Hello everyone, I contribute to implementing the apache james mailqueue contract on top of pulsar. I encountered a surprising situation and I would like to know if it's my expectations or if I'm doing something wrong.
A bit of context : I have a producer configured on a persistent topic with batching explicitly disabled I have a subscription on that same topic, the consumers of that subscription are configured with a subscription type shared and an initial position earliest. I have a negative ack delay of one second and I acknowledge messages individually. I made sure that I only have a single shared subscription on the topic. With this setup I expect to be able to do out of order acks. And indeed message production and consumption works as expected and the consumers act as a worker pool. The issue: I want to be able to list all the **unacked** messages that are in the topic waiting to be consumed. To do that, I create a reader on the topic which starts at the earliest and iterate over the messages. However this yields acked messages which I didn't expect. I have tried adding an artificial delay between the ack of the consumed message and the read to no avail I built a junit integration test which uses test containers to start a pulsar 2.9.1 (and we use the corresponding pulsar client) and connects the mailqueue implementation to it. I have uploaded the logs (which mix the log output of the pulsar container and the logs of our implementation) at https://logpaste.com/8uahGvHm The part we try to figure out starts at 12:44:07.426 which is a call to the admin api to get the topic stats which shows among other things: msgBacklog=0, backlogSize=0, msgBacklogNoDelayed=0 and lastAckedTimestamp=1642160642430, lastMarkDeleteAdvancedTimestamp=1642160642442 yet at 12:44:07.456 the reader received a message is there a way to filter out messages that have been acked when reading ? Thanks jean
