[ https://issues.apache.org/jira/browse/CASSANDRA-8835?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14329950#comment-14329950 ]
Pavel Yaskevich commented on CASSANDRA-8835: -------------------------------------------- So I did similar thing to Netty, it detects if selector returned N times in a row with empty selection of keys then replaces selector to newly allocated one and migrates keys there, work is pushed to https://github.com/xedin/disruptor_thrift_server/tree/100_percent_cpu_epoll (that also has patch for complete read operation reverted), so [~rbranson] please give it a try and let me know. > 100% CPU spikes from disruptor thrift server > -------------------------------------------- > > Key: CASSANDRA-8835 > URL: https://issues.apache.org/jira/browse/CASSANDRA-8835 > Project: Cassandra > Issue Type: Bug > Components: Core > Environment: 2.0.11, hsha, JDK7u65, ~6,000 connections per machine > Reporter: Rick Branson > Assignee: Pavel Yaskevich > > Seeing big CPU spikes (to 100%) inside of the TDisruptorServer when running > hsha. perf top loaded with JVM symbols shows this one floating to the top for > anywhere between 15 and 90 seconds and then falling down: > Lcom/thinkaurelius/thrift/TDisruptorServer$SelectorThread;.selectorIterationComplete > in Lcom/thinkaurelius/thrift/TDisruptorServer$AbstractSelectorThread;.select -- This message was sent by Atlassian JIRA (v6.3.4#6332)