It doesn’t happen often but every once in a while you may be the lucky person to find a previously unknown bug in SQL Server.
It was a normal morning for me, checking the status of our servers going over any failure messages waiting for the day to ramp up. That’s when one of our lead developers came around the corner and told me he had an error when he had tried to create an index on a table he was working on. The more he tried to explain the error the more I started to worry. I had him send me the code and the error statement.
Location: BtreeMgr.cpp:5372
Expression: bufferLen > currOffset + ACCESSSOR_OVERHEAD
SPID: 116
Process ID: 5016
Msg 3624, Level 20, State 1, Line 2
A system assertion check has failed. Check the SQL Server error log for details. Typically, an assertion failure is caused by a software bug or data corruption. To check for database corruption, consider running DBCC CHECKDB. If you agreed to send dumps to Microsoft during setup, a mini dump will be sent to Microsoft. An update might be available from Microsoft in the latest Service Pack or in a QFE from Technical Support.
Msg 0, Level 20, State 0, Line 0
A severe error occurred on the current command. The results, if any, should be discarded.
I had what we like to call in the high availability space a “pucker moment”. This wasn’t your normal, I typed something wrong and got an error, kind of problem. This was a real SEV 20 with an assert, the core engine had just puked on something it shouldn’t have.
Like all good DBA’s the first thing I did was run a DBCC on the database this error was generated from.
While that was going on I asked my very good friend, Google, if he had seen this particular assert before. For the first time in a very long time Google failed me! In the last few years if I hit this kind of hard error someone else has too and it is ether being fixed in a hot fix or addressed in the next version of SQL Server, but not this time.
So, we have this same schema on another server and the developer tried the exact same code there and had the exact same error.
I had him document the steps he took to get to this point and to his credit the steps were clear, concise and easily reproducible.
The DBCC finished with zero problems detected, which let me calm down a bit. That coupled with the fact it looked like I had a repeatable test case When the second database had cleared the DBCC I set about my task of reproducing the error and trying to find a work around. Lucky for us it was a simple matter of column organization in the index and we were able to apply it successfully and carry on with life.
I bundled up the documentation I had accumulated, ran the test case confirmed the bug and sent it off to the powers that be at Microsoft. Since we had a work around and it wasn’t a show stopper I didn’t raise it as a critical server down issue but Microsoft still worked it in a timely fashion.
So, what was the problem you say? It was an interesting edge condition.
We have a table that contains a composite primary key and the rest is made up of bit fields, a flag table.
We had to add a new column, another bit flag, to the table.
The non-clustered covering index was dropped the column was added to the end of the table.
The index was updated with the new column at the end of the column list and then *POOF* it blew up.
I think it has to do with two specific things.
First, bit fields are stored in a compact manor where multiple bits share a byte and aren’t truly separate from every other but field. It would be a huge waste of space to store each bit in it’s own byte but it would make things like this index issue less likely to happen.
Secondly we did a column add but didn’t drop and recreate the table repopulating it in the process so things at the page level weren’t nice and neat. The underlying clustered index wasn’t effected but when we tried to add an index back with the new field it couldn’t do it. The fix was simple, change the column order in the non-clustered index moving the new column up one. We verified the data without the index was correct and with the index was correct.
I haven’t tried it yet, but I am betting included columns won’t suffer the assert ether since the items don’t have to be sorted in the index.
So there you go! Having been on the software side of things a couple of times I always find it interesting when I find bugs in others products and work the issue to conclusion.
What is your take away from all of this? Never be afraid to submit a bug report to Microsoft. I have heard people say to the effect someone else will or has hit the bug and they will submit it. DON’T RELY ON THE ACTIONS OF OTHERS! Reporting bugs helps the community as a whole and makes the product better. When you assume someone else is worked it you are putting YOUR production servers in the hands of strangers. If someone has submitted it and it didn’t turn up in a search they will let you know, and be very kind about it to boot. You will get piece of mind that it is being worked on and it is a bug, or that you may keep someone else from stumbling onto this and not having the knowledge to fix it or work around it.
Wes