[ 
https://issues.apache.org/jira/browse/PHOENIX-6298?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17259050#comment-17259050
 ] 

Geoffrey Jacoby commented on PHOENIX-6298:
------------------------------------------

I just want to point out that this only affects the old indexing framework, 
because the new one doesn't auto-disable indexes. (It repairs write failures at 
read time.) 

Since existing indexes have to be manually upgraded to the new framework 
(either by dropping/recreating or by using the IndexUpgradeTool), we'll need to 
support the old Indexer-style indexes for a little while yet, so still good to 
fix bugs / race conditions when we find them. 

> Use timestamp of PENDING_DISABLE_COUNT to calculate elapse time for 
> PENDING_DISABLE state
> -----------------------------------------------------------------------------------------
>
>                 Key: PHOENIX-6298
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-6298
>             Project: Phoenix
>          Issue Type: Bug
>    Affects Versions: 5.0.0, 4.15.0
>            Reporter: Ankit Singhal
>            Assignee: Viraj Jasani
>            Priority: Major
>             Fix For: 5.1.0, 4.16.0
>
>
> Instead of taking indexDisableTimestamp to calculate the elapsed time, we 
> should be considering the last time we incr/decremented the counter for 
> PENDING_DISABLE_COUNT. as if the application writes failures span more than 
> the default threshold of 30 seconds, the index will unnecessarily get 
> disabled even though the client could have retried and made it active.
> {code}
>                     long elapsedSinceDisable = 
> EnvironmentEdgeManager.currentTimeMillis() - Math.abs(indexDisableTimestamp);
> // on an index write failure, the server side transitions to PENDING_DISABLE, 
> then the client
>                     // retries, and after retries are exhausted, disables the 
> index
>                     if (indexState == PIndexState.PENDING_DISABLE) {
>                         if (elapsedSinceDisable > pendingDisableThreshold) {
>                             // too long in PENDING_DISABLE - client didn't 
> disable the index, so we do it here
>                             IndexUtil.updateIndexState(conn, 
> indexTableFullName, PIndexState.DISABLE, indexDisableTimestamp);
>                         }
>                         continue;
>                     }
> {code}



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to