There's no formal definition of a pipeline in Go; it’s just one of many kinds of concurrent programs. Informally, a pipeline is a series of stages connected by channels, where each stage is a group of goroutines running the same function. In each stage, the goroutines:
- receive values from upstream via inbound channels
- perform some function on that data, usually producing new values
- send values downstream via outbound channels
Each stage has any number of inbound and outbound channels, except the first and last stages, which have only outbound or inbound channels, respectively. The first stage is sometimes called the source or producer; the last stage, the sink or consumer.
example pipeline
Consider this pipeline with three stages:
The first stage, gen, is a function that converts a list of integers to a channel that emits the integers in the list. The gen function starts a goroutine that sends the integers on the channel and closes when all the values have been sent.
func gen(nums ...int) <-chan int {
out := make(chan int)
go func() {
for _, n := range nums {
out <- n
}
close(out)
}()
return out
}
The second stage, sq, receives integers from a channel and returns a channel that emits the square of each received integer. After the inbound channel is closed and this stage has sent all values downstream, it closes the outbound channel.
func sq(in <-chan int) <-chan int {
out := make(chan int)
go func() {
for n := range in {
out <- n * n
}
close(out)
}()
return out
}
The main function sets up the pipelines and runs the final stage: it receives values from the second stage and prints each one, until the channel is closed:
func main() {
// Set up the pipeline.
c := gen(2, 3)
out := sq(c)
// Consume the output.
fmt.Println(<-out) // 4
fmt.Println(<-out) // 9
// We could instead write main as a range loop like the other stages. Also
// since sq has the same type for it's inbound and outbound channels, we can
// compose it any number of times.
for n := range sq(sq(gen(2, 3))) {
fmt.Println(n) // 16 then 81
}
}
fan-out, fan-in
Multiple functions can read from the same channel until that channel is closed; this is called fan-out. This provides a way to distribute work amongst a group of workers to parallelize CPU use and I/O.
A function can read from multiple inputs and proceed until all are closed by multiplexing the input channels onto a single channel that's closed when all the inputs are closed. This is called fan-in.
We can change the pipeline to run two instances of sq
, each reading from the
same input channel.
func main() {
in := gen(2, 3)
// Distribute the sq work across two goroutines that both read from in.
c1 := sq(in)
c2 := sq(in)
// Consume the merged output from c1 and c2.
for n := range merge(c1, c2) {
fmt.Println(n)
}
}
The merge function converts a list of channels to a single channel by starting a goroutine for each inbound channel that copies the values to the sole outbound channel. Once all the output goroutines have been started, merge starts one more goroutine to close the outbound channel after all sends on that channel are done. Sends on a closed channel cause a panic, so it's important to ensure all sends are done before calling close. The sync.WaitGroup type provides a simple way to arrange this.
func merge(cs ...<-chan int) <-chan int {
var wg sync.WaitGroup
out := make(chan int)
// Start an output goroutine for each input channel in cs. output copies
// values from c to out until c is close, then calls wg.Done.
output := func(c <-chan int) {
for n := range c {
out <- n
}
wg.Done()
}
wg.Add(len(cs))
for _, c := range cs {
go output(c)
}
// Start a goroutine to close out once all the output goroutines are done.
// This must start after the wg.Add call.
go func() {
wg.Wait()
close(out)
}()
return out
}
stopping short
There is a pattern to out pipeline functions:
- stages close their outbound channels when all the send operations are done.
- stages keep receiving values from inbound channels until those channels are closed.
This pattern allows each receiving stage to be written as a range loop and ensures that all goroutines exit once all values have been successfully sent downstream.
However, in real pipelines, stages don't always receive all the inbound values. Sometimes this is by design: the receiver may only need a subset of values to make progress. More often, a stage exits early because an inbound value represents an error in an earlier stage. In either case the receiver should not have to wait for the remaining values to arrive, and we want earlier stages to stop producing values that later stages don't need.
In the example pipeline, if a stage fails to consume all inbound values, the goroutines attempting to send those values will block forever:
out := merge(c1, c2)
// Consume only a single value from the output.
fmt.Println(<-out)
return
// Since we didn't receive the second value from out,
// one of the output goroutines is hung attempting to send it.
This is a resource leak: goroutines consume memory and runtime resources, and heap references in goroutine stacks keep data from being garbage collected. Goroutines are not garbage collected and must exit on their own!
buffers
We'd like to arrange for the upstream stages of our pipeline to exit even when the downstream stages fail to receive all the inbound values. One way to do this is to change the outbound channels to have a buffer. A buffer can hold a fixed number of values; send operations complete immediately if there's room in the buffer:
c := make(chan int, 2) // buffer size 2
c <- 1 // succeeds immediately
c <- 2 // succeeds immediately
c <- 3 // blocks until another goroutine does <-c and receives 1
When the number of values to be sent is known at channel creation time, a buffer
can simplify the code. For example, we can rewrite gen
to copy the list of
integers into a buffered channel and avoid creating a new goroutine:
func gen(nums ...int) <-chan int {
out := make(chan int, len(nums))
for _, n := range nums {
out <- n
}
close(out)
return out
}
While useful buffers cannot solve all cases. We cannot know how large to make
the buffer for merge
as it would depend on knowing exactly how many numbers
will be received per channel. (You could attempt to hardcode it but that results
in very fragile, intertwined, and generally messy code).
explicit cancellation
When main
decides to exit without receiving all the values from out
, it must
tell the goroutines in the upstream stages to abandon the values they're trying
to send. It does so by sending values on a channel called done
. It sends two
values since there are potentially two blocked senders.
func main() {
in := gen(2, 3)
// Distribute the sq work across two goroutines that both read from in.
c1 := sq(in)
c2 := sq(in)
// Consume the first value from output.
done := make(chan struct{}, 2)
out := merge(done, c1, c2)
fmt.Println(<-out) // 4 or 9
// Tell the remaining senders we're leaving.
done <- struct{}{}
done <- struct{}{}
}
The sending goroutines replace their send operation with a select
statement
that proceeds either when the send on out
happens or when they receive a value
from done
. The value type of done
is the empty struct (by convention) as the
value doesn't matter and it consumes little space. It is the receive event that
indicates the send on out
should be abandoned. The output
goroutines
continue looping on their inbound channel, c
, so the upstream stages are not
blocked.
func merge(done <-chan struct{}, cs ...<-chan int) <-chan int {
var wg sync.WaitGroup
out := make(chan int)
// Start an output goroutine for each input channel in cs. output
// copies values from c to out until c is closed or it receives a value
// from done, then output calls wg.Done.
output := func(c <-chan int) {
for n := range c {
select {
case out <- n:
case <-done:
}
}
wg.Done()
}
// ... the rest is unchanged ...