Understanding Duplicate Matching

[emilyf]

Recently, at my organization a staff member started going through Duplicates using the dupe finder for individuals. A lot of the dupes listed were staff members and site admins who also have Drupal user accounts.

When the staff member used the merge feature on these kinds of people (with a drupal account), we ended up with loads of errors. Here is an example:

My contact id was '2', and my drupal user account is also '2'. At some point, I ended up with 2 other records that had same first, last, and email address (occurred because another staff member did an import ignoring dupes). Those contact ids were somewhere in the ballpark '1822' and '2018'.  So I came up in the list of duplicates. The staff member choose to 'merge'. Everything appeared fine to the staff member after that. But a day later, when I logged in and went to any CiviCRM screen or my Drupal user screen, I got a CiviCRM non-recoverable error.

Because I have reverted since then, I can't recall the exact error (if there is reason to, I can try to reproduce again on a dev site), but it was an issue with the civicrm_uf_match table. Something about duplicate entry keys; it was pretty similar to this issue: http://issues.civicrm.org/jira/browse/CRM-1816, except that I recall the wording of the particular error being more like "duplicate entries found" or something. In any case, when I looked at the civicrm_uf_match table, I noticed my primary record of contact id '2' was gone, and it was replaced by another record with one of the higher contact_ids.

This error was only showing up for me when I was logged in, and for other users who had the same thing happen.

Which brings me to the question of exactly how duplicate merging is handled, and especially when one of the records has a drupal id associated with it. This is the main wiki page I found associated with dupes, and it's pretty sparse: http://wiki.civicrm.org/confluence/display/CRMDOC/Find+and+Merge+Duplicate+Contacts. I'd really like to update it with more complete information once I understand the whole process. If someone can explain, fabulous!

So just wondering what you guys think of this. I can attempt to recreate it in full on a dev site if you need specifics (if you think it's warranted to pursue this). It seems to be related to dupes with Drupal accounts specifically that get merged.

Thanks for the input.
emily