[jbossts-issues] [JBoss JIRA] Commented: (JBTM-821) Cope With Zero Length Transaction Log Files
Jonathan Halliday (JIRA)
jira-events at lists.jboss.org
Thu Jan 27 05:15:50 EST 2011
[ https://issues.jboss.org/browse/JBTM-821?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12578175#comment-12578175 ]
Jonathan Halliday commented on JBTM-821:
The file system based stores will create and file and then write it, these actions being distinct and non-atomic. Thus a crash may leave a zero length file in the store. This is currently regarded as valid, although it's impossible to activate the corresponding object. This causes recovery problems where e.g. an XAResource may be regarded as 'owned' by a tx and hence not rolledback, although we do not in fact have the parent tx log and hence will never commit it.
By patching the store implementation's allObjUids method we can render zero length files invisible to recovery scanning, thereby treating them as though no log exists, which is a more accurate representation of reality. This removes 'activation failed' log messages for the zero length records and allows orphaned subordinates to be correctly terminated by recovery. However, it also leads to subtle inconsistencies in the objectstore API, since the files remain visible to e.g. currentState. This is not currently believed to cause problems, but we add a config option to restore the old scanning behaviour just in case.
<properties depends="common" name="arjuna">
<property name="com.arjuna.ats.arjuna.objectstore.scanZeroLengthFiles" value="YES"/>
> Cope With Zero Length Transaction Log Files
> Key: JBTM-821
> URL: https://issues.jboss.org/browse/JBTM-821
> Project: JBoss Transaction Manager
> Issue Type: Task
> Security Level: Public(Everyone can see)
> Components: Recovery, Transaction Core
> Affects Versions: 4.6.1.CP07, 4.14.0
> Reporter: James Wilson
> Assignee: Jonathan Halliday
> Fix For: 4.15.0, 4.6.1.CP09
> In certain situations, zero-length transactions log files can be created. When this happens there are repeated warning messages in the logs. Can these be dealt with another way and/or possibly ignored?
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
More information about the jbossts-issues