I have a goroutine that will be run multiple times. But it can only run one at a time (single instance). What is the correct/idiomatic way to make sure a certain goroutine can run only one at a time?
Here is my contrived example code to illustrate the point:
func main() {
// Contrived example!!!!!!
// theCaller() may be run at multiple, unpredictable times
// theJob() must only be run one at a time
go theCaller()
go theCaller()
go theCaller()
}
func theCaller() {
if !jobIsRunning { // race condition here!
jobIsRunning = true
go theJob()
}
}
var jobIsRunning bool
// Can run multiple times, but only one at a time
func theJob() {
defer jobDone()
do_something()
}
func jobDone() {
jobIsRunning = false
}
Based on question and other comments from the OP, it looks like the goal is to start a new job if and only if a job is not already running.
Use a boolean variable protected by a sync.Mutex to record the running state of of the job. Set the variable to true when starting a job and to false when the job completes. Test this variable to determine if a job should be started.
var (
jobIsRunning bool
JobIsrunningMu sync.Mutex
)
func maybeStartJob() {
JobIsrunningMu.Lock()
start := !jobIsRunning
jobIsRunning = true
JobIsrunningMu.Unlock()
if start {
go func() {
theJob()
JobIsrunningMu.Lock()
jobIsRunning = false
JobIsrunningMu.Unlock()
}()
}
}
func main() {
maybeStartJob()
maybeStartJob()
maybeStartJob()
}
The lower-level sync/atomic package can also be used and may have better performance than using a mutex.
var jobIsRunning uint32
func maybeStartJob() {
if atomic.CompareAndSwapUint32(&jobIsRunning, 0, 1) {
go func() {
theJob()
atomic.StoreUint32(&jobIsRunning, 0)
}()
}
}
The sync/atomic package documentation warns that the functions in the package require great care to use correctly and that most applications should use the sync package.