Le 27/02/16 18:08, Buddy Butterfly a écrit : > Obviously netty fixed this with a workaround.
Thanks for the interesting analysis and pointers. It seems that in some case, we are still looping on the select(). Can you do one more thing to confirm that it's teh real problem ? Set the log to WARN, you should get messages like : Create a new selector. Selected is 0, delta = XXX This would be the signal that we detected a stalled selector (ie a selector that return immediately on select(), even if no event has arrived, leading to a fast loop eating all the CPU). In thsi case, we recreate a new selector. In your case, and accordingly to the bug report, there might be a case where creating a new selector does not solve the issue. Many thanks !