x0id / erlsem Goto Github PK

Hi Dmitriy,
I see that you found the cause of the Shackle backlog leak: x0id/shackle@570bf12. I assume this project was created to fix the race condition that existed in the Erlang code?

One thing I noticed is that your "backlog re-implementation" doesn't allow for the same PID to take the lock since it returns {error,duplicate_pid}. This is a breaking change from the current backlog implementation. It is totally acceptable that one process calls multiple time the cast function for a specific client. In the case that the pool is very large it might still work, but for small pools it's going to limit the single process throughput significantly.

Thoughts?

Cheers,
LP