Indexing is the most important tool you have for speeding up queries. There are other techniques available to you, too, but generally the one thing that will make the most difference is the proper use of indexes. On the MySQL mailing list, people often ask for help in making a query run faster. In a surprisingly large number of cases, there are no indexes on the tables in question, and adding indexes often solves the problem immediately. It doesn't always work like that, because optimization isn't always simple. Nevertheless, if you don't use indexes, in many cases you're just wasting your time trying to improve performance by other means. Use indexing first to get the biggest performance boost and then see what other techniques might be helpful.
This section describes what an index is and how indexing improves query performance. It also discusses the circumstances under which indexes might degrade performance and provides guidelines for choosing indexes for your table wisely. In the next section, we'll discuss MySQL's query optimizer. It's good to have some understanding of the optimizer in addition to knowing how to create indexes because then you'll be better able to take advantage of the indexes you create. Certain ways of writing queries actually prevent your indexes from being useful, and generally you'll want to avoid having that happen. (Not always, though. Sometimes you'll want to override the optimizer's behavior. We'll cover some of those cases, too.)
Let's consider how an index works by beginning with a table that has no indexes. An unindexed table is simply an unordered collection of rows. For example, Figure 4.1 shows the ad table that we first saw in Chapter 1, "Getting Started with MySQL and SQL." There are no indexes on this table, so to find the rows for a particular company, it's necessary to examine each row in the table to see if it matches the desired value. This involves a full table scan, which is slow as well as tremendously inefficient if the table is large but contains only a few records matching the search criteria.
Figure 4.2 shows the same table but with the addition of an index on the company_num column in the ad table. The index contains an entry for each row in the ad table, but the index entries are sorted by company_num value. Now, instead of searching through the table row by row looking for items that match, we can use the index. Suppose we're looking for all rows for company 13. We begin scanning the index and find three rows for that company. Then we reach the row for company 14, a value higher than the one we're looking for. Index values are sorted, so when we read the record containing 14, we know we won't find any more matches and can quit looking. Thus, one efficiency gained by using the index is that we can tell where the matching rows end and can skip the rest. Another efficiency is that there are positioning algorithms for finding the first matching entry without doing a linear scan from the start of the index (for example, a binary search is much quicker than a scan). That way, we can quickly position to the first matching value and save a lot of time in the search. Databases use various techniques for positioning to index values quickly, but it's not so important here what those techniques are. What's important is that they work and that indexing is a good thing.
You may be asking why we don't just sort the data file and dispense with the index file. Wouldn't that produce the same type of improvement in search speed? Yes, it would--if the table had a single index. But you might want to add a second index, and you can't sort the data file two different ways at once. (For example, you might want one index on customer names and another on customer ID numbers or phone numbers.) Using indexes as entities separate from the data file solves the problem and allows multiple indexes to be created. In addition, rows in the index are generally shorter than data rows. When you insert or delete new values, it's easier to move around shorter index values to maintain the sort order than to move around the longer data rows.
The example just described corresponds in general to the way MySQL indexes tables, although the particular details vary for different table types. For example, for a MyISAM or ISAM table, the table's data rows are kept in a data file, and index values are kept in an index file. You can have more than one index on a table; if you do, they're all stored in the same index file. Each index in the index file consists of a sorted array of key records that are used for fast access into the data file. By contrast, the BDB and InnoDB table handlers do not separate data rows and index values in the same way, although both maintain indexes as sets of sorted values. The BDB handler uses a single file per table to store both data and index values, and the InnoDB handler uses a single tablespace within which it manages data and index storage for all InnoDB tables.
The preceding discussion describes the benefit of an index in the context of single-table queries, where the use of an index speeds searches significantly by eliminating the need for full table scans. However, indexes are even more valuable when you're running queries involving joins on multiple tables. In a single-table query, the number of values you need to examine per column is the number of rows in the table. In a multiple-table query, the number of possible combinations skyrockets because it's the product of the number of rows in the tables.
Suppose you have three unindexed tables, t1, t2, and t3, each containing a column c1, c2, and c3, respectively, and each consisting of 1000 rows that contain the numbers 1 through 1000. A query to find all combinations of table rows in which the values are equal looks like this:
SELECT t1.c1, t2.c2, t3.c3 FROM t1, t2, t3 WHERE t1.c1 = t2.c2 AND t1.c1 = t3.c3;
The result of this query should be 1000 rows, each containing three equal values. If we process the query in the absence of indexes, we have no idea which rows contain which values. Consequently, we must try all combinations to find the ones that match the WHERE clause. The number of possible combinations is 1000x1000x1000 (1 billion!), which is a million times more than the number of matches. That's a lot of wasted effort, and this query is likely to be very slow, even for a database such as MySQL that is very fast. And that is with only 1000 rows per table. What happens when you have tables with millions of rows? As tables grow, the time to process joins on those tables grows even more if no indexes are used, leading to very poor performance. If we index each table, we can speed things up considerably because indexing allows the query to be processed as follows:
Select the first row from table t1 and see what value the row contains.
Using the index on table t2, go directly to the row that matches the value from t1. Similarly, using the index on table t3, go directly to the row that matches the value from t1.
Proceed to the next row of table t1 and repeat the preceding procedure until all rows in t1 have been examined.
In this case, we're still performing a full scan of table t1, but we're able to do indexed lookups on t2 and t3 to pull out rows from those tables directly. The query runs about a million times faster this way?literally. (This example is contrived for the purpose of making a point, of course. Nevertheless, the problems it illustrates are real, and adding indexes to tables that have none often results in dramatic performance gains.)
MySQL uses indexes as just described to speed up searches for rows matching terms of a WHERE clause or rows that match rows in other tables when performing joins. It also uses indexes to improve the performance of other types of operations:
For queries that use the MIN() or MAX() functions, the smallest or largest value in a column can be found quickly without examining every row if the column is indexed.
MySQL can often use indexes to perform sorting and grouping operations quickly for ORDER BY and GROUP BY clauses.
Sometimes MySQL can use an index to avoid reading data rows entirely. Suppose you're selecting values from an indexed numeric column in a MyISAM table and you're not selecting other columns from the table. In this case, by reading an index value from the index file, you've already got the value you'd get by reading the data file. There's no reason to read values twice, so the data file need not even be consulted.
In general, if MySQL can figure out how to use an index to process a query more quickly, it will. This means that, for the most part, if you don't index your tables, you're hurting yourself. You can see that I'm painting a rosy picture of the benefits of indexing. Are there disadvantages? Yes, there are. In practice, these drawbacks tend to be outweighed by the advantages, but you should know what they are.
First, an index takes up disk space, and multiple indexes take up correspondingly more space. This may cause you to reach a table size limit more quickly than if there are no indexes:
For ISAM and MyISAM tables, indexing a table heavily may cause the index file to reach its maximum size more quickly than the data file.
For BDB tables, which store data and index values together in the same file, adding indexes will certainly cause the table to reach the maximum file size more quickly.
InnoDB tables all share space within the InnoDB tablespace. Adding indexes depletes storage within the tablespace more quickly. However, as long as you have additional disk space, you can expand the tablespace by adding new components to it. (Unlike files used for ISAM, MyISAM, and BDB tables, the InnoDB tablespace is not bound by your operating system's file size limit, because it can comprise multiple files.)
Second, indexes speed up retrievals but slow down inserts and deletes as well as updates of values in indexed columns. That is, indexes slow down most operations involving writing. This occurs because writing a record requires writing not only the data row, it requires changes to any indexes as well. The more indexes a table has, the more changes need to be made, and the greater the average performance degradation. In the "Loading Data Efficiently" section later in this chapter, we'll go into more detail about this phenomenon and what you can do about it.
The syntax for creating indexes was covered in the "Creating and Dropping Indexes" section of Chapter 3, "MySQL SQL Syntax and Use." I assume here that you've read that section. But knowing syntax doesn't in itself help you determine how your tables should be indexed. That requires some thought about the way you use your tables. This section gives some guidelines on how to identify candidate columns for indexing and how best to set up indexes:
Index columns that you use for searching, sorting, or grouping, not columns you display as output. In other words, the best candidate columns for indexing are the columns that appear in your WHERE clause, columns named in join clauses, or columns that appear in ORDER BY or GROUP BY clauses. Columns that appear only in the output column list following the SELECT keyword are not good candidates:
SELECT col_a not a candidate FROM tbl1 LEFT JOIN tbl2 ON tbl1.col_b = tbl2.col_c candidates WHERE col_d = expr; a candidate
The columns that you display and the columns you use in the WHERE clause might be the same, of course. The point is that appearance of a column in the output column list is not in itself a good indicator that it should be indexed.
Columns that appear in join clauses or in expressions of the form col1 = col2 in WHERE clauses are especially good candidates for indexing. col_b and col_c in the query just shown are examples of this. If MySQL can optimize a query using joined columns, it cuts down the potential table-row combinations quite a bit by eliminating full table scans.
Use unique indexes. Consider the spread of values in a column. Indexes work best for columns with unique values and most poorly with columns that have many duplicate values. For example, if a column contains many different age values, an index will differentiate rows readily. An index probably will not help much for a column that is used to record sex and contains only the two values 'M' and 'F'. If the values occur about equally, you'll get about half of the rows whichever value you search for. Under these circumstances, the index may never be used at all because the query optimizer generally skips an index in favor of a full table scan if it determines that a value occurs in more than about 30 percent of a table's rows.
Index short values. If you're indexing a string column, specify a prefix length whenever it's reasonable to do so. For example, if you have a CHAR(200) column, don't index the entire column if most values are unique within the first 10 or 20 bytes. Indexing the first 10 or 20 bytes will save a lot of space in the index, and probably will make your queries faster as well. A smaller index involves less disk I/O, and shorter values can be compared more quickly. More importantly, with shorter key values, blocks in the index cache hold more key values, so MySQL can hold more keys in memory at once. This improves the likelihood of locating rows without reading additional index blocks from disk. (You want to use some common sense, of course. Indexing just the first character from a column isn't likely to be that helpful because there won't be very many distinct values in the index.)
Take advantage of leftmost prefixes. When you create an n-column composite index, you actually create n indexes that MySQL can use. A composite index serves as several indexes because any leftmost set of columns in the index can be used to match rows. Such a set is called a leftmost prefix. (This is different than indexing a prefix of a column, which is using the first n bytes of the column for index values.)
Suppose you have a table with a composite index on columns named state, city, and zip. Rows in the index are sorted in state/city/zip order, so they're automatically sorted in state/city order and in state order as well. This means that MySQL can take advantage of the index even if you specify only state values in a query or only state and city values. Thus, the index can be used to search the following combinations of columns:
state, city, zip state, city state
MySQL cannot use the index for searches that don't involve a leftmost prefix. For example, if you search by city or by zip, the index isn't used. If you're searching for a given state and a particular Zip code (columns 1 and 3 of the index), the index can't be used for the combination of values, although MySQL can narrow the search using the index to find rows that match the state.
Don't over-index. Don't index everything in sight based on the assumption "the more, the better." That's a mistake. Every additional index takes extra disk space and hurts performance of write operations, as has already been mentioned. Indexes must be updated and possibly reorganized when you modify the contents of your tables, and the more indexes you have, the longer this takes. If you have an index that is rarely or never used, you'll slow down table modifications unnecessarily. In addition, MySQL considers indexes when generating an execution plan for retrievals. Creating extra indexes creates more work for the query optimizer. It's also possible (if unlikely) that MySQL will fail to choose the best index to use when you have too many indexes. Maintaining only the indexes you need helps the query optimizer avoid making such mistakes.
If you're thinking about adding an index to a table that is already indexed, consider whether the index you're thinking about adding is a leftmost prefix of an existing multiple-column index. If so, don't bother adding the index because, in effect, you already have it. (For example, if you already have an index on state, city, and zip, there is no point in adding an index on state.)
Consider the type of comparisons you perform on a column. Generally, indexes are used for <, <=, =, >=, >, and BETWEEN operations. Indexes are also used for LIKE operations when the pattern has a literal prefix. If you use a column only for other kinds of operations, such as STRCMP(), there is no value in indexing it. For HEAP tables, indexes are hashed and are used only for equality comparisons. If you perform a range search (such as a < b) with a HEAP table, an index will not help.
Use the slow-query log to identify queries that may be performing badly. This log can help you find queries that may benefit from indexing. Use the mysqldumpslow utility to view this log. (See Chapter 11, "General MySQL Administration" for a discussion of MySQL's log files.) If a given query shows up over and over in the slow-query log, that's a clue that you've found a query that may not be written optimally. You may be able to rewrite it to make it run more quickly. Keep the following points in mind when assessing your slow-query log:
"Slow" is measured in real time, so more queries will show up in the slow-query log on a heavily loaded server than on a lightly loaded one. You'll need to take this into account.
If you use the --log-long-format option in addition to enabling slow-query logging, the log also will include queries that execute without using any index. These queries aren't necessarily slow. (No index may be needed for small tables, for example.)