[jira] [Commented] (GEODE-5437) Recovery of async persistent regions may result in inconsistency
[ https://issues.apache.org/jira/browse/GEODE-5437?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16562250#comment-16562250 ] ASF subversion and git services commented on GEODE-5437: Commit aaf030d81b14e5fbcbc661a262ae0d63240af1ba in geode's branch refs/heads/develop from agingade [ https://gitbox.apache.org/repos/asf?p=geode.git;h=aaf030d ] GEODE-5437: change async region gii to be synchronous (#2219) When a region uses async disk persistence it used to write its data to disk async but some of the gii meta-data synchronously. This could cause data inconsistency during recovery because it looked like the disk had the entire gii, when it may not have been able to write all the data. To fix this we now write all data to the disk store during gii synchronously for regions that are persistent even if they are configured async. DUnit tests have been added for both backup/restore and killing during gii. Unit tests have been added to cover the product changes. Also added bounceForcibly to the test framework which enables bounce to do a mean vm kill unlike bounce which always does a nice vm kill. > Recovery of async persistent regions may result in inconsistency > > > Key: GEODE-5437 > URL: https://issues.apache.org/jira/browse/GEODE-5437 > Project: Geode > Issue Type: Bug > Components: persistence >Reporter: Kirk Lund >Assignee: Anilkumar Gingade >Priority: Major > Labels: pull-request-available > Time Spent: 0.5h > Remaining Estimate: 0h > > The region initialized flag is written to IF file synchronously, but the data > is written to CRF asynchronously, then servers may become inconsistent during > recovery. > Proposed fix: > Async region should perform disk store flush BEFORE writing region > initialized flag to the IF file. -- This message was sent by Atlassian JIRA (v7.6.3#76005)
[jira] [Commented] (GEODE-5437) Recovery of async persistent regions may result in inconsistency
[ https://issues.apache.org/jira/browse/GEODE-5437?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16562342#comment-16562342 ] ASF subversion and git services commented on GEODE-5437: Commit aaf030d81b14e5fbcbc661a262ae0d63240af1ba in geode's branch refs/heads/feature/GEODE-5400 from agingade [ https://gitbox.apache.org/repos/asf?p=geode.git;h=aaf030d ] GEODE-5437: change async region gii to be synchronous (#2219) When a region uses async disk persistence it used to write its data to disk async but some of the gii meta-data synchronously. This could cause data inconsistency during recovery because it looked like the disk had the entire gii, when it may not have been able to write all the data. To fix this we now write all data to the disk store during gii synchronously for regions that are persistent even if they are configured async. DUnit tests have been added for both backup/restore and killing during gii. Unit tests have been added to cover the product changes. Also added bounceForcibly to the test framework which enables bounce to do a mean vm kill unlike bounce which always does a nice vm kill. > Recovery of async persistent regions may result in inconsistency > > > Key: GEODE-5437 > URL: https://issues.apache.org/jira/browse/GEODE-5437 > Project: Geode > Issue Type: Bug > Components: persistence >Reporter: Kirk Lund >Assignee: Anilkumar Gingade >Priority: Major > Labels: pull-request-available > Fix For: 1.7.0 > > Time Spent: 0.5h > Remaining Estimate: 0h > > The region initialized flag is written to IF file synchronously, but the data > is written to CRF asynchronously, then servers may become inconsistent during > recovery. > Proposed fix: > Async region should perform disk store flush BEFORE writing region > initialized flag to the IF file. -- This message was sent by Atlassian JIRA (v7.6.3#76005)