this is an extension of the fix done for ENTMQBR-7422. With multiple consumers this time.

When the messages are placed back from multiple consumers, the messages are spread between two consumers, and when they are sent back to the queue they are shuffled what makes the addSorted to ignore the last optimization and do the addSortedScan, and if you have multiple messages in the queue already, and is delivering back a large buffer that will create a massive return to the queue buffer.

This time it will not crash the broker like it did in ENTMQBR-7422 but the process will take a while depending on the size because of the scan.

I have changed the last test I added in ENTMQBR-7422 to use multiple consumers. The logic is a small enhancement on top of the previous logic and it will be valid in both cases.

You have to disable flow control to hit this issue where there would be a lot of pending messages in the queue, however this could have a negative impact in OpenWire since that protocol disables flow control by credits and only takes the TCP flow control, allowing more messages to be pending in the buffer.