Why use atomic.AddInt32() and sync.Mutex at the same time

[zhu121]

File：pond.go Function：decrementWorkerCount() and incrementWorkerCount() Question： In Go language, atomic packages supply lower-level atomic memory that is helpful is implementing synchronization algorithms. The AddInt32() function in Golang is used to atomically add delta to the *addr. In addition, using the sync.Mutex to ensure the mutually exclusive execution of a piece of code. Aren't these two equivalent？

[alitto]

Hey @zhu121,

The reason the mutex lock is needed there is to prevent decrementWorkerCount decreasing the workerCount variable while this block is executed in incrementWorkerCount, which could bring undesirable consequences, such as adding more workers than necessary or not starting new ones when needed. And for this to happen, the runningWorkerCount variable in that block must reflect the actual number of worker goroutines during the entirety of the block. A similar thing could happen if incrementWorkerCount was executed simultaneously from 2 different threads. Does it make sense?

[zhu121]

Hey @zhu121,

The reason the mutex lock is needed there is to prevent decrementWorkerCount decreasing the workerCount variable while this block is executed in incrementWorkerCount, which could bring undesirable consequences, such as adding more workers than necessary or not starting new ones when needed. And for this to happen, the runningWorkerCount variable in that block must reflect the actual number of worker goroutines during the entirety of the block. A similar thing could happen if incrementWorkerCount was executed simultaneously from 2 different threads. Does it make sense?

Hey @alitto, thank you for your reply, I understand it