Used Akka Projection (slick db, actor handler, AkkaProjectionVersion = 1.2.2) with 2 cluster singleton projections per node, 2 nodes in total. Noticed a time when half of the partitions started processing from offset 0 again.
AtLeastOnceAsync was used with the ff. setting for offset saving:
save-offset-after-envelopes = 100
save-offset-after-duration = 200 ms
Any insights as to what might have happened?