Inspirations

The idea behind this project was to get accustomed to Go programming language and its concurrency model by implementing a thread pool. Worth noting that there are multiple ways you can write a thread pool, and the way I did it, probably not how an experienced Golang programmer would do. My current solution is based around thread-safe queues, but that could be easily replaced with channels, the core mechanism for passing data between go routines and synchronizing them.

NOTE This project was written exclusively for learning purposes and should never be used in a production.

Overall description

As mentioned above, a decision was made to use thread-safe queues for tasks submission and processing, though use of channels will be more natural. The core data type looks like this:

type ThreadPool struct {
	maxThreads uint32

	submitQueue  *Queue[ThreadFunc]
	waitingQueue *Queue[ThreadFunc]
	workQueue    *Queue[ThreadFunc]

	wg          sync.WaitGroup
	doneCh      chan struct{}
	threadCount uint32

	metrics Metrics

	waiting int32

	blocked bool

	logsEnabled bool
	*Logger
}

Where maxThreads is the maximum number of goroutines running concurrently. waitingQueue is used when all the workers (goroutines) are busy and no new can be spawned, a task is put into a waiting queue. submitQueue is responsible for tasks submission. workQueue a queue to pull work from. The rest of the data are internals and easily understandable by looking at code.

All the logic is happening inside processTasks() function, which is itself is executed in a separate go routine. This was mainly done to add a possibility to process tasks on the background while some more still could be submitted.

The flow is pretty straightforward, we pull tasks from a submitQueue and enqueue them into workQueue. Spawned workers constantly polling a work queue for available tasks and execute them. If the amount of workers is equal to maxThreads all the subsequent tasks are pushed into a waitQueue instead. No new workers are spawned until a wait queue is empty.

Example:

const nThreads = 8
p := NewPool(nThreads)
for i := 0; i < (1 << 10); i++ {
	p.SubmitTask(func(){
		// do something
	})	
}
p.Wait()

IMPORTANT Each call to NewPool(...) should be supplemented with Wait() after all the tasks have been submitted.

Example

A simple web-crawler was implemented to demonstrate the functionality of a thread pool in action. An example could be found in example.go file. The programm traverses a specified url up to a certain depth (supplied from the command line) in a breadth-first search fashion, and outputs all href(s) to stdout.

In orde to achieve that goal I had to implement a simple, generic stack with Push/TryPop/Empty and Size methods. Here is the stack data type and its core function declarations:

type Stack[T any] struct {
	count int
	data  []T
}

func (s *Stack[T]) Push(v T)
func (s *Stack[T]) TryPop(v *T) bool
func (s *Stack[T]) Empty() bool
func (s *Stack[T]) Size() int

Running the example:

go build
./example -depth 3 -url https://golang.com

NOTE For more examples look at the thread_pool_test.go, where I implemented filling in a giant (4GB) buffer of bytes concurrently and parallelized some sorting algorithms.

Logging

zerolog is used as an underlying system for logging with custom settings to produce nicely formatted logs:

EXAMPLE 24 Mar 24 10:32 CET24 Mar 24 10:32 CET |DEBUG| Msg: worker finished CurrentThreads: 33