This is a quick blog post to let you know about a bug in SQL Server 2012 SP1 that can cause data loss when performing index maintenance.
The data loss issue can happen in some circumstances when you do a parallel online rebuild of a clustered index while there are concurrent data modifications happening on the table AND you also hit a deadlock and another error. Nasty when it occurs, but that should hopefully be a rare combination.
The workaround is to limit the online rebuild operation to be single threaded using the WITH (MAXDOP = 1) option.
There is a hotfix available – see KB 2969896 for more details.
Depending on which build you are on of SQL Server 2012 or 2014, the best option for you will vary. See my friend Aaron Bertrand’s post for comprehensive details.