SQL Problems Requiring Cursors
From Seo Wiki - Search Engine Optimization and Programming Languages
|This article may need to be wikified to meet Wikipedia's quality standards. Please help by adding relevant internal links, or by improving the article's layout. (June 2008)|
|This article is an orphan, as few or no other articles link to it. Please introduce links to this page from related articles; suggestions are available. (June 2008)|
| This article does not cite any references or sources.
Please help improve this article by adding citations to reliable sources. Unsourced material may be challenged and removed. (June 2008)
The trouble with cursor processing is that the row-by-row nature cannot easily be made to occur in parallel by database optimizers. This results in a failure of the cursor processing to take full advantage of the processing power available on most database platforms.
To overcome this problem, cursor logic can often be converted into set-based SQL queries. Set-based SQL uses the syntax "SELECT...FROM..." with various clauses to perform the data manipulation work. Database optimizers have little trouble dividing such queries into parallel threads, thus fully utilizing the database hardware.
The resulting performance gain can be literal orders of magnitude: hours become minutes, and minutes become seconds.
Despite the tremendous performce gains from set-based SQL, it can be challenging to find the SQL that replaces fairly straigh-forward cursors.
This article discusses examples of such conversions.
In this article, the following constraints apply:
- The term "cursor" includes all cursor-like behavior. For example, using a loop in a shell script that loops across SQL queries or the output of SQL queries is cursor-like behavior and does not achieve the goal of true set-based processing within the database.
- All set-based SQL must be ANSI SQL. A number of vendors provide some extremely powerful proprietary extensions. The goal is to avoid such extensions in favor of ANSI SQL.
- The solution must be generalizable. In one or more examples below, specific values may be used for demonstration purposes, but any solution must scale to any number feasible within the power of the database software and machine resources.
Example: Insert rows based on a count in the table itself
The table below represents marbles. The four text columns represent four characteristics of marbles. Each characteristic has two values for a total of 16 types of marbles.
The "quantity" column represents how many of that marble we have. But instead of having a quantity, we want to have one row for each marble.
Thus, the target table would have the four text columns, and a total of 40 + 20 + 20 + 10 + ... + 10 + 5 = 270 rows.Source table:
QUANTITY TEXTURE APPEARANCE SHAPE COLOR ---------- ---------- ---------- ---------- ----- 40 smooth shiny round blue 20 smooth shiny warped blue 20 smooth dull round blue 10 smooth dull warped blue 20 rough shiny round blue 10 rough shiny warped blue 10 rough dull round blue 5 rough dull warped blue 40 rough dull warped red 20 rough dull round red 20 rough shiny warped red 10 rough shiny round red 20 smooth dull warped red 10 smooth dull round red 10 smooth shiny warped red 5 smooth shiny round redTable to generate:
TEXTURE APPEARANCE SHAPE COLOR ---------- ---------- ---------- ----- smooth shiny round blue -- 1 smooth shiny round blue -- 2 ... -- and so on smooth shiny round blue -- 40 smooth shiny warped blue -- 1 smooth shiny warped blue -- 2 ... -- and so on smooth shiny warped blue -- 20 ... -- and so on smooth shiny round red -- 1 smooth shiny round red -- 2 smooth shiny round red -- 3 smooth shiny round red -- 4 smooth shiny round red -- 5
Solution in cursor form
Generating the target table with a cursor is fairly simple.
cursor c is select * from marbles_seed;
for r in c loop for i in 1..r.quantity loop
r.accuracy, r.coverage ); end loop; end loop;
Solution in ANSI set-based SQL
Solving the problem with ANSI SQL is a bit more code, but requires a bit more creative thought than the nested loop approach of cursors.
To reach the solution requires an intermediate table. The table has one column of type NUMBER that has the values 0 to whatever number of rows is needed. For this discussion, we'll limit it to one million rows. The code is as follows:
Assume the source table above is named marbles_seed and the target table is named marbles. The code that generates the needed 270 rows is:
insert into marbles select m.texture, m.appearance, m.shape, m.color_actual, m.hits, m.color_predicted, m.accuracy, m.coverage from marbles_seed m, numbers n where m.quantity > n.n