[jira] [Commented] (CASSANDRA-4925) Failure Detector should log or ignore sudden time change to the past

["Jonathan Ellis (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/CASSANDRA-4925?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13788869#comment-13788869 ] 

Jonathan Ellis commented on CASSANDRA-4925:
-------------------------------------------

Thanks!

What about this? Looks like we're mixing millis + nanos.

{code}
        if (tLast != 0L)
        {
            interArrivalTime = (value - tLast);
        }
        else
        {
            interArrivalTime = Gossiper.intervalInMillis / 2;
        }
{code}


> Failure Detector should log or ignore sudden time change to the past
> --------------------------------------------------------------------
>
>                 Key: CASSANDRA-4925
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-4925
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>            Reporter: sankalp kohli
>            Assignee: sankalp kohli
>            Priority: Minor
>         Attachments: trunk-4925.patch
>
>   Original Estimate: 1h
>  Remaining Estimate: 1h
>
> If a machine goes back in time all of a sudden because of a problem, Gossip will insert a negative interArrivalTime. 
> This will decrease the mean value and can cause this machine to mark other nodes as down and then mark them up as time passed. 
> But we should log such occurrences.



--
This message was sent by Atlassian JIRA
(v6.1#6144)