I came across with the following scenario:
The system started to perform very poorly. I checked sys.dm_exec_requests and there were several UPDATE calls waiting on the resource KEY XX:YYYYYYYYYYYYY.
The key points to a non-clustered index on a table with over 28 million rows that is heavily used. Since we didn't want to put the system down, we decided to rebuild the index online. This action took aprox. 45 min and after that the system was back to normal speed for the next 40-50 min and then the same problem happened again. At this point, we decided to drop/create the index which fixed the problem.
I was wondering why the rebuilding index online was not enough to fix the problem. Any suggestions?