hi, i would like to ask a question that is not really hadoop related.
i am working on a message system and find that every hour there happens 3 full gc, more or less. and after each full gc, there is an message burst, and because there is an pipeline to consume the messages, and it needs time to finish working on analyzing and even being persisted on disk. always there encountered queue overflow, lot of messages are dropped even though we have second cached queue in the pipeline. since we have no way to completely avoiding full gc, can any one offer an approach to solve these issues? any comment is welcomed. thanks! uknow.
