运行Go异步操作并写入map

时间:2017-07-28 16:40:04

标签: python asynchronous go google-cloud-platform goroutine

我有this project试图在Go中同时运行无限大量查询。父项目都是Python。我需要能够跟踪查询结果,就像在地图中一样。

Input:

{
 'reports_portal': 'select * from reports_portal',
 'billing_portal': 'select * from billing_portal',
}

output:

{
 'reports_portal': [23, 123, 5234, 632],
 'billing_portal': [23, 123, 5234, 632],
}

等等

这些大型查询需要异步运行,因为它们非常慢(从UI角度来看,SRE等待15-30秒才能获得结果。

我首先尝试将项目异步写入地图:

package main

import (
    "fmt"
)


func add_to_map(m map[string] string, word string) map[string]string {
    added_word := word + " plus more letters"
    m[word] = added_word
    return m
}


func main() {
    words_map := make(map[string]string)
    words := []string{"giraffe", "cat", "dog", "turtle"}
    for _, this_word := range words {
        go add_to_map(words_map, this_word)
    }
    fmt.Println(words_map)
}

爆炸如下:

$ go run try_asynchronous.go 
fatal error: concurrent map writes

goroutine 7 [running]:
runtime.throw(0x10b3b96, 0x15)
    /usr/local/Cellar/go/1.8.1/libexec/src/runtime/panic.go:596 +0x95 fp=0xc420032eb8 sp=0xc420032e98
runtime.mapassign(0x109ad20, 0xc420016270, 0xc420032fa0, 0x10b3268)
    /usr/local/Cellar/go/1.8.1/libexec/src/runtime/hashmap.go:499 +0x667 fp=0xc420032f58 sp=0xc420032eb8
main.add_to_map(0xc420016270, 0x10b1ba0, 0x3, 0x0)
    /tmp/golang-w-python/try_asynchronous.go:10 +0xa3 fp=0xc420032fc0 sp=0xc420032f58
runtime.goexit()
    /usr/local/Cellar/go/1.8.1/libexec/src/runtime/asm_amd64.s:2197 +0x1 fp=0xc420032fc8 sp=0xc420032fc0
created by main.main
    /tmp/golang-w-python/try_asynchronous.go:19 +0xc8

goroutine 1 [runnable]:
fmt.(*pp).fmtString(0xc42001e0c0, 0x10b1f52, 0x7, 0x76)
    /usr/local/Cellar/go/1.8.1/libexec/src/fmt/print.go:424 +0x1a2
fmt.(*pp).printValue(0xc42001e0c0, 0x10953c0, 0xc42000e260, 0x98, 0x76, 0x1)
    /usr/local/Cellar/go/1.8.1/libexec/src/fmt/print.go:729 +0x27aa
fmt.(*pp).printValue(0xc42001e0c0, 0x109ad20, 0xc420016270, 0x15, 0x76, 0x0)
    /usr/local/Cellar/go/1.8.1/libexec/src/fmt/print.go:750 +0x103d
fmt.(*pp).printArg(0xc42001e0c0, 0x109ad20, 0xc420016270, 0x76)
    /usr/local/Cellar/go/1.8.1/libexec/src/fmt/print.go:682 +0x217
fmt.(*pp).doPrintln(0xc42001e0c0, 0xc420045f28, 0x1, 0x1)
    /usr/local/Cellar/go/1.8.1/libexec/src/fmt/print.go:1138 +0xa1
fmt.Fprintln(0x1108140, 0xc42000c018, 0xc420045f28, 0x1, 0x1, 0xc420045ef0, 0xc420045ee0, 0x1087218)
    /usr/local/Cellar/go/1.8.1/libexec/src/fmt/print.go:247 +0x5c
fmt.Println(0xc420045f28, 0x1, 0x1, 0x10b1e6f, 0x6, 0x0)
    /usr/local/Cellar/go/1.8.1/libexec/src/fmt/print.go:257 +0x57
main.main()
    /tmp/golang-w-python/try_asynchronous.go:21 +0x132
exit status 2

基于需要一次运行多个查询并尝试按名称跟踪结果,我希望在异步期间写入映射。但是fatal error: concurrent map writes说你不能。

我不明白

  1. 为什么不
  2. 我应该做什么同时运行这些大型查询。
  3. 编辑:

    我最接近返回结果的东西不是异步的:

    package main
    
    import (
        "fmt"
        "math/rand"
        "sync"
        "time"
    )
    
    var mutex sync.Mutex
    var wg sync.WaitGroup
    
    func random_sleep() {
        r := rand.Intn(3000)
        time.Sleep(time.Duration(r) * time.Millisecond)
    }
    
    func add_to_map(m map[string] string, word string) {
        defer wg.Done()
        added_word := word + " plus more letters"
        mutex.Lock()
        defer mutex.Unlock()
        fmt.Println("Before sleep")
        random_sleep()
        m[word] = added_word
        fmt.Println("Added word %v", word)
    }
    
    
    func main() {
        words_map := make(map[string]string)
        words := []string{"giraffe", "cat", "dog", "turtle"}
        for _, this_word := range words {
            wg.Add(1)
            go add_to_map(words_map, this_word)
        }
        wg.Wait()
        fmt.Println(words_map)
    }
    

    结果是错误的:

    cchilders:~/work_projects/metricsportal/golang_integration (feature/golang-query) 
    $ go run try_async.go 
    Before sleep
    Added word %v turtle
    Before sleep
    Added word %v cat
    Before sleep
    Added word %v giraffe
    Before sleep
    Added word %v dog
    map[dog:dog plus more letters turtle:turtle plus more letters cat:cat plus more letters giraffe:giraffe plus more letters]
    
    cchilders:~/work_projects/metricsportal/golang_integration (feature/golang-query) 
    $ go run try_async.go 
    Before sleep
    Added word %v turtle
    Before sleep
    Added word %v cat
    Before sleep
    Added word %v giraffe
    Before sleep
    Added word %v dog
    map[dog:dog plus more letters turtle:turtle plus more letters cat:cat plus more letters giraffe:giraffe plus more letters]
    

    结果应该非常快,不超过3秒(我认为随机的最大值):

    Expectation - 
    
    Before sleep
    Before sleep
    Before sleep
    Before sleep
    Added word %v cat
    Added word %v giraffe
    Added word %v turtle
    Added word %v dog
    

3 个答案:

答案 0 :(得分:1)

您的代码中有两个不同的问题:

1)即使您总是写不同的密钥,也不能在不锁定地图的情况下同时执行此操作:https://golang.org/doc/faq#atomic_maps

因此,您需要确保在访问地图时获得对地图的独占访问权。

2)你需要在打印地图之前完成所有goroutine的完成(这就是为什么你在编辑的代码中得到不一致的结果)

根据您的示例解决这两个问题的简单方法:

package main

import (
    "fmt"
    "sync"
)

var mutex sync.Mutex
var wg sync.WaitGroup

func add_to_map(m map[string] string, word string) {
    defer wg.Done()
    added_word := word + " plus more letters"
    mutex.Lock()
    defer mutex.Unlock()
    m[word] = added_word
}


func main() {
    words_map := make(map[string]string)
    words := []string{"giraffe", "cat", "dog", "turtle"}
    for _, this_word := range words {
        wg.Add(1)
        go add_to_map(words_map, this_word)
    }
    wg.Wait()
    fmt.Println(words_map)
}

答案 1 :(得分:1)

好的,让我澄清一些事情并帮助你。

你不需要从这里返回修改后的地图,因为你的 function获取对map的引用而不是它的副本。 (让我们忽略这个事实 你完全无视返回值)

func add_to_map(m map[string] string, word string) map[string]string {
    added_word := word + " plus more letters"
    m[word] = added_word
    return m
}

接下来就是你需要同步访问地图。您可以使用 mutex为此。

import "sync"

var mutex sync.Mutex //glabal variable but can be created as local also

func add_to_map(m map[string] string, word string) {
    added_word := word + " plus more letters"
    // here you can do long to compute task and calculate result
    // calc here
    mutex.Lock() //result ready lock mutex
    defer mutex.Unlock() // unlock mutex when we return from function
    m[word] = added_word // result write to shared map
}

请注意,在Go 1.9中会有Concurrent Map类型。

修改: 你需要等待所有的例程完成,因为你的main()现在已经完成了。您可以使用WaitGroup

执行此操作
package main

import (
    "fmt"
    "sync"
)

var mutex sync.Mutex
var wg sync.WaitGroup

func add_to_map(m map[string] string, word string) {
    defer wg.Done()
    added_word := word + " plus more letters"
    // do heavy work here
    //
    mutex.Lock()
    defer mutex.Unlock()
    m[word] = added_word
}


func main() {
    words_map := make(map[string]string)
    words := []string{"giraffe", "cat", "dog", "turtle"}
    for _, this_word := range words {
        wg.Add(1)
        go add_to_map(words_map, this_word)
    }
    wg.Wait()
    fmt.Println(words_map)
}

答案 2 :(得分:0)

(代表OP发布解决方案)

我对假延迟的使用是错误的,解决方案都有效。谢谢你:

package main

import (
    "fmt"
    "math/rand"
    "sync"
    "time"
)

var mutex sync.Mutex
var wg sync.WaitGroup

func random_sleep() {
    r := rand.Intn(3000)
    time.Sleep(time.Duration(r) * time.Millisecond)
}


func add_to_map(m map[string] string, word string) {
    defer wg.Done()
    added_word := word + " plus more letters"
    fmt.Println("Before sleep")
    random_sleep()
    mutex.Lock()
    defer mutex.Unlock()
    m[word] = added_word
    fmt.Println("Added word %v", word)
}


func main() {
    words_map := make(map[string]string)
    words := []string{"giraffe", "cat", "dog", "turtle"}
    for _, this_word := range words {
        wg.Add(1)
        go add_to_map(words_map, this_word)
    }
    wg.Wait()
    fmt.Println(words_map)
}

结果:

$ go run try_async.go 
Before sleep
Before sleep
Before sleep
Before sleep
Added word %v dog
Added word %v giraffe
Added word %v cat
Added word %v turtle
map[turtle:turtle plus more letters dog:dog plus more letters giraffe:giraffe plus more letters cat:cat plus more letters]