You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@zookeeper.apache.org by "Flavio Junqueira (JIRA)" <ji...@apache.org> on 2014/06/02 00:04:01 UTC

[jira] [Commented] (ZOOKEEPER-1928) add configurable throttling to the number of snapshots concurrently sent by a leader

    [ https://issues.apache.org/jira/browse/ZOOKEEPER-1928?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14015101#comment-14015101 ] 

Flavio Junqueira commented on ZOOKEEPER-1928:
---------------------------------------------

Thanks for the changes, Edward. I had a more careful look at the patch and here are some other comments, hope you don't mind:

# The configuration this is adding is only taking parameters through the command line. Although there has been some configuration introduced that doesn't use the config file, we typically prefer to have both options. Would you be ok with adding the config file option?
# In beginSnapshot, if an interrupted exception is thrown, shouldn't we also throw an exception, either snapshot throttle or extend it also throw interrupted exception?
# Minor: if I'm not mistaken, we are still using 80 characters per line, and there are some pretty long lines in leaner throttler. 

Other than that, it looks good to me.

> add configurable throttling to the number of snapshots concurrently sent by a leader
> ------------------------------------------------------------------------------------
>
>                 Key: ZOOKEEPER-1928
>                 URL: https://issues.apache.org/jira/browse/ZOOKEEPER-1928
>             Project: ZooKeeper
>          Issue Type: Improvement
>          Components: server
>            Reporter: Edward Carter
>            Assignee: Edward Carter
>         Attachments: ZOOKEEPER-1928.patch, ZOOKEEPER-1928.patch
>
>   Original Estimate: 168h
>  Remaining Estimate: 168h
>
> We want to add configurable throttling to the number of snapshots concurrently sent by a leader.  Without this, when recovering from a partial outage or network partition, the leader can become overloaded and unresponsive due to its attempts to send snapshots to too many followers and observers all at once.  The throttle will operate by terminating the connection of any observer receiving a snapshot deemed to be in excess of the throttle.  Followers should be allowed to receive snapshots unconditionally, though those snapshots do count against the quota.
> I have a patch ready which implements this.



--
This message was sent by Atlassian JIRA
(v6.2#6252)