Category: Indexing

Brent Ozar generates 999 indexes:

The CHARACTER_MAXIMUM_LENGTH <> -1 OR IS NULL stuff is because I don’t want to try to index NVARCHAR(MAX) fields, but I do want to index NVARCHAR(50) fields.

The ORDER BY is because I wanted to get a wide variety of leading fields. If I just ordered by c1.COLUMN_NAME, I wouldn’t get all of the first fields covered in the first 999 indexes. (And I’m not even using a large table.)

Sometimes I think I’ve worked on systems which used this script to build indexes.  But then I read the index names:  “dta.”  And it all makes sense.

Jeremiah Peschka looks at a scenario in which a heap might be superior to a clustered index:

In this case, we have to assume that Event IDs may be coming from anywhere and, as such, may not arrive in order. Even though we’re largely appending to the table, we may not be appending in a strict order. Using a clustered index to support the table isn’t the best option in this case – data will be inserted somewhat randomly. We’ll spend maintenance cycles defragmenting this data.

Another downside to this approach is that data is largely queried by Owner ID. These aren’t unique, and one Owner IDcould have many events or only a few events. To support our querying pattern we need to create a multi-column clustering key or create an index to support querying patterns.

This result is not intuitive to me, and I recommend reading the whole thing.

Tara Kizer looks at forwarded records on heaps:

Forwarded records are rows in a heap that have been moved from the original page to a new page, leaving behind a forwarding record pointer on the original page to point at the new page. This occurs when you update a column that increases the size of the column and can no longer fit on the page. UPDATEs can cause forwarded records if the updated data does not fit on the page. Forwarding pointers are used to keep track of where the data is.

The comments are also worth reading.  Except for the terrible puns.

Kendra Little has a great diagram showing which indexes are disk-based and which are memory-resident:

I was looking through some terms in SQL Server documentation the other day, thinking about what it’s like to learn about SQL Server’s indexes when you’re new to the field. I jotted down a note: B-tree = Rowstore = Disk Based

And then I realized that’s not quite right.

Not all disk based indexes are traditional clustered and nonclustered indexes. Columnstore indexes are also disk based. Updatable Columnstore indexes use special rowstore B-trees behind the scenes. And Books Online says “rowstore” also refers to Memory-Optimized tables.

If you’re new to indexing, this picture will save you some learning time.

Shane O’Neill answers one of my favorite interview questions:

I used to think that this would be a complex requirement, possibly requiring aTRIGGER or two to check the inserted value against whatever is already there; but there is a way to have this functionality and have it the way that SQL Server normally would enforce a uniqueness on a column; by using a UNIQUE INDEX.

In case you’re thinking…

“Oh, a unique index doesn’t check what’s already there, is that it?”

I’m afraid that’s not the case.

This is one of my favorite uses of filtered indexes:  “limited” uniqueness.  In other words, I’m okay with an unlimited number of NULL values but all non-NULL values need to be unique.

Derik Hammer looks at the power of clustered indexes:

The data in a clustered index is logically sorted but does not guarantee that it will be physically sorted. The physical sorting is simply a common misconception. In fact, the rows on a given page are not sorted even though all rows contained on that page will be appropriate to its place in the logical sort order. Also, the pages on disk are not guaranteed to be sorted by the logical key either.

The most likely time where you will have a clustered index that is physically sorted is immediately after an index rebuild operation. If you are trying to optimize for sequential reads, setting a fill factor to leave free space on your pages will help limit how often you have pages physically out of order at the expense of disk space.

Derik also discusses four qualities for a good clustered index.  My preferred acronym is NUSE (Narrow, Unique, Static, Ever-increasing); Derik uses slightly different terms.

SQLWayne has a script to help find unused indexes:

Here’s some code that can show you what indexes are unused or empty.  An empty index just means that there’s no data in that table right now, it may always be populated later, so I would not drop an empty index.  Besides, how much space would an empty index take?

For my personal preferences, I order the output by table then index name, also I put a u.* at the end of the select statement so the more interesting usage stat columns can be seen.

If an index truly is unused, it’s a waste of resources.  The problem is, sometimes you’ll think an index is unused but it’s really a vital part of month-end reporting or used for the CEO’s favorite dashboard.