You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@helix.apache.org by "Kanak Biscuitwala (JIRA)" <ji...@apache.org> on 2014/03/18 00:22:44 UTC
[jira] [Created] (HELIX-409) Some messaging patterns are unreliable
Kanak Biscuitwala created HELIX-409:
---------------------------------------
Summary: Some messaging patterns are unreliable
Key: HELIX-409
URL: https://issues.apache.org/jira/browse/HELIX-409
Project: Apache Helix
Issue Type: Bug
Reporter: Kanak Biscuitwala
There is a usage pattern where a task is started, and Helix's messaging framework is used to post status updates. Specifically, the controller serves as the liason for status updates, as it checks completion status, and updates an overall status to indicate completion.
Having the controller in the middle of this workflow is the cause of several problems:
- Polling of zookeeper
- Writes to zookeeper for every status change
- If the controller dies, the controller that comes up doesn't know that it has something to process, as much of this management work is done in memory
Generally speaking, this is probably not the right way at all to use Helix messaging, as "send and forget" point-to-point messaging works more naturally. However, if we actually want to support use cases like the one causing these problems, we need a fix.
--
This message was sent by Atlassian JIRA
(v6.2#6252)