GitHub user mdykman opened a pull request: https://github.com/apache/kafka/pull/34
changed List type in RequestPurgatory.Watchers class Watchers is declared, having member private val requests = new util.ArrayList[T] in purgeSatisfied(), an iterator is created which conditionally removes elements from the list as it goes: RequestPurgatory.scala, lines 193-201: val iter = requests.iterator() var purged = 0 while(iter.hasNext) { val curr = iter.next if(curr.satisfied.get()) { iter.remove() purged += 1 } } Using the .remove operation on an ArrayList iterator is very expensive as the ArrayList promises a contiguous backing array and all higher elements must be shifted on every operation. A LinkedList is optimized for for the removal of arbitrary elements. Therefore recommending: - private val requests = new util.ArrayList[T] + private val requests = new util.LinkedList[T] This case was identified due to an observed failure In a high-load production environment, which found one or more cores pinned (ie. @100 usage) insider this loop. Once the pin was established, it tended to remain until the system was restarted. As the loop is within a synchronized block, checkAndMaybeAdd (0.8.2) or add (0.8.1) which are also synchronized on the same object, may become blocked by this inefficiency. You can merge this pull request into a Git repository by running: $ git pull https://github.com/mdykman/kafka 0.8.2 Alternatively you can review and apply these changes as the patch at: https://github.com/apache/kafka/pull/34.patch To close this pull request, make a commit to your master/trunk branch with (at least) the following in the commit message: This closes #34 ---- commit ade6abea822b87ac72309ecd6c43cfe4adc81e22 Author: Michael Dykman <mdyk...@gmail.com> Date: 2014-10-03T20:11:14Z changed List type in RequestPurgatory.Worker ---- --- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. ---