The problem is to do with the optimisation in the BlockingWaitStrategy that tries to avoid waking up the consumers if none of them are waiting. However due to a bug in that code it possible for the consumers to miss a potential wake-up and the ring buffer ends up with a message sitting unconsumed in the buffer and it won't be processed until the next message arrives. This actually happened in our production system. We had a latency spike of 6 seconds do to a missed wake-up.

Definitely upgrade, either to 2.10.4 - which removes the optimisation altogether or even better use the latest version 3 of the Disruptor which has some significant performance improvements for multiple publishers.

@mikeb01 , sincerely thank you for your help.

We are trying to apply ver2.10.4 and use waitfor instead of waitfor(timeout). So we want to do unit test to make sure this bug is fixed. But I can't reproduce it using ver2.10.1. So is there any way to reproduce this race condition?

I can't think of a test that will deterministically fail for the v2.10.1. This code ran fine in production for over 7 months before we noticed the issue. To reproduce I think you need to have a producer sleep for a short (random) amount of time then publish a bunch of messages. Then wait to see if they've all been consumed. Eventually you see a situation where the consumer will be sitting idle and not all of the messages will have been consumed.