because the C consumer library acknowledges messages ... so the application layer cannot.
When we test failures (but restarting sarracenia, or the broker) the C consumer (aka sr3_cpump) is likely to lose messages. The C consumer is not currently used operationally anywhere, it's more of a demonstrator (can replace a shovel or a winnow) but with the caveat from that it doesn't deal with failure well:
because the C consumer library acknowledges messages ... so the application layer cannot. When we test failures (but restarting sarracenia, or the broker) the C consumer (aka sr3_cpump) is likely to lose messages. The C consumer is not currently used operationally anywhere, it's more of a demonstrator (can replace a shovel or a winnow) but with the caveat from that it doesn't deal with failure well:
https://github.com/MetPX/sarrac/issues/121
So remove the tests that require C consumer to be 100% reliable, and the flow tests should pass more often.