Symptoms

• Users appear twice in the People Directory, in shares, and mentions
• Users have two USERINFO records with no 'prevver' value (ie two current records), eg:

  select * from content where contenttype like 'USERINFO' and prevver is null;
  

  USERINFO records that do not have an entry in prevver are the currently active records. This is a little counter-intuitive given the name of the field, however older records that are no longer active contain the id of the superseding record in prevver. If prevver is null, the record is considered active.

Steps to Reproduce

This may not be the only way to have this happen, but doing the following sequence will cause this to happen:

1. Create a user in an LDAP server, which you want to test with
2. Set up the LDAP server in Confluence
3. Synchronise
4. Log in as that user and possibly add some extra profile information such as a picture/description
5. Disable the LDAP directory
6. Rename that user in the LDAP server
7. Create a new LDAP directory exactly the same
8. Synchronise

Known Triggers

This has been confirmed to occur on upgrade to 5.1.3, and possibly earlier or later versions.

Case Study

• These 3 users were duplicated as part of the upgrade to 5.1.3 (some irrelevant columns have been removed for ease of reading):

  contentid	contenttype	version	creator	creationdate	lastmodifier	lastmoddate	prevver	content_status	username
  22217497	USERINFO	1	an.admin	10/05/2013	an.admin	10/05/2013		current	USER1
  24248321	USERINFO	1		7/06/2013		7/06/2013		current	user1
  917505	USERINFO	3	admin	12/03/2012	user2	14/12/2012		current	user2
  24248322	USERINFO	1		7/06/2013		7/06/2013		current	USER2
  1409789	USERINFO	1	USER3	29/03/2012	USER3	29/03/2012		current	USER3
  24248323	USERINFO	1		7/06/2013		7/06/2013		current	user3

NB: Note the casing above - the 1st and 3rd records were originally uppercase, the second was originally lowercase.

• All 3 duplicates were created on June 7th. This was in the logs:

  2013-06-07 19:40:39,244 INFO [main] [confluence.upgrade.upgradetask.UserIndexingUpgradeTask] doUpgrade Found 3 usernames that need PersonalInformation created.
  

• I originally suspected that the casing of the usernames may have changed, but the users are currently in the original casing in cwd_user (upperfor USER1 and USER3, and lower for user2).

• There was a warning regarding a case change for user2, going back as far as October 2012 (found in old support case) :

  2013-06-10 09:22:11,856 WARN [http-8080-40] [atlassian.crowd.directory.DelegatedAuthenticationDirectory] updateLdapUser remote username [ USER2 ] casing differs from local username [ user2 ]. User details will be kept updated, but the username cannot be updated
  

• All 3 users are LDAP user accounts. This is a Delegated Authentication directory, and not a synchronising Connector type.

Diagnosis - is this affecting you?

First, try a content reindex. If this doesn't help, run this query (on any version). If it returns any results, this bug is the cause of your duplicates.

SELECT c.username
      FROM CONTENT c
      WHERE
        contenttype LIKE 'USERINFO'
        AND prevver is null
          GROUP BY c.username
          HAVING count(c.username)>1;

If this query returns nothing, have a look at CONF-36018 as well.

Work Around

1. Stop Confluence
2. Make a full database and file system backup
3. Run this command to find the usernames of affected users - ensure these match what you are seeing in the People Directory before deleting

   // FOR CONFLUENCE 5.3 AND ABOVE
   SELECT m.username FROM content c
    JOIN user_mapping m ON c.username = m.user_key
     WHERE c.username IN
       (SELECT c.username
         FROM CONTENT c
         WHERE
           contenttype LIKE 'USERINFO'
           AND prevver is null
             GROUP BY c.username
             HAVING count(c.username)>1)
       GROUP BY m.username;
   
   // FOR CONFLUENCE 5.2 AND BELOW
   SELECT c.username FROM CONTENT c
     WHERE c.username IN
       (SELECT c.username
         FROM CONTENT c
         WHERE
           contenttype LIKE 'USERINFO'
           AND prevver is null
             GROUP BY c.username
             HAVING count(c.username)>1)
       GROUP BY c.username;
   

1. Run this command to delete the duplicate entries (all versions). The row with the higher contentid will be deleted.

   DELETE FROM CONTENT
   WHERE prevver IN
       (SELECT DISTINCT c1.contentid AS "pageid"
        FROM CONTENT c1
        JOIN CONTENT c2 ON c1.username = c2.username
        WHERE c1.contenttype = 'USERINFO'
        AND c2.contenttype = 'USERINFO'
        AND c1.prevver is null 
        AND c2.prevver is null 
        AND c1.contentid > c2.contentid);
   
   DELETE FROM CONTENT
   WHERE contentid IN
       (SELECT DISTINCT c1.contentid AS "pageid"
        FROM CONTENT c1
        JOIN CONTENT c2 ON c1.username = c2.username
        WHERE c1.contenttype = 'USERINFO'
        AND c2.contenttype = 'USERINFO'
        AND c1.prevver is null 
        AND c2.prevver is null 
        AND c1.contentid > c2.contentid);

1. Restart Confluence.
2. If you are still experiencing problems, have a look at CONF-36018 as well.

NB: If you have any trouble running the delete query because of foreign key constraints on the ATTACHMENT or CONTENT tables, do not remove the entries from that table. Instead, please get in touch with Support and we can help you remove the duplicates, and update the queries to be more robust.