You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@storm.apache.org by "Brian Johnson (JIRA)" <ji...@apache.org> on 2015/07/16 02:29:04 UTC

[jira] [Created] (STORM-940) Lost messages with netty

Brian Johnson created STORM-940:
-----------------------------------

             Summary: Lost messages with netty
                 Key: STORM-940
                 URL: https://issues.apache.org/jira/browse/STORM-940
             Project: Apache Storm
          Issue Type: Bug
    Affects Versions: 0.9.5
            Reporter: Brian Johnson


We have a topology that mysteriously stopped several times. On the most recent occasion it had a max spout pending of 4 and there were 4 transactions in zookeeper. We are using transactional topologies and the final bolt in the topology is a global grouping with only one task. We have verified that the first transaction finished successfully right up to the last line of the finish batch, but the tx is still in zookeeper and all of the other transactions are waiting on it. I know transactional topologies have been deprecated, but regardless this shouldn't happen unless the acks are being dropped. We have had a bunch of similar issues since moving to the netty based versions and I think there is a serious reliability issue with them. Sorry I can't provide code to replicate the issue because the problem is intermittent. 



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)