failure of MessageConsumerTest.testRedelMessageListener2

[jmesnil]

I've started to investigate this long-standing failure.

The failure is due to a incorrect redelivery after a message rollback.
The test expects the following sequence: "a RB a b c" (the "a" message is received, a rollback occurs, the "a" message is received again, then "b" and finally "c").

On my machine, the test fails intermittently with the sequence: "a RB b a c".
The "b" message is received *before* the message "a" is redelivered.
It also fails from time to time with a sequence: "a RB c a b". So it seems we have a race condition between our rollback code and message listener delivery.

I continue to investigate...

[clebert.suconic]

This test aways fail on my workstation with 8 cores.


And it aways pass if enable Nagle's algorithm.

[timfox]

This should be fixed now.

This means we should get a clear run apart from one AIO test that Clebert is looking into.

From now, the test suite should stay at green.