You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@geode.apache.org by "Jason Huynh (JIRA)" <ji...@apache.org> on 2017/03/16 23:13:41 UTC

[jira] [Created] (GEODE-2679) Lucene asynchronous disk writes for aeq can lead to data mismatch after compacting

Jason Huynh created GEODE-2679:
----------------------------------

             Summary: Lucene asynchronous disk writes for aeq can lead to data mismatch after compacting
                 Key: GEODE-2679
                 URL: https://issues.apache.org/jira/browse/GEODE-2679
             Project: Geode
          Issue Type: Bug
          Components: lucene
            Reporter: Jason Huynh


Due to the way krf and drf oplogs are created/updated, if the aeq has disk synchronous as false, there is a window of time where a krf is rolled/created with the current region entries, but events that had been removed have not yet been written to the drf.

We should probably change the aeq to be disk sync = true.  This will alleviate some of the data mismatch where we have potential extra keys after compacting the oplogs.




--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

Re: [jira] [Created] (GEODE-2679) Lucene asynchronous disk writes for aeq can lead to data mismatch after compacting

Posted by Michael Stolz <ms...@pivotal.io>.
Wow. This appears to be another GemFire basic plumbing issue.

I am so happy we embarked on this project.

It appears to be finding all kinds of issues in the AEQ implementation.

Please keep up this great work.

--
Mike Stolz
Principal Engineer - Gemfire Product Manager
Mobile: 631-835-4771

On Mar 16, 2017 7:20 PM, "Jason Huynh (JIRA)" <ji...@apache.org> wrote:

Jason Huynh created GEODE-2679:
----------------------------------

             Summary: Lucene asynchronous disk writes for aeq can lead to
data mismatch after compacting
                 Key: GEODE-2679
                 URL: https://issues.apache.org/jira/browse/GEODE-2679
             Project: Geode
          Issue Type: Bug
          Components: lucene
            Reporter: Jason Huynh


Due to the way krf and drf oplogs are created/updated, if the aeq has disk
synchronous as false, there is a window of time where a krf is
rolled/created with the current region entries, but events that had been
removed have not yet been written to the drf.

We should probably change the aeq to be disk sync = true.  This will
alleviate some of the data mismatch where we have potential extra keys
after compacting the oplogs.




--
This message was sent by Atlassian JIRA
(v6.3.15#6346)