[jira] [Commented] (HDFS-6841) Use Time.monotonicNow() wherever applicable instead of Time.now()

["Kihwal Lee (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HDFS-6841?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14365300#comment-14365300 ] 

Kihwal Lee commented on HDFS-6841:
----------------------------------

We have seen multiple occasions of dead datanode coming back and trying to resend an IBR every 1 second. Normally the offerService loop makes it hreatbeat and re-register, but these nodes were skipping heartbeat and kept re-sending IBR.  I got a hold of one such node and got a heap dump to see what's going on.

{noformat}
lastBlockReport (J) : 1426164231598
lastCacheReport (J) : 0
lastDeletedReport (J) : 1426206179858
lastHeartbeat (J) : 2693080084078
{noformat}

Since {{lastHeartbeat}} is huge, it wasn't sending heartbeat.  We should move forward with this.

> Use Time.monotonicNow() wherever applicable instead of Time.now()
> -----------------------------------------------------------------
>
>                 Key: HDFS-6841
>                 URL: https://issues.apache.org/jira/browse/HDFS-6841
>             Project: Hadoop HDFS
>          Issue Type: Bug
>            Reporter: Vinayakumar B
>            Assignee: Vinayakumar B
>         Attachments: HDFS-6841-001.patch, HDFS-6841-002.patch, HDFS-6841-003.patch, HDFS-6841-004.patch
>
>
> {{Time.now()}} used  in many places to calculate elapsed time.
> This should be replaced with {{Time.monotonicNow()}} to avoid effect of System time changes on elapsed time calculations.



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