The C++ broker appears to segfault during MultipleTransactedBatchProducerTest. Several instances can be found in the history from our CI jobs:
During course of the latest test failure in #496 (logs attached), the broker logged receiving yet another TxCommit command:
After this nothing else is ever logged from the broker process, but exceptions start flying from the client indicating the connections were reset:
When the test tearDown occurs, it is logged that the broker process exited with an abnormal looking exit code, which a quick google suggests is the result of a segfault:
The test fails after 75sec, which is the exact time the test itself allows for the process undertaken to complete (artificially high to allow for extreme slowdowns on the CI nodes which happened occasionally in the past and also the C++ broker being slower at the test due to use of client side selectors; passing results show this usually takes 1-3sec on the Java broker and <10sec on the [transient] C++ broker), however the test harness tries to clean up the connections use in the test during tearDown and actually ends up reporting issues cleaning those up rather than the fact that the CountDownLatch used in the test wasnt decremented enough times.