Bevan Philip

> > Debian GNU/Linux 10 (buster), aarch64 (but the executable is arm32), pulsar-client-cpp 3.3.0 > > We've used the same broker and test with an x86_64 Pulsar 3.5.0 Pulsar client,...

So, looking at the internal stats for the topic, it does appear that the mark-delete position indicates that the full batch is indeed acknowledged for our aarch32 runs. ### Unsuccessful...

OK, this turns out to have been some strange oddity I can't reproduce in isolation. We were putting our MessageIds into a `pulsar::MessageId[]` array, and then acknowledging them all after...