[ 
https://issues.apache.org/jira/browse/PHOENIX-4924?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Bin Shi updated PHOENIX-4924:
-----------------------------
    Comment: was deleted

(was: h2. Conclusion:
 # The test result clearly shows that the change doesn't introduce any new 
performance regression and issues.
 # The test result clearly shows that the pattern of “Periodical query spikes” 
observed in blocking stats cache (see the design document [Use Asynchronous 
Refresh to Provide Non-blocking Phoenix Stats 
Cache|https://salesforce.quip.com/rxokAkVatiQO] for details) is gone.
 # When analyzing the noises in the test result of the Flighting, I suspected 
that it is because multiple loading stats async tasks are triggered for the 
same cached entry during the same period. By exploring the code base of Google 
Guava Cache, I found this negative case won't happen, because Google Guava 
Cache won't try to reload a cache entry when there is another thread performing 
the refresh.

h2. Test Result:
h3. Baseline 1st run

!x8q4HN0IRAZ0wAAAABJRU5ErkJggg==!
h3. Flighting 1st RUN

!B8EntiMHCbVvAAAAAElFTkSuQmCC!
h3. Baseline 4th Run

!0HRAABAKBQCCweRHwyX0jwmTzykK0PBAIBFYfgf8PqgsrMqhZgIIAAAAASUVORK5CYII=!
h3. Flighting 4th Run

!j8FfZ 2FQd2ugAAAABJRU5ErkJggg==!
h2.  )

> Provide strong guarantee for Stats Write to prevent inconsistent and 
> incomplete data issue
> ------------------------------------------------------------------------------------------
>
>                 Key: PHOENIX-4924
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-4924
>             Project: Phoenix
>          Issue Type: Improvement
>            Reporter: Bin Shi
>            Assignee: Bin Shi
>            Priority: Major
>
> The Stats could be inconsistent and incomplete due to region servers going 
> down, RPC failures when committing Guidepost data to Stats table and the lack 
> of retry and atomic operation. The expected behavior from the Platform is “We 
> need a strong guarantee that on stats write there are sufficient retries and 
> monitoring in place so that stats writes are resilient such that we will not 
> end up with inconsistent, partial or empty stats until the next run of UPDATE 
> STATISTICS”. 



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to