question regarding how to upgrade storm without downtime

[이승진 <sw...@navercorp.com>]

We have several topologies that should run continuously without halting.
 
1. how to release next version of topology without halting?
 
In some cases, we can run two same topologies in a cluster so that when running next version of topology jar, we can restart one by one not to halt entire service.
 
But if we cannot seperate incoming spout, this strategy cannot be applied also.
 
2. main question, how to upgrade storm without downtime?
 
Currently when we upgrade storm, we stop all the topologies then kill supervisors, nimbus, and launch new nimbus, supervisor, then run topologies again.
 
Is there any way to avoid this?
 
Maybe relaunching supervisor one by one and finally replacing nimbus would work without killing running topology, but I'm not sure about that.
 
do you have any idea?