TDM merger fails due to infinite recursive loop

Regression of CONFSERVER-32164

Problem summary

When Collaborative editing is disabled, or in a version of Confluence below 6.0, under specific conditions, Confluence can run out of memory when a user goes to edit a Confluence page or view his drafts (Username > Drafts), following a sequence of events:

1. User A goes edit a page (Version 1)
2. Confluence auto-saves a draft based on Version 1, and user does not save the page
3. User B edits the same page and saves, promoting the page to Version 2
4. User A views his drafts (or edits the page again)
5. Confluence tries to merge the three versions (Version 1, Version 1 draft, and Version 2), and results in Confluence hanging

Stack trace:

2016-04-28 12:13:09,528 WARN [com.atlassian.confluence.plugins.merge.TDMMerger:thread-311] [confluence.plugins.merge.TDMMerger] mergeContentInternal Merge failed.
 -- url: /confluence/pages/editpage.action | referer: https://XXXX/confluence/display/SPACE/PAGE | userName: user | page: 1052053228 | action: editpage
java.lang.RuntimeException: ASSERTION FAILED: Merge.treeMerge(). Infinitive recursive loop.
        at tdm.lib.Merge.treeMerge(Merge.java:107)
        at tdm.lib.Merge.treeMerge(Merge.java:173)
        at tdm.lib.Merge.treeMerge(Merge.java:173)
        at tdm.lib.Merge.treeMerge(Merge.java:173)
        at tdm.lib.Merge.treeMerge(Merge.java:173)
        at tdm.lib.Merge.merge(Merge.java:82)
        at com.atlassian.confluence.plugins.merge.TDMMerger.mergeContentInternal(TDMMerger.java:99)
        at com.atlassian.confluence.plugins.merge.TDMMerger.access$000(TDMMerger.java:46)
        at com.atlassian.confluence.plugins.merge.TDMMerger$1.call(TDMMerger.java:120)
        at com.atlassian.confluence.plugins.merge.TDMMerger$1.call(TDMMerger.java:116)
        at com.atlassian.sal.core.executor.ThreadLocalDelegateCallable.call(ThreadLocalDelegateCallable.java:42)
        at java.util.concurrent.FutureTask.run(FutureTask.java:266)
        at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
        at java.lang.Thread.run(Thread.java:745)

Details:

• https://jira.atlassian.com/browse/CONF-32164?focusedCommentId=854595&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-854595
• https://jira.atlassian.com/browse/CONF-32164?focusedCommentId=904073&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-904073

Workaround

6.0 and later

Make sure that Collaborative Editing is enabled and working, as the affected merge code is not used unless it is disabled.

Warning:
The below workarounds involve deleting one or more drafts. Drafts contain edits that have not yet been saved, so important work can be lost by deleting them.

5.5 and later
If a user has been identified causing this problem, they can go to /users/viewmydrafts.action page and discard either all of their drafts or just the draft for the page that is known to trigger this problem.

All versions
Alternatively, all the drafts can be deleted directly from the DB:

1. Back up Confluence DB as a precaution
2. Run the following DELETE statements against the DB:

   DELETE FROM BODYCONTENT WHERE contentid IN (SELECT contentid FROM CONTENT WHERE contenttype = 'DRAFT');
   DELETE FROM CONTENT WHERE contenttype = 'DRAFT';
   

1. Confluence Admin > Cache Statistics > Flush all (down at the bottom)