Query Processing

Read in Elmasri & Navathe (EN)


How should a database organize the execution of a query?
How can we identify slow queries?
How can we make them run faster?



In this section we will assume that multiple records are stored per disk block, and that the goal of query optimization is to minimize the number of block reads.

Ideally this material would be read after the files_and_indexes.html material, but this is not essential. A B-Tree index is a very "fat" tree that is ordered by the key value in question. The ordering means that a B-Tree can be used either for single-value queries, e.ssn = '111223333', or range queries: e.ssn between '111223333' and '111229999'.

To begin, note it is relatively easy to sort records. Typically we will use merge-sort, where we subdivide into blocks small enough to fit into memory. We sort those with, say, quicksort. We then make passes until each pair of consecutive blocks is merged.

Select

Some examples
  1. select * from employee e where e.ssn = '123456789'
  2. select * from department where dnumber >=5
  3. select * from employee where dno = 5
  4. select * from employee where dno = 5 and salary > 30000 and sex = 'F'
  5. select * from works_on where essn='123456789' and pno=10
Methods of implementation:

Linear search: we read each disk block once. This corresponds to a search for e in employee ....
Binary search: this is an option if the selection condition is an equality test on a key attribute, and the file is ordered by that attribute.
Primary index: same requirement as above, plus we need an index
Hash search: equality comparison on any attribute for which we have a hash index (need not be a key attribute)
Primary index and ordered comparison: use the index to find all departments with dnumber > 5
Clustered index (includes hash indexes)
B-tree index

If we are searching for "dno=5 AND salary=4000" we can use an index for either attribute, or a composite index if one exists.

In general, we would like to start with the condition that leads to the fewest results, and then continue with the remaining conditions.

For OR conditions, we just have to search for each.

For the examples above, here are some notes on which select implementation is likely to be most efficient.
  1. For this query, an index on employee.ssn is likely to be fastest. Any index will do.
  2. For this query, the range comparison (dnumber >= 5) means that unless we have an "ordered" index (that is, a B-tree or a "primary index") we might find linear search hard to beat. Note that this is a search by key attribute.
  3. For this query, an index on dno would be a clustering index as in figs 18.2, 18.3 and 18.5. This index could be used, and if the employee file were ordered by dno then the minimum number of blocks would need to be retrieved. But this is unlikely; and that in turn means that using the index might still involve retrieving nearly ever block of the employee file (for example, if almost every block contained at least one member of department 5). Linear search in such a case would be hard to beat.
  4. In this case we have three selection criteria. We are best off applying the most selective criterion first. Unless we know something about the distribution of salaries, we can assume that "salary>30000" is about 50% likely to be true; the same might be true for "sex = 'F'". We are best off starting with dno=5, and then searching only those records for the ones that also satisfy the other two criteria.
  5. Note that the primary key to works_on is the pair (essn,pno). If we have a primary-key index, we can simply look this up. But if we keep the file ordered by essn, and with an index on essn, then we can also use that to find the answer; most employees work on only a small number of projects so works_on.lookup(essn) will be small.

Scans

Here are three types of scans:

1. Heap scan (or sequential scan): we go through the table a block at a time

2. Index scan: we go through the table via accessing the index, a record at a time. Normally, going through record by record is inefficient, but we may not actually need to access anything besides the index. Often, we can eliminate many records based only on information available in the index.

In some cases the scan requires only the information found in the index; we don't have to retrieve data blocks from the table at all. In this case, the scan is an index-only scan.

3. Bitmap heap scan: we go through the index, but then sort the record addresses by block, and then go through the records block-by-block. This resembles a heap scan, but we can do the following:


Joins

Mostly we focus on equijoins:

  1. select * from employee e join department d on e.dno = d.dnumber
  2. select * from department d join employee e on d.mgr_ssn = e.ssn
In each case we are joining a key in one table (underlined) to a foreign key in another. However, it is likely that the employee table is much larger than the department table. In the first example, the number of records in the join will be (about) the size of the employee table (that is, large); in the second example the number of records in the join will be (about) the size of the department table, much smaller. Thus, while both are joins of the employee and department tables, the joins are in fact quite different.

We will consider four methods of implementation.

Numeric examples for the four methods (EN6 p 690-693): Suppose we have 6,000 employee records in 2,000 blocks, and 50 departments in 10 blocks. We have indexes on both tables. The employee.ssn index takes 4 accesses, and the dept.dnumber index takes 2.

1. Nested-loop join: this is where we think of a for-loop:

    for (e in employee) {
       for (d in department) {
          if (e.dno = d.dnumber) print(e,d);
       }
    }

The time needed here is NM, where N is the size of employee and M is the size of department; loosely speaking, we might say this is quadratic. If we go through the employee table record by record, that amounts to 2,000 block accesses. For each employee record the loop above would go through all 10 blocks of departments; that's 6,000 × 10 = 60,000 blocks. Doing it the other way, we go through 50 department records, and, for each department record, we search 2,000 blocks of employees for 50×2,000 = 100,000 block accesses. In other words, the complexity is not symmetric!

However, note that we can also do this join block-by-block on both files (sometimes called the block-nested-loop approach):
Done this way, the number of block accesses is 2,000 × 10 = 20,000 blocks, or, in general, NM, where N and M are now the sizes of the respective tables in blocks.

Realistically, doing operations block-by-block instead of record-by-record is very important!

Performance improves rapidly if we can keep the smaller table entirely in memory: we then need only 2,010 block accesses! (The EN6 analysis makes some additional assumptions about having spare buffer blocks to work with.)

All joins in MySQL (5.7) are considered to be nested-loop joins, although MySQL considers index joins to be a special case of nested-loop joins. MySQL does not implement hash or sort-merge joins. Block-nested-loop joins are used where possible.

2. Index join: if we have an index on one of the attributes, we can use it to find the joined records. Here is an example using the index on d.dnumber; it might be used to implement the query select * from department d join employee e on d.dnumber = e.dno.

    for (e in employee) {
       d = lookup(department, e.dno);
       if (d != null) print(e,d);
    }

Note that any index will do, but that this may involve retrieving several disk blocks for each e and will almost certainly involve retrieving at least one disk block (from department) for every e in employee. It may or may not be better than Method 1.

Consider the first query above. Suppose we have a primary index on department.dno that allows us to retrieve a given department in 2 accesses. Naively, we go through 6,000 employees and retrieve the department of each; that's 6,000×2 = 12,000 block accesses. There is no obvious way to take advantage of the grouping of employees into blocks (why?).

Now consider the second query, and suppose we can find a given employee in 4 accesses. Then we go through 50 × 4 = 200 block accesses (for every department d, we look up d.mgr_ssn in table employee). Of course, as we noted at the beginning, the final result of the second query is a much smaller table.


3. Sort-merge join: we sort both files on the attribute in question, and then do a join-merge. This takes a single linear pass, of size the number of blocks in the two files put together file. This is most efficient if the files are already sorted, but note that it's still faster than 1 (and possibly faster than 2) if we have to sort the files. Assume both tables are sorted by their primary key, and assume we can sort in with N log(N) block accesses, where N is the number of blocks in the file. Then query 1 requires us to sort table employee in time 2,000×11 = 22,000; the actual merge time is much smaller. Query 2 requires us to sort table department in time 10×4 = 40; the merge then takes ~2,000 blocks for the employee table.

4. Hash (or partition-hash) join: Let the relations (record sets) be R and S. We partition R into disjoint sets Ri = {r ∈ R | hash(r) = i}, and S into Si = {s∈S | hash(s) = i}. Now we note that the join R ⋈ S is simply the disjoint union of the Ri ⋈ Si. In most cases, either Ri or Si will be small enough to fit in memory.

If R is small enough that we can keep all the Ri in memory, the process is even simpler: we create the Ri, and then make a single scan through S. For each record of s, we use the hash function to figure out which Ri we need to search, and then search that Ri for a match.

A good hash function helps; we want the buckets to be relatively uniform. We won't get uniform distribution, but we should hope for Poisson distribution. With N things in K buckets, we expect an average of a=N/K per bucket. The probability of i things is aie-a/i! and the expected number of i-sized buckets is about N*aie-a/i!.

An important feature of hash joins is that an index is not necessary.


The join selection factor is the fraction of records that will participate in the join. In query 2 above,

    select * from department d, employee e where d.mgr_ssn = e.ssn

all departments will be involved in the join, but almost no employees. So we'd rather go through the departments, looking up managing employees, rather than the other way around.

In the most common case of joins, the join field is a foreign key in one file and a primary key in the other. Suppose we keep all files sorted by their primary key. Then for any join of this type, we can traverse the primary-key file block by block; for each primary-key value we need to do a lookup of the FK attribute in the other file. This would be method 2 above; note that we're making no use of the primary index.


Query optimization

As with compilers, the word "optimization" is used loosely. However, SQL statements can usually be implemented a number of ways.

A major rule is to apply select/project before joins, and in general do operations first that have greatest potential for reducing the size of the number of records. But if projection makes an index unavailable, and the index would speed up the join, do it later (that is, use the index).

Query tree: See fig 19.4.

Select lname from employee e, works_on w, project p
where e.ssn = w.essn and w.pno = p.pnumber and p.pname = 'Aquarius';

See fig 19.5, but note that doing cross-products was never on the agenda. However, it still does help to do the second join first.

Some heuristic rules:


For information on how MySQL organizes queries, including the explain command, see admin1.html#explain.


Selecting Indexes

Much of this information comes from www.slideshare.net/myxplain/mysql-indexing-best-practices-for-mysql and www.slideshare.net/billkarwin/how-to-design-indexes-really.

The first step is identifying which queries are underperforming. If a query is slow, but you don't run it often, it doesn't matter. The MySQL slow-query log (/var/log/mysql/mysql-slow.log) can be useful here.

If a query is not running fast enough, options available include

The straight_join option (above) is often useful for the first category, as can be the addition of a WITH INDEX index_name option. Most query-optimizing efforts, however, are in the last category: adding indexes.

Be aware that indexes add write costs for all updates; don't create more than you need.

A B-Tree index can also be used to look up any prefix of the key. The prefix might be a single attribute A in an index on trow attributes (A,B), or else a string prefix: e.ssn LIKE '55566%'.

B-tree indexes can also be used for range queries: key < 100, key BETWEEN 50 AND 80, for example.

Join comparisons do best when the fields being compared are both strings and are both the same length. Joining on ssn where one table declares ssn to be an Integer and the other a CHAR(9) is not likely to work well.

A non-key index has high selectivity if each index value corresponds to only a few records. An index on lname might qualify. The standard example of a low-selectivity index is one on sex.

Many MySQL indexes on a non-key field A are really dual-attribute indexes on (A,PK) where PK is the primary key.

In a multi-attribute index, MySQL can use attributes up until the first one not involved in an equality comparison (that is, not used or involved in an inequality). If the index is on attributes (A,B,C), then the index can be used to answer:
    A>10
    A=10 and B between 30 and 50
    A=12 and B=40 and C<> 200

The telephone book is an index on (lname, fname). It can find people with lname='dordal' quickly, but is not very useful in finding people with fname='peter'.

Sometimes a query can be answered by using ONLY the index. In that case, the index is said to be a covering index for that query. For example, if we have an index for EMPLOYEE on (dno,ssn), then we can answer the following query using the index alone:

    select e.ssn from employee e where e.dno = 5

The (dno,ssn) index is a covering index for this query.

For the phone book, an index on (lname, fname, phone) would be a covering index for phone-number lookups, but not, say, for address lookups.

For the works_on table, an index on (essn,pno,hours) ((essn,pno) is the primary key) is a covering index for the whole table.

If the query is 

select e.lname from employee e where e.salary = 30000 and e.dno = 5;

then while the ideal index is a common index on either (salary,dno) or (dno,salary), two separate indexes on (salary) alone and on (dno) alone are almost as useful, as we can look up the employees with salary = 30000 and the employees in dno 5 and then intersect the two sets. The strategy here is sometimes called "index merge".

Sometimes, if the key is a long string (eg varchar(30)), we build the index on just a prefix of the string (maybe the first 10 characters). The prefix length should be big enough that it is "selective": there should be few records corresponding to each prefix. It should also be short enough to be fast. If a publisher database has an index for book_title, of max length 100 characters, what is a good prefix length? 10 chars may be too short: "An Introdu" matches a lot of books. How about 20 characters? "An Introduction to C" is only marginally more specific.