[
https://issues.apache.org/jira/browse/AMQ-3982?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Gary Tully updated AMQ-3982:
----------------------------
Comment: was deleted
(was: a variant of the concurrent add iterate test that fails.
Syncing all updates, seeks and writes in pagefile should resolve, but there is
somehting more:
{code}
synchronized (this) {
readFile.seek(toOffset(pageId))
readFile.readFully(data);
}
...
synchronized (this) {
writeFile.seek(toOffset(w.page.getPageId()));
writeFile.write(w.getDiskBound(), 0, pageSize);
}{code})
> Overlapping PList iterators can read wrong data or throw exceptions about
> chunk streams not existing.
> -----------------------------------------------------------------------------------------------------
>
> Key: AMQ-3982
> URL: https://issues.apache.org/jira/browse/AMQ-3982
> Project: ActiveMQ
> Issue Type: Bug
> Components: Broker
> Affects Versions: 5.6.0
> Reporter: Timothy Bish
> Assignee: Timothy Bish
> Priority: Critical
> Fix For: 5.7.0
>
> Attachments: new.txt
>
>
> When multiple PList iterators are in use and happen have an overlapping read
> of PageFile data on disk they can reset each others seek offset and cause an
> exception to be thrown about "chunk stream does not exist" or read of a page
> in the Free list. The reason is that the hasNext of PList iterator needs to
> by sync'd on the index lock to prevent two iterators from entering into a
> readPage call concurrently.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
