1brc解析 - 技术栈

一、问题解析

原文链接：

The One Billion Row Challenge - Gunnar Morlinghttps://www.morling.dev/blog/one-billion-row-challenge/一个文件中有10亿行数据，文件大小约为13G

文件内容如下：

复制代码

Hamburg;12.0
Bulawayo;8.9
Palembang;38.8
St. John's;15.2
Cracow;12.6

;前面是城市名，后面代表城市的测量值，值的范围是-99.9到99.9

要求输出每个城市的名称及测量值的最大值、最小值、平均值

类似这样

复制代码

Abha=5.0/18.0/27.4
Abidjan=15.7/26.0/34.1

二、数据准备

在机器上安装jdk21
git clone GitHub - gunnarmorling/1brc: 1️⃣🐝🏎️ The One Billion Row Challenge -- A fun exploration of how quickly 1B rows from a text file can be aggregated with Java1️⃣🐝🏎️ The One Billion Row Challenge -- A fun exploration of how quickly 1B rows from a text file can be aggregated with Java - gunnarmorling/1brchttps://github.com/gunnarmorling/1brc.git
./create_measurements.sh 1000000000

生成测试数据

三、代码实现

GitHub - fusugongzi/1brc

1. 文件读取

使用两个协程读取文件，先找到文件中间后的首行偏移量，一个协程读取文件开始到文件中间，另一个协程读取文件中间到文件结束。

复制代码

func readFile(offset, size, chunkSize int64, bytesChan chan []byte) {
	var readSpendTime int64 = 0
	defer func() {
		fmt.Println("read file spend time" + strconv.FormatInt(readSpendTime, 10))
	}()
	file, err := os.Open("measurements.txt")
	if err != nil {
		panic(err)
	}
	_, err = file.Seek(offset, 0)
	if err != nil {
		panic(err)
	}
	buf := make([]byte, chunkSize)
	leftover := make([]byte, 0, chunkSize)
	var readTotal int64 = 0
	for {
		readStart := time.Now().UnixMilli()
		singleRead, err := file.Read(buf)
		readSpendTime = readSpendTime + (time.Now().UnixMilli() - readStart)
		if err != nil {
			if errors.Is(err, io.EOF) {
				break
			}
			panic(err)
		}
		buf = buf[:singleRead]
		readTotal = readTotal + int64(singleRead)
		if readTotal > size {
			idx := int64(singleRead) - (readTotal - size)
			buf = buf[:idx]
		}

		toSend := make([]byte, singleRead)
		copy(toSend, buf)

		lastNewLineIndex := bytes.LastIndex(buf, []byte{'\n'})

		toSend = append(leftover, buf[:lastNewLineIndex+1]...)
		leftover = make([]byte, len(buf[lastNewLineIndex+1:]))
		copy(leftover, buf[lastNewLineIndex+1:])

		if readTotal >= size {
			toSend = append(toSend, leftover...)
			bytesChan <- toSend
			break
		} else {
			bytesChan <- toSend
		}
	}
}

2. 数据处理

使用unsafe实现了byte到string的转换，减少内存分配

将float转化为int存储，加快执行速度

复制代码

func process(readBytes []byte, dataChan chan map[string]*measurement) {
	m := make(map[string]*measurement)
	start := 0
	var city string
	var sign bool
	var processDom bool

	for idx, v := range readBytes {
		if v == byte(';') {
			city = unsafeString(unsafe.Pointer(&readBytes[start]), start, idx)
			start = idx + 1
		}
		if v == byte('\n') || idx == len(readBytes)-1 {
			if city != "" {
				sign = true
				processDom = false

				var measure int64 = 0
				for i := start; i < idx; i++ {
					if readBytes[i] == '-' {
						sign = false
					} else if readBytes[i] == '.' {
						processDom = true
					} else if !processDom {
						measure = measure*10 + int64(readBytes[i]-'0')*10
					} else {
						measure = measure + int64(readBytes[i]-'0')
					}
				}
				if !sign {
					measure = 0 - measure
				}

				if exist, ok := m[city]; !ok {
					m[city] = &measurement{
						min: measure,
						max: measure,
						sum: measure,
						cnt: 1,
					}
				} else {
					if measure < exist.min {
						exist.min = measure
					} else if measure > exist.max {
						exist.max = measure
					}
					exist.sum = exist.sum + measure
					exist.cnt++
				}
				city = ""
			}
			start = idx + 1
		}
	}

	dataChan <- m
}