Storm的一些重要的bug issues,stormbugissues


Fix cascading Storm failure by improving reconnection strategy and buffering messages https://issues.apache.org/jira/browse/STORM-329
上面的isusse是storm从0.9.3升级到0.9.4主要的优化,去掉了0.9.3中对于worker之间通信的bug。0.9.3对于worker之间如果连接不上就直接throw一个exception出来,导致该worker crash,那么和该worker 连接的worker也有可能在连接超时时crash,从而导致worker级联crash,job无法正常工作,数据丢失。
0.9.4改进的是不抛出异常,直接丢弃message,等待刷新重连,具体讨论可见该issue

Make worker receiver thread number configurable

https://issues.apache.org/jira/browse/STORM-272
Nathan Marz added a comment - 04/Apr/14 01:28 It would be a good experiment, but we'd have to make sure that incoming messages from the same task always go to the same receiver thread (otherwise messages sent by a task to another task could be received out of order)
Storm Performance cannot be scaled up by adding more CPU cores https://issues.apache.org/jira/browse/STORM-297 https://github.com/apache/storm/pull/103

相关内容