I am really not sure - this is the third cluster I have found it on myself (all in the last couple releases). oomkills are rather prevalent on our systems with billions of messages being inserted (which is another problem I would like to solve). The only thing I can think of is a oomkill does not allow badger to close correctly or something. As I put above, this is on GKE. I have seen it on systems with one group and systems with multiple groups. All of the systems had 3 shard replicas in a group (in case it has to do with tablet replication)