Golang concurrency
This page is about the methods of concurrency provided by go.
If you're looking for synchronization primitives (ex. mutexes, semaphores, ... see golang synchronization)
Documentation
Chan
https://pkg.go.dev/go/types@go1.18.3#Chan
Not Present
Go does not:
- expose OS-threads, you only have access to it's green-threads
- abstract multiprocessing, but you could roll your own with a subprocess and IPC if you wanted to
- provide a message-queue implementation, use channels instead
Goroutines
Usage
Goroutines use green-threads rather than os-threads.
An OS thread is relatively expensive in setup and memory. One thread is reserved for a particular stack.
Go abstracts threads/threadpools with goroutines to make threads relatively cheap.func doThing() { fmt.Println("hi") } func main() { go sayHello() // <-- run in thread }Go functions default to using value objects rather than references.
Depending on your datastructure, this makes goroutines fairly concurrency-safe, since it operates on a copy of the data, rather than the same data.func printThing(a string) { fmt.Println(a) } go printThing("abc")Testing
go run -race foo.go # run, checking for race conditions
Goroutines cannot be stopped externally, you must use the poison pill pattern for eventloops.
If spinning up eventloops while testing, make sure to explicitly stop the goroutine after each test.Limits
Threads are a finite resource. You only have so many CPU cores, and CPU cores can only evaluate one thread at a time. Go defaults to allowing one thread per core, but you can generally get additional performance by increasing this.
require "runtime" runtime.GOMAXPROCS(-1) // show configured max-number of threads runtime.GOMAXPROCS(2) // set max-number of threadscpu core count
runtime.NumCPU()
Channels
Channels serve as a message queue for go's goroutines.
Channels are typed, and you may optionally restrict it to direction (ex. read/write only).Basics
Create channel
ch := make(chan int) // channel (sends/recvs ints, enqueues a max of 1 int at a time) ch := make(chan int, 100) // bufferred channel (sends/recvs ints, enqueues a max of 100 ints at a time)Read/write channel
num := <- ch // read next item from channel num, ok := <- ch // read next item from channel (ok false when channel is closed) ch <- 123 // append next item to channelChannel functions
ch := make(chan int) close(ch) // close it so no more items can be written/read len(ch) // number of unread elements enqueuedChannel direction in a method signature
go func(ch <-chan int) { ... } // read-only channel go func(ch chan<- int) { ... } // write-only channelDeadlock Protection
Go tries to protect you from deadlocks.
You can iterate over enqueued messages in the channel,
but if it is impossible for your application to enqueue any more items,
and you are still looping through enqueued messages, go will panic.In order to avoid this, you must close the channel when you know you are done sending messages.
require "sync" var wg = sync.WaitGroup{} func recv(ch <-chan int) { for i:= range ch { // <-- iterate over enqueued messages, as they become ready fmt.Println(i) } wg.Done() } func send(ch chan<- int) { for i := 0; i < 10; i++ { ch <- i } close(ch) // <-- indicate that no more items will be sent } func main() { ch := make(chan int, 50) wg.Add(1) go send(ch) go recv(ch) wg.Wait() }Select on Read
Go's implementation of
select
is fairly unique.
// blocking select statement for { select { case signal := <- signals_chan // ... case command := <- commands_chan // ... } } // optionally, if you pass in a 'default' section, // the select statement becomes non-blocking. // (your default code is run whenever no channels have data) for { select { case signal := <- signals_chan // ... case command := <- commands_chan // ... default: // ... }Channel Buffer on Writes
You can enqueue as much as you'd like to a buffered channel.
Your program will automatically pause until there is room for more items in the buffer.