Minimalistic and High-performance goroutine worker pool written in Go
This library is meant to provide a simple way to limit concurrency when executing some function over a limited resource or service.
Some common scenarios include:
- Executing queries against a Database with a limited no. of connections
- Sending HTTP requests to a a rate/concurrency limited API
- Zero dependencies
- Create pools with fixed or dynamic size
- Worker goroutines are only created when needed (backpressure detection) and automatically purged after being idle for some time (configurable)
- Minimalistic APIs for:
- Creating worker pools with fixed or dynamic size
- Submitting tasks to a pool in a fire-and-forget fashion
- Submitting tasks to a pool and waiting for them to complete
- Submitting tasks to a pool with a deadline
- Submitting a group of tasks and waiting for them to complete
- Submitting a group of tasks associated to a Context
- Getting the number of running workers (goroutines)
- Stopping a worker pool
- Task panics are handled gracefully (configurable panic handler)
- Supports Non-blocking and Blocking task submission modes (buffered / unbuffered)
- Very high performance and efficient resource usage under heavy workloads, even outperforming unbounded goroutines in some scenarios (See benchmarks)
- Configurable pool resizing strategy, with 3 presets for common scenarios: Eager, Balanced and Lazy.
- Complete pool metrics such as number of running workers, tasks waiting in the queue and more.
- New (since v1.7.0): configurable parent context and graceful shutdown with deadline.
- API reference
go get -u github.com/alitto/pondpackage main
import (
"fmt"
"github.com/alitto/pond"
)
func main() {
// Create a buffered (non-blocking) pool that can scale up to 100 workers
// and has a buffer capacity of 1000 tasks
pool := pond.New(100, 1000)
// Submit 1000 tasks
for i := 0; i < 1000; i++ {
n := i
pool.Submit(func() {
fmt.Printf("Running task #%d\n", n)
})
}
// Stop the pool and wait for all submitted tasks to complete
pool.StopAndWait()
}package main
import (
"fmt"
"github.com/alitto/pond"
)
func main() {
// Create an unbuffered (blocking) pool with a fixed
// number of workers
pool := pond.New(10, 0, pond.MinWorkers(10))
// Submit 1000 tasks
for i := 0; i < 1000; i++ {
n := i
pool.Submit(func() {
fmt.Printf("Running task #%d\n", n)
})
}
// Stop the pool and wait for all submitted tasks to complete
pool.StopAndWait()
}package main
import (
"fmt"
"github.com/alitto/pond"
)
func main() {
// Create a pool
pool := pond.New(10, 1000)
defer pool.StopAndWait()
// Create a task group
group := pool.Group()
// Submit a group of tasks
for i := 0; i < 20; i++ {
n := i
group.Submit(func() {
fmt.Printf("Running group task #%d\n", n)
})
}
// Wait for all tasks in the group to complete
group.Wait()
}This feature provides synchronization, error propagation, and Context cancelation for subtasks of a common task. Similar to errgroup.Group from golang.org/x/sync/errgroup package with concurrency bounded by the worker pool.
package main
import (
"context"
"fmt"
"net/http"
"github.com/alitto/pond"
)
func main() {
// Create a worker pool
pool := pond.New(10, 1000)
defer pool.StopAndWait()
// Create a task group associated to a context
group, ctx := pool.GroupContext(context.Background())
var urls = []string{
"https://www.golang.org/",
"https://www.google.com/",
"https://www.github.com/",
}
// Submit tasks to fetch each URL
for _, url := range urls {
url := url
group.Submit(func() error {
req, err := http.NewRequestWithContext(ctx, http.MethodGet, url, nil)
resp, err := http.DefaultClient.Do(req)
if err == nil {
resp.Body.Close()
}
return err
})
}
// Wait for all HTTP requests to complete.
err := group.Wait()
if err != nil {
fmt.Printf("Failed to fetch URLs: %v", err)
} else {
fmt.Println("Successfully fetched all URLs")
}
}Specifies the minimum number of worker goroutines that must be running at any given time. These goroutines are started when the pool is created. The default value is 0. Example:
// This will create a pool with 5 running worker goroutines
pool := pond.New(10, 1000, pond.MinWorkers(5))Defines how long to wait before removing idle worker goroutines from the pool. The default value is 5 seconds. Example:
// This will create a pool that will remove workers 100ms after they become idle
pool := pond.New(10, 1000, pond.IdleTimeout(100 * time.Millisecond))Allows to configure a custom function to handle panics thrown by tasks submitted to the pool. The default handler just writes a message to standard output using fmt.Printf with the following contents: Worker exits from a panic: [panic] \n Stack trace: [stack trace]). Example:
// Custom panic handler function
panicHandler := func(p interface{}) {
fmt.Printf("Task panicked: %v", p)
}
// This will create a pool that will handle panics using a custom panic handler
pool := pond.New(10, 1000, pond.PanicHandler(panicHandler)))Configures the strategy used to resize the pool when backpressure is detected. You can create a custom strategy by implementing the pond.ResizingStrategy interface or choose one of the 3 presets:
- Eager: maximizes responsiveness at the expense of higher resource usage, which can reduce throughput under certain conditions. This strategy is meant for worker pools that will operate at a small percentage of their capacity most of the time and may occasionally receive bursts of tasks. This is the default strategy.
- Balanced: tries to find a balance between responsiveness and throughput. It's suitable for general purpose worker pools or those that will operate close to 50% of their capacity most of the time.
- Lazy: maximizes throughput at the expense of responsiveness. This strategy is meant for worker pools that will operate close to their max. capacity most of the time.
// Example: create pools with different resizing strategies
eagerPool := pond.New(10, 1000, pond.Strategy(pond.Eager()))
balancedPool := pond.New(10, 1000, pond.Strategy(pond.Balanced()))
lazyPool := pond.New(10, 1000, pond.Strategy(pond.Lazy()))Configures a parent context on this pool to stop all workers when it is cancelled. The default value context.Background(). Example:
// This creates a pool that is stopped when myCtx is cancelled
pool := pond.New(10, 1000, pond.Context(myCtx))The following chart illustrates the behaviour of the different pool resizing strategies as the number of submitted tasks increases. Each line represents the number of worker goroutines in the pool (pool size) and the x-axis reflects the number of submitted tasks (cumulative).
As the name suggests, the "Eager" strategy always spawns an extra worker when there are no idles, which causes the pool to grow almost linearly with the number of submitted tasks. On the other end, the "Lazy" strategy creates one worker every N submitted tasks, where N is the maximum number of available CPUs (GOMAXPROCS). The "Balanced" strategy represents a middle ground between the previous two because it creates a worker every N/2 submitted tasks.
There are 3 methods available to stop a pool and release associated resources:
pool.Stop(): stop accepting new tasks and signal all workers to stop processing new tasks. Tasks being processed by workers will continue until completion unless the process is terminated.pool.StopAndWait(): stop accepting new tasks and wait until all running and queued tasks have completed before returning.pool.StopAndWaitFor(deadline time.Duration): similar toStopAndWaitbut with a deadline to prevent waiting indefinitely.
Each worker pool instance exposes useful metrics that can be queried through the following methods:
pool.RunningWorkers() int: Current number of running workerspool.IdleWorkers() int: Current number of idle workerspool.MinWorkers() int: Minimum number of worker goroutinespool.MaxWorkers() int: Maxmimum number of worker goroutinespool.MaxCapacity() int: Maximum number of tasks that can be waiting in the queue at any given time (queue capacity)pool.SubmittedTasks() uint64: Total number of tasks submitted since the pool was createdpool.WaitingTasks() uint64: Current number of tasks in the queue that are waiting to be executedpool.SuccessfulTasks() uint64: Total number of tasks that have successfully completed their exection since the pool was createdpool.FailedTasks() uint64: Total number of tasks that completed with panic since the pool was createdpool.CompletedTasks() uint64: Total number of tasks that have completed their exection either successfully or with panic since the pool was created
In our Prometheus example we showcase how to configure collectors for these metrics and expose them to Prometheus.
- Creating a worker pool with dynamic size
- Creating a worker pool with fixed size
- Creating a worker pool with a Context
- Exporting worker pool metrics to Prometheus
- Submitting a group of tasks
- Submitting a group of tasks associated to a context
Full API reference is available at https://pkg.go.dev/github.com/alitto/pond
See Benchmarks.
Here are some of the resources which have served as inspiration when writing this library:
- http://marcio.io/2015/07/handling-1-million-requests-per-minute-with-golang/
- https://brandur.org/go-worker-pool
- https://gobyexample.com/worker-pools
- https://github.com/panjf2000/ants
- https://github.com/gammazero/workerpool
Feel free to send a pull request if you consider there's something which can be improved. Also, please open up an issue if you run into a problem when using this library or just have a question.