説明
I'm not positive on this, but @sergeybykov said to put in an issue to research/investigate possibilities:
@Kritner:
it seems like utilizing LoadShedding would prevent new clients from connecting in times of overloading... is there a way to disable that particular portion of load shedding?
@sergeybykov:
When load shedding kicks in, it starts rejecting client requests. I wonder if what you are seeing is not rejection of a client connection per se, but of it's first call to retrieve the grain type map. Could you open an issue on that please. Maybe we should consider special treatment of those calls (if my guess is correct).
I can try to put together a repro at some point, but from what I can recall, I set my LoadSheddingLimit value to a low amount - low enough that the silo would always be overloaded in my local testing. With the single silo cluster running (using development clustering) I was not successfully connecting to the silo from with a client. Once disabling LoadShedding or setting my threshold higher, connections went through w/o issue.
I guess if it's not the client connection failing, but the call @sergeybykov referenced about retrieving a grain type map, then that could complicate things a bit. I guess I was hoping that client connections (and everything that a connection entails) would be kept separate from the load shedding.