The pattern, similar to RWMutex. With low latency between RLock and WLock, as well as calling a slow lazy asynchronous data loader. The code looks like a WLock call inside RLock. Please, readme golang/go#4026
go get github.com/jenchik/fallbackpackage main
import (
"github.com/jenchik/fallback"
"fmt"
"time"
)
var cacheDB map[int]string = make(map[int]string)
func getUserName(f fallback.Fallback, userId int) (name string, found bool) {
f.Do(func(h fallback.Helper) {
// sample code in shared lock
name, found = cacheDB[userId]
if found {
return
}
load := make(chan struct{})
h.Exclusive(func() {
// sample code in exclusive lock
cacheDB[userId] = name
}, func() {
// for example we load from DB or HTTP
time.Sleep(time.Millisecond * 100)
name = fmt.Sprintf("User ID: %d", userId)
close(load)
})
<-load
})
return
}
func main() {
start := time.Now()
f := fallback.NewBatchFallback(1)
for i := 0; i < 100; i++ {
go func(id int) {
time.Sleep(time.Millisecond)
getUserName(f, id)
}(i)
}
for i := 100; i > 0; i-- {
fmt.Println(getUserName(f, i))
}
fmt.Println("Duration:", time.Since(start).String())
}package main
import (
"github.com/jenchik/fallback"
"github.com/jenchik/listener"
// ...
)
type (
// T ...
)
// ...
// GetSampleResult with competitive access for each 'conditionKey' value will be called only once AsyncLoader and ExclusiveHandler
func GetSampleResult(f fallback.Fallback, listeners listener.Listeners, conditionKey interface{}) (result T, found bool) {
var l listener.Listener
f.Do(func(h fallback.Helper) {
// sample code in shared lock
if /* some condition */ {
// result = ...
return
}
if l, found = listeners.GetOrCreate(conditionKey); found {
return
}
h.Exclusive(func() {
// sample code in exclusive lock
// ... = l.Wait().(T)
listeners.Delete(conditionKey)
}, func() {
// for example may be async load from DB or HTTP
var val T
// val = ...
l.Broadcast(val)
})
})
if l != nil {
return l.Wait().(T)
}
return
}with lazyLoaderTimeout = 500ms
BenchmarkThreadsDoClassic-4 10000 100202 ns/op 0.02 MB/s 166 B/op 4 allocs/op
BenchmarkThreadsDoBatchSizeN1-4 10000 100485 ns/op 0.02 MB/s 161 B/op 4 allocs/op
BenchmarkThreadsDoBatchSizeN1000-4 10000 100200 ns/op 0.02 MB/s 155 B/op 3 allocs/op
BenchmarkThreadsConcurrentMap-4 10000 350299 ns/op 0.01 MB/s 18 B/op 0 allocs/op
with lazyLoaderTimeout = 100ms
BenchmarkThreadsDoClassic-4 10000000 173 ns/op 11.55 MB/s 32 B/op 1 allocs/op
BenchmarkThreadsDoBatchSizeN1-4 10000000 176 ns/op 11.30 MB/s 32 B/op 1 allocs/op
BenchmarkThreadsDoBatchSizeN1000-4 10000000 154 ns/op 12.95 MB/s 32 B/op 1 allocs/op
BenchmarkThreadsConcurrentMap-4 10000000 108 ns/op 18.51 MB/s 0 B/op 0 allocs/op
with lazyLoaderTimeout = 50ms
BenchmarkThreadsDoClassic-4 10000000 179 ns/op 11.14 MB/s 32 B/op 1 allocs/op
BenchmarkThreadsDoBatchSizeN1-4 10000000 154 ns/op 12.94 MB/s 32 B/op 1 allocs/op
BenchmarkThreadsDoBatchSizeN1000-4 10000000 168 ns/op 11.88 MB/s 32 B/op 1 allocs/op
BenchmarkThreadsConcurrentMap-4 50000000 24.6 ns/op 81.30 MB/s 0 B/op 0 allocs/op
with lazyLoaderTimeout = 10ms
BenchmarkThreadsDoClassic-4 10000000 153 ns/op 13.04 MB/s 32 B/op 1 allocs/op
BenchmarkThreadsDoBatchSizeN1-4 10000000 159 ns/op 12.56 MB/s 32 B/op 1 allocs/op
BenchmarkThreadsDoBatchSizeN1000-4 10000000 155 ns/op 12.86 MB/s 32 B/op 1 allocs/op
BenchmarkThreadsConcurrentMap-4 100000000 20.0 ns/op 100.21 MB/s 0 B/op 0 allocs/op
with lazyLoaderTimeout = 1ms
BenchmarkThreadsClassicWithMutex-4 20000 54138 ns/op 0.04 MB/s 39 B/op 0 allocs/op
BenchmarkThreadsDoClassic-4 10000000 188 ns/op 10.59 MB/s 36 B/op 1 allocs/op
BenchmarkThreadsDoBatchSizeN1-4 10000000 147 ns/op 13.55 MB/s 32 B/op 1 allocs/op
BenchmarkThreadsDoBatchSizeN1000-4 10000000 141 ns/op 14.17 MB/s 32 B/op 1 allocs/op
BenchmarkThreadsConcurrentMap-4 100000000 18.3 ns/op 109.36 MB/s 0 B/op 0 allocs/op
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent