Oh and the upside to actually getting this right beyond some bug prevention
and minor better behavior, is that the system can weather high load
situations dramatically better. For instance, it’s not uncommon if you are
hammering the system with indexing to start running into exceptions around
losing the zk connection that fail updates and then spiral into chaos. None
of this starts with a session expiration though. In a proper zk client
architecture you would almost never see that. Some connection loss would
happen, it would almost always recover before expiration, and all the
indexing would go through fine.
