[jira] [Updated] (HBASE-6950) TestAcidGuarantees system test now flushes too aggressively

["Gregory Chanan (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/HBASE-6950?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Gregory Chanan updated HBASE-6950:
----------------------------------

    Attachment: HBASE-6950.patch
    
> TestAcidGuarantees system test now flushes too aggressively
> -----------------------------------------------------------
>
>                 Key: HBASE-6950
>                 URL: https://issues.apache.org/jira/browse/HBASE-6950
>             Project: HBase
>          Issue Type: Bug
>          Components: test
>    Affects Versions: 0.92.2, 0.94.2, 0.96.0
>            Reporter: Gregory Chanan
>            Assignee: Gregory Chanan
>            Priority: Minor
>             Fix For: 0.96.0
>
>         Attachments: HBASE-6950.patch
>
>
> HBASE-6552 caused the TestAcidGuarantees system test to flush more aggressively, because flushes are where ACID problems have occurred in the past.
> After some more cluster testing, it seems like this too aggressive; my clusters eventually can't keep up with the number of flushes/compactions and start getting SocketTimeoutExceptions.  We could try to optimize the flushes/compactions, but since this workload would never occur in practice, I don't think it is worth the effort.  Instead, let's just only flush once a minute.  This is arbitrary, but seems to work.
> Here is my comment in the (upcoming) patch:
> {code}
> // Flushing has been a source of ACID violations previously (see HBASE-2856), so ideally,
> // we would flush as often as possible.  On a running cluster, this isn't practical:
> // (1) we will cause a lot of load due to all the flushing and compacting
> // (2) we cannot change the flushing/compacting related Configuration options to try to
> // alleviate this
> // (3) it is an unrealistic workload, since no one would actually flush that often.
> // Therefore, let's flush every minute to have more flushes than usual, but not overload
> // the running cluster.
> {code}

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira