We are seeing some weird behavior: for one of topics, counts are perfectly doubled in the real-time layer. I’ve seen a discussion recently in this user group, which says that changing the number of partitions can cause some duplication, but we didn’t change the number for the last week or so. Anyway, segment granularity is defined to be daily, so if we would change it the counts were become correct on the day after, but this isn’t the case.
Events are sent via Tranquility API. We use 3 partitions and 2 replicants for this topic.
How can I debug this issue?
(Ingestion spec of one of the tasks is attached).
ingestion_spec.json (3.16 KB)