Pool#join not working ?

[meh]

Pool#join not working ?

[meh]

Yeah, that solution is unacceptable, I know about the issue on MRI, it's a bug in the deadlock inference, I opened a bug on Ruby's issue tracker but I was unable to come up with a reduced testcase, so they ignored it.

I'll try to find a proper solution, but it really isn't easy, pools are supposed to be used in long running applications rather than to achieve parallelization in simple scripts.

[meh]

No wait, now that I think of it, you should use #shutdown and not #join there, the pool is never going to be closed if you join.

I think I should make #join private, or work differently, as in "wait for all the tasks to be consumed", what do you think?

[Kris-LIBIS]

The wait for all the tasks to be consumed is actually what I wanted to achieve.

I have a batches of workload that comes in at irregular times. Each batch should be handled in different stages. I need to be sure that a batch has finished stage 1 completely before I start with stage 2. Then stage 3 and so on. When the batch is completely processed, it will wait for the next batch.

I want to use the same Pool for stage 1, stage 2, stage 3, .... which is why I cannot call shutdown. Actually, I wanted to have 4 threads minimum with a max of up to 20 threads, but - of course - any minimum not 0 does not work with join.

I'm actually wondering what the join method is supposed to do, currently? AFAIK it will just hang. I replaced worker.join with a simple sleep(0.3) and trim and that works just as well. worker.join seems to never return.

[Kris-LIBIS]

I've forked the code and trying to implement a wait that returns when (@todo.empty? and @waiting == @Spawned). That should do the trick, not?

[meh]

#join right now is useful if you call #shutdown! and want to wait for the workers to end and it's used internally by #shutdown.

So yes, it's kind of useless and not what you want as is.

I think the best way to go is to merge the current #join in #shutdown and make #join wait for the current tasks to finish.

Depending on how you implement the waiting, having @todo.empty? && @waiting == @spawned should be a good check.

[Kris-LIBIS]

OK. I'll have a go. I'll let you know when I'm done.

[Kris-LIBIS]

BTW. Do you agree that the @trim_requests -= 1 is a bug? Or am I mistaken?

[meh]

You're mistaken, the number of trim requests is decreased when one of the workers is closed, if you increment it all the workers will die.

[Kris-LIBIS]

Are you sure? I was talking about the line in the #trim method. Shouldn't the trim_requests go up when a new trim request is made?

[Kris-LIBIS]

Auto_trim did not work for me unless I changed that line to += ...

[meh]

Oh, yeah, I'm sorry, you're right, I was looking in the worker instead of #trim.