Make write pipeline retry times configurable.

Now recover the pipeline five times in a row for the same packet, will close the stream, but i think it should be configurable for different cluster needed.