If Bolt B’s receive queue is full the tuple will be put into an overflow buffer. If back pressure is not enabled it will keep on filling the overflow buffer and eventually cause an OOM.
You might also want to see the proposed changes in STORM-2306 where the receive queues are going to be bounded and the back pressure is allowed to propagate backwards up to the spout at which point the framework would stop invoking the nextTuple. Thanks, Arun From: Lawrence Craft <lawrence.cr...@gmail.com> Reply-To: "user@storm.apache.org" <user@storm.apache.org> Date: Tuesday, September 26, 2017 at 2:01 PM To: "user@storm.apache.org" <user@storm.apache.org> Subject: Behavior of Storm when buffers fill Hi all, I'm trying to get a better understanding of how Storm behaves in adverse conditions. I read http://www.michael-noll.com/blog/2013/06/21/understanding-storm-internal-message-buffers/, describing the various buffers between different workers and components. My question: if bolt A emits a tuple that is routed to Bolt B, but Bolt B's executor buffer is full, how does the system react? I would guess it'd be kept in the worker's receive buffer, but what happens when that fills? Does Bolt A's worker send thread block and stall? Obviously this is not an ideal situation and buffers/backpressure watermarks should be set precisely to avoid something like this. I'd just like to get an idea of what happens. Lawrence
