Many Thanks Ryan, You the man!!
--g On Jun 24, 2011, at 10:58 , Ryan Zezeski wrote: Gordon, Gilbert, and all you Search fans out there, I've patched this bug in the riak_search-0.14 branch. Below you'll find a link to the pull request. The bug was a little tricky to find but is fairly "obvious" once you see what is happening. The leak occurs when you perform an intersection query (i.e. `AND`) where one side is an empty result and the other side has results (which side is empty does not matter). Essentially, the query executor in Search will "short-circuit" the execution on either side and stop once one side has been exhausted but leave the results for the other side just hanging around. That is, the process that collected the results will wait indefinitely for another process to pull them off. Gordon, this is why your gist (https://gist.github.com/1010384) alleviated the leak. However, it's not an appropriate fix because you could lose data in more extreme conditions. My fix was to make sure both sides of an `AND` are exhausted after an answer has been found. TL;DR - Using an `AND` clause in your query where one side is empty (or small) and the other side has results (or is larger than the other side) will cause processes and it's corresponding msgs to sit around, waiting indefinitely. https://github.com/basho/riak_search/pull/66 -Ryan P.S. Gordon & Gilbert, thank you for taking the time to share your findings with us and especially Gordon for taking the time to make a reproducible test case which was key in helping me quickly solve this.
_______________________________________________ riak-users mailing list riak-users@lists.basho.com http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
