This issue is coming up frequently of testing non-standard configurations of the server since the errors are tending to happen early on, so it is something I need fixed. I looked at the RootLoggerService and followed the logging down to the LoggerNode.publish(ExtLogRecord) to get a quick idea of what the path from the log messaes to the handlers is.
What if the LoggerNode.clearHandlers() accepts a List<LogRecord> and this is populated with all log msgs that attempt to be published while the handlers are being cleared, and then the RootLoggerService republishes these to the server.log. I'm thinking that duplicates between the logs are ok if it simplifies how this is done. Right now I really don't know all that is involved to capture these msgs, so I'll start along this approach. If it is the wrong way to go, chime in!