Missing entries (events) in cassandra messages table

Hi, I have a strange issue with my persistence/eventsourcing/cluster/sharding system.

I have millions of persistent actors that are short-living (2 to 7 days, mostly only 2) that model state. The idea is to have them passivated after some time and thus have only active in memory. There are no snapshots involved.

I noticed that after some time restarting k8s pods lead to errors like the following in logs:

java.lang.IllegalStateException: Sequence number [1] still missing after [10.00 s], saw unexpected seqNr [3] for persistenceId [Entity|02040044000060].

Checking the journal (the cassantra messages table) shows that there are indeed only events with sequenceNr 3, 4 and 5 for given peristence Id.

The sharding is configured as following:

    sharding {
      role = "state"
      state-store-mode = ddata
      passivate-idle-entity-after = 48 h

and the actor is implemented very close to the example given in the documentation:

Behaviors.setup { context =>
  EventSourcedBehavior.withEnforcedReplies[EntityCommand, EntityEvent, EntityState](
    persistenceId  = PersistenceId(EntityKey.name, EntityId),
    emptyState     = EntityState.empty,
    commandHandler = commandHandler(context),
    eventHandler   = eventHandler(context)
  .onPersistFailure(SupervisorStrategy.restartWithBackoff(minBackoff = 5.seconds, maxBackoff = 30.seconds, randomFactor = 0.1))

I’d be very grateful for any suggestions or ideas about fixing this issue.