `heapTotal`    | V8 heap allocated                        | Grows then stabilizes        |
+| `heapUsed`     | V8 heap actually used                    | Should not continuously grow |
+| `external`     | Memory for C++ objects bound to JS       | Fluctuates with I/O          |
+| `arrayBuffers` | Memory for ArrayBuffer/TypedArray        | Fluctuates with I/O          |
+
+## Memory Leak Indicators
+
+### Leak Detected:
+
+- `heapUsed` continuously growing without returning to baseline
+- `rss` continuously growing over time
+- ENOMEM errors in strings output
+- EMFILE (too many open files) errors
+- Thousands of duplicate object references
+
+### No Leak (Healthy):
+
+- `heapUsed` fluctuates but returns to baseline
+- `rss` stabilizes after initial growth
+- `external` and `arrayBuffers` fluctuate with I/O operations
+- GC running regularly (check GC stats in log)
+
+## Example Analysis Report
+
+```
+=== Memory Analysis Report ===
+
+Sample count: 147
+Duration: 60 seconds
+Operations: 9200 download/upload cycles
+
+Memory State:
+- Initial RSS: 235 MB
+- Final RSS: 328 MB
+- Max RSS: 360 MB
+- Growth: 93 MB (40%) - NORMAL (initial allocation)
+
+V8 Heap (heapUsed): 12-20 MB - STABLE (no leak)
+External Memory: 5-85 MB - FLUCTUATING (normal for I/O)
+ArrayBuffers: 0-74 MB - FLUCTUATING (normal for file ops)
+
+Conclusion: NO MEMORY LEAK DETECTED
+```
+
+## Troubleshooting
+
+### Coredump not generated
+
+```bash
+# Check ulimit
+docker exec container ulimit -c
+
+# Set unlimited
+docker run --ulimit core=-1 --privileged ...
+```
+
+### Architecture mismatch (Rosetta)
+
+If running on Apple Silicon with x86_64 container:
+
+- Use `strings` extraction method
+- Or run native ARM64 container: `--platform linux/arm64`
+
+### Missing symbols in GDB/LLDB
+
+- Use Node.js debug build
+- Or rely on string extraction methods

benchmark/stream_download/Dockerfile

@@ -0,0 +1,50 @@
+FROM node:24.12.0
+
+# 安装 nginx 和其他必要工具
+RUN apt-get update && apt-get install -y \
+    nginx \
+    curl \
+    vim \
+    gdb \
+    procps \
+    && rm -rf /var/lib/apt/lists/* \
+    && apt-get clean
+
+# 配置 coredump
+RUN echo "ulimit -c unlimited" >> /etc/bash.bashrc \
+    && mkdir -p /tmp/cores \
+    && chmod 777 /tmp/cores
+
+# 创建 nginx 配置目录
+RUN mkdir -p /etc/nginx/conf.d
+
+# 复制 nginx 配置文件
+COPY nginx.conf /etc/nginx/sites-available/default
+
+# 创建 nginx 工作目录
+RUN mkdir -p /var/www/html
+
+# 创建启动脚本
+COPY start-nginx.sh /usr/local/bin/start-nginx.sh
+RUN chmod +x /usr/local/bin/start-nginx.sh
+
+# 暴露端口
+EXPOSE 80 9229
+
+# 设置工作目录
+WORKDIR /var/www/html
+
+# 健康检查
+HEALTHCHECK --interval=30s --timeout=3s --start-period=5s --retries=3 \
+    CMD curl -f http://localhost/health || exit 1
+
+RUN mkdir -p /root/workspace
+
+COPY gc.js /root/workspace/gc.js
+COPY benchmark.js /root/workspace/benchmark.js
+COPY benchmark_undici.js /root/workspace/benchmark_undici.js
+
+RUN cd /root/workspace && npm i urllib --registry https://registry.npmmirror.com
+
+# 启动命令
+CMD ["/usr/local/bin/start-nginx.sh"]

benchmark/stream_download/README.md

@@ -0,0 +1,115 @@
+# Nginx 下载/上传测试服务器
+
+## 快速开始
+
+> **注意**: 请先切换到 `benchmark/stream_download` 目录下执行以下命令
+
+### 构建镜像
+
+```bash
+docker build --platform linux/amd64 -t nginx-node-benchmark .
+```
+
+### 运行容器
+
+```bash
+docker run --rm -d --platform linux/amd64 \
+  --name nginx-node-benchmark \
+  -p 8080:80 \
+  -v $(pwd)/nginx:/var/www/html \
+  nginx-node-benchmark
+```
+
+### 测试
+
+```bash
+# 下载测试
+curl -v http://localhost:8080/download/test-file.txt
+
+# 上传测试
+curl -v -X POST -d "test" http://localhost:8080/upload/
+```
+
+### 停止
+
+```bash
+docker stop nginx-node-benchmark && docker rm nginx-node-benchmark
+```
+
+### 运行生成大文件
+
+```bash
+sh generate_50mb_file.sh
+```
+
+### 运行 node 测试
+
+```bash
+docker exec -ti nginx-node-benchmark bash
+
+cd /root/workspace
+node benchmark.js
+```
+
+## 内存分析 (Memory Leak Analysis)
+
+### 运行 benchmark 并生成 coredump
+
+运行 benchmark 60 秒后自动生成 heap snapshot 和 coredump:
+
+```bash
+# 使用默认 60 秒
+./run-benchmark-with-coredump.sh
+
+# 或指定运行时间（秒）
+./run-benchmark-with-coredump.sh 120
+```
+
+### 手动复制 coredump 文件
+
+如果需要手动复制 coredump 和 heap snapshot 文件:
+
+```bash
+./copy-coredump.sh
+```
+
+### 分析 heap snapshot
+
+1. 打开 Chrome DevTools -> Memory 标签
+2. 点击 "Load" 加载 `coredumps/*.heapsnapshot` 文件
+3. 分析内存分配情况
+
+### 分析 coredump
+
+详细的 coredump 分析指南请参考 [COREDUMP_ANALYSIS.md](./COREDUMP_ANALYSIS.md)
+
+快速分析方法:
+
+```bash
+# 使用 strings 提取内存信息
+strings coredumps/core.* | grep -E "(heapUsed|rss|external):" | tail -20
+
+# 查找内存相关错误
+strings coredumps/core.* | grep -E "(ENOMEM|EMFILE|leak)" | head -20
+
+# 统计对象引用数量
+strings coredumps/core.* | grep -oE "[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}" | wc -l
+
+# 在容器内使用 gdb 分析
+docker exec -it nginx-benchmark-server gdb /usr/local/bin/node /tmp/core.*
+
+# 或将 coredump 复制到本地后使用 lldb 分析
+lldb -c coredumps/core.*
+```
+
+### 手动触发 heap snapshot
+
+在 benchmark 运行时发送 SIGUSR2 信号生成 heap snapshot:
+
+```bash
+# 获取 benchmark 进程 PID
+docker exec nginx-benchmark-server cat /tmp/benchmark.pid
+
+# 发送信号生成 heap snapshot
+docker exec nginx-benchmark-server kill -SIGUSR2 <PID>
+```