apple
WorkingPool unexpected behaviour when workers are on different node
Description WorkingPool unexpected behaviour when workers are on different node
Steps to reproduce https://github.com/akbashev/WorkerPoolTest
Two nodes:
- Master
- Worker
- Join nodes and wait to be up.
- Spawn master with working pool on Master node.
- Spawn workers on Worker node.
If you run an example and submit some work—WorkingPool will terminate all workers in selectWorker() function, seems like actor is none here:
if let worker = self.workers[selectedWorkerID]?.actor {
Expected behavior Pool is routing job to workers, e.g. will log:
2023-07-20T07:36:38+0200 info worker : cluster/node=sact://[email protected]:1111 [WorkingPoolTest] Done check for /user/Worker-d
Environment MacOS 14.0 Beta (23A5286i), Xcode 15.0 Beta 4 (15A5195m), Swift 5.9
btw think I've pushed a error in SPM in example before 🙈 fixed that, now should work
~~Ok, after a bit of testing and checking repo around, think this PR and particular WeakWhenLocal wrapper can fix this issue. Will double check.~~
Probably there should be a better way to fix :)
Actually looking back again into issue and thinking a bit more about, introduction of some type like WeakWhenLocal makes sense.
Making worker reference either just weak or just strong will both give you a problem:
- Weak references of remote actors will just be cleaned up by local system as there are no other references to this actor.
- Strong references of local actors will create unwanted reference between worker and worker pool and won't be cleaned up from memory.
So you (or system) need to know if it's local or remote reference for WorkerPool. And this PR is actually should fix it. 🤔