I currently can't reproduce this in the non-http code, so it may be a bug in the refactored http server's use of SinkResponder and SourceResponder. Another theory is that it relates to the order of events in the http code, with receiving of a new batch being nested within a flush.
I currently can't reproduce this in the non-http code, so it may be a bug in the refactored http server's use of SinkResponder and SourceResponder. Another theory is that it relates to the order of events in the http code, with receiving of a new batch being nested within a flush.