(Note that sometimes, killing the fluent-bit pods and letting them restart suddenly results in a flood of backlogged messages hitting my fluentd, so I suspect something's up with the notification of log changes)