ambari-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Siddharth Wagle (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (AMBARI-8769) Aggregator checkpoint logic should take into account the checkPointDelay
Date Wed, 17 Dec 2014 21:03:13 GMT

     [ https://issues.apache.org/jira/browse/AMBARI-8769?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Siddharth Wagle updated AMBARI-8769:
------------------------------------
    Description: 
Aggregator checkpoint logic should take into account the checkPointDelay. When the checkpoint
is saved it is done with a currentTime minus the checkPointDelay (this is a configurable value),
when it is compared the next iteration the checkPointDelay is not taken into account.

In effect when the delay is the same or greater than interval, aggregator will not work. 

Example configuration that should work, and currently is broken (can be edited using Ambari
on metrics collector service page):
{noformat}
timeline.metrics.cluster.aggregator.minute.checkpointCutOffMultiplier = 2
timeline.metrics.cluster.aggregator.minute.interval = 120
{noformat}

Solution:
Fix the logic that checks if checkpoint is too old 

  was:
Aggregator checkpoint logic should take into account the checkPointDelay. When the checkpoint
is saved it is done with a currentTime minus the checkPointDelay (this is a configurable value),
when it is compared the next iteration the checkPointDelay is not taken into account.

In effect when the delay is the same or greater than interval, aggregator will not work. 

Example configuration that should work, and currently is broken (can be edited using Ambari
on metrics collector service page):
{noformat}
timeline.metrics.cluster.aggregator.minute.checkpointCutOffMultiplier = 2
timeline.metrics.cluster.aggregator.minute.interval = 120
{noformat}

A solution:
We should either fix the logic that checks if checkpoint is too old, or ensure that the delay
is always much smaller than interval. 


> Aggregator checkpoint logic should take into account the checkPointDelay
> ------------------------------------------------------------------------
>
>                 Key: AMBARI-8769
>                 URL: https://issues.apache.org/jira/browse/AMBARI-8769
>             Project: Ambari
>          Issue Type: Bug
>          Components: ambari-metrics
>    Affects Versions: 2.0.0
>            Reporter: Siddharth Wagle
>            Assignee: Siddharth Wagle
>             Fix For: 2.0.0
>
>         Attachments: AMBARI-8769.patch
>
>
> Aggregator checkpoint logic should take into account the checkPointDelay. When the checkpoint
is saved it is done with a currentTime minus the checkPointDelay (this is a configurable value),
when it is compared the next iteration the checkPointDelay is not taken into account.
> In effect when the delay is the same or greater than interval, aggregator will not work.

> Example configuration that should work, and currently is broken (can be edited using
Ambari on metrics collector service page):
> {noformat}
> timeline.metrics.cluster.aggregator.minute.checkpointCutOffMultiplier = 2
> timeline.metrics.cluster.aggregator.minute.interval = 120
> {noformat}
> Solution:
> Fix the logic that checks if checkpoint is too old 



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

Mime
View raw message