Firebird SQL best practices

Firebird SQL best practices

Review of some SQL features available and that people often forget about

Author: Philippe Makowski IBPhoenix

Email: [email protected]
Licence: Public Documentation License
Date: 2016-09-29

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices

Common table expression

Syntax

WITH [RECURSIVE] -- new keywords

CTE_A -- first table expression’s name
[(a1, a2, ...)] -- fields aliases, optional
AS ( SELECT ... ), -- table expression’s definition
CTE_B -- second table expression
[(b1, b2, ...)]
AS ( SELECT ... ),
...
SELECT ... -- main query, used both
FROM CTE_A, CTE_B, -- table expressions
TAB1, TAB2 -- and regular tables
WHERE ...

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices

Emulate loose index scan

The term "loose indexscan" is used in some other databases for the operation of using
a btree index to retrieve the distinct values of a column efficiently; rather than scanning
all equal values of a key, as soon as a new value is found, restart the search by
looking for a larger value. This is much faster when the index has many equal keys.
A table with 10,000,000 rows, and only 3 differents values in row.

CREATE TABLE HASH

(
ID INTEGER NOT NULL,
SMALLDISTINCT SMALLINT,
PRIMARY KEY (ID)
);
CREATE ASC INDEX SMALLDISTINCT_IDX ON HASH (SMALLDISTINCT);

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices
Without CTE :

SELECT DISTINCT SMALLDISTINCT FROM HASH

SMALLDISTINCT
=============
0
1
2

PLAN SORT ((HASH NATURAL))

Prepared in 0.001 sec, processed in 13.234 sec
HASH 10000000 Non-Indexed reads

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices
Emulate loose index scan with recursive CTE :

WITH RECURSIVE
t AS (SELECT min(smalldistinct) AS smalldistinct FROM HASH
UNION ALL
SELECT (SELECT min(smalldistinct) FROM HASH
WHERE smalldistinct > t.smalldistinct)
FROM t WHERE t.smalldistinct IS NOT NULL)
SELECT smalldistinct FROM t WHERE smalldistinct IS NOT NULL
UNION ALL
SELECT NULL FROM RDB$DATABASE
WHERE EXISTS(SELECT 1 FROM HASH WHERE smalldistinct IS NULL)

PLAN (T HASH ORDER SMALLDISTINCT_IDX INDEX (SMALLDISTINCT_IDX))

PLAN (HASH INDEX (SMALLDISTINCT_IDX))
Prepared in 0.001 sec, processed in 3.312 sec
HASH 3 Indexed reads
RDB$DATABASE 1 Non-Indexed read

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices

MERGE
The purpose of MERGE is to read data from the source and INSERT, UPDATE or
DELETE in the target table according to a condition.
The source may be table, a view or "anithing you can select from" in general. Each
source record will be used to update, or delete one or more target record, insert a new
record in the target table, or neither.

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices

Example for MERGE

create table stock ( item_id int not null primary key, balance int);
insert into stock values (10, 2200);
insert into stock values (20, 1900);
commit;
select * from stock;

ITEM_ID BALANCE
========== ============
10 2200
20 1900

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices

create table buy ( item_id int not null primary key, volume int);
insert into buy values (10, 1000);
insert into buy values (30, 300);
commit;
select * from buy;

ITEM_ID VOLUME
========== ============
10 1000
30 300

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices

create table sale ( item_id int not null primary key, volume int);
insert into sale values (10, 2200);
insert into sale values (20, 1000);
commit;
select * from sale;

ITEM_ID VOLUME
========== ============
10 2200
20 1000

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices
Update the stock with what we bought.

select * from stock;

ITEM_ID BALANCE
========== ============
10 2200
20 1900

MERGE INTO stock USING buy ON stock.item_id = buy.item_id

WHEN MATCHED THEN UPDATE SET balance = balance + buy.volume
WHEN NOT MATCHED THEN INSERT VALUES (buy.item_id, buy.volume);
SELECT * FROM stock ORDER BY item_id;
ITEM_ID BALANCE
========== ============
10 3200
20 1900
30 300

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices
Then update the stock with what we sale.

SELECT * FROM stock ORDER BY item_id;

ITEM_ID BALANCE
========== ============
10 3200
20 1900
30 300

MERGE INTO stock USING sale ON stock.item_id = sale.item_id

WHEN MATCHED AND balance - volume > 0 THEN UPDATE SET balance = balance - volume
WHEN MATCHED THEN DELETE;
SELECT * FROM stock ORDER BY item_id;
ITEM_ID BALANCE
========== ============
10 1000
20 900
30 300

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices
See the DELETE in action :

rollback;
SELECT * FROM stock ORDER BY item_id;
ITEM_ID BALANCE
========== ============
10 2200
20 1900

select * from sale;

ITEM_ID VOLUME
========== ============
10 2200
20 1000

MERGE INTO stock USING sale ON stock.item_id = sale.item_id

WHEN MATCHED AND balance - volume > 0 THEN UPDATE SET balance = balance - volume
WHEN MATCHED THEN DELETE;
SELECT * FROM stock ORDER BY item_id;
ITEM_ID BALANCE
========== ============
20 900

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices

What are Windowing Functions?

• Similar to classical aggregates but does more!
• Provides access to set of rows from the current row
• Introduced SQL:2003 and more detail in SQL:2008
• Supported by PostgreSQL, Oracle, SQL Server, Sybase and DB2
• Used in OLAP mainly but also useful in OLTP

• Analysis and reporting by rankings, cumulative aggregates

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices

Windowed Table Functions

• Windowed table function

• operates on a window of a table

• returns a value for every row in that window
• the value is calculated by taking into consideration values from the set of
rows in that window
• 8 new windowed table functions
• In addition, old aggregate functions can also be used as windowed table
functions
• Allows calculation of moving and cumulative aggregate values.

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices

A Window
• Represents set of rows that is used to compute additionnal attributes
• Based on three main concepts
• partition
• specified by PARTITION BY clause in OVER()
• Allows to subdivide the table, much like GROUP BY clause
• Without a PARTITION BY clause, the whole table is in a single partition
• order
• defines an order with a partition
• may contain multiple order items
• Each item includes a value-expression
• NULLS FIRST/LAST defines ordering semantics for NULL
• this clause is independant of the query's ORDER BY clause

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices

• frame (Firebird don't implement frame yet)

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices

Built-in Windowing Functions

• RANK () OVER ...
• DENSE_RANK () OVER ...
• LAG () OVER ...
• LEAD () OVER ...
• ROW_NUMBER () OVER ...
• FIRST_VALUE () OVER ...
• LAST_VALUES () OVER ...
• NTH_VALUE () OVER ...

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices

Set Functions as Window Functions

Who are the highest paid relatively compared with the department average?

select emp_no, dept_no, salary,

avg(salary) over (partition by dept_no) as dept_avg,
salary - avg(salary) over (partition by dept_no) as diff
from employee
order by diff desc;
EMP_NO DEPT_NO SALARY DEPT_AVG DIFF
======= ======= ===================== ===================== =====================
118 115 7480000.00 6740000.00 740000.00
105 000 212850.00 133321.50 79528.50
107 670 111262.50 71268.75 39993.75
2 600 105900.00 66450.00 39450.00
85 100 111262.50 77631.25 33631.25
4 621 97500.00 69184.87 28315.13
46 900 116100.00 92791.31 23308.69
9 622 75060.00 53409.16 21650.84

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices

Performance
List orders, quantity ordered and cumulative quantity ordered by day

ORDER_DATE PO_NUMBER QTY_ORDERED QTY_CUMUL_DAY

=========== ========= ============ =====================
1991-03-04 V91E0210 10 10
1992-07-26 V92J1003 15 15
1992-10-15 V92E0340 7 7
1992-10-15 V92F3004 3 10
1993-02-03 V9333005 2 2
1993-03-22 V93C0120 1 1
1993-04-27 V9333006 5 5
1993-08-01 V93H3009 3 3
1993-08-09 V9324200 1000 1000
1993-08-09 V93C0990 40 1040

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices
Without window function

SELECT ORDER_DATE, CUST_NO,QTY_ORDERED,

(SELECT SUM(QTY_ORDERED)
FROM SALES AS Si
WHERE Si.ORDER_DATE = S.ORDER_DATE
AND Si.CUST_NO <= S.CUST_NO)
AS QTY_CUMUL_DAY
FROM SALES AS S
ORDER BY S.ORDER_DATE, S.CUST_NO

PLAN (SI INDEX (RDB$FOREIGN25))

PLAN SORT (S NATURAL)
SALES 591 indexed reads
SALES 33 non indexed reads

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices
With window function

SELECT ORDER_DATE, PO_NUMBER,QTY_ORDERED,

SUM(QTY_ORDERED)
OVER (PARTITION BY ORDER_DATE
ORDER BY PO_NUMBER)
AS QTY_CUMUL_DAY
FROM SALES
ORDER BY ORDER_DATE, PO_NUMBER

PLAN SORT (SALES NATURAL)

SALES 33 non indexed reads

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices
And you can extend it nearly without cost

SELECT ORDER_DATE, PO_NUMBER,QTY_ORDERED,

SUM(QTY_ORDERED)
OVER (PARTITION BY ORDER_DATE
ORDER BY PO_NUMBER)
AS QTY_CUMUL_DAY,
SUM(QTY_ORDERED)
OVER (PARTITION BY EXTRACT(YEAR FROM ORDER_DATE),EXTRACT(MONTH FROM ORDER_DATE)
ORDER BY ORDER_DATE, PO_NUMBER)
AS QTY_CUMUL_MONTH,
SUM(QTY_ORDERED)
OVER (PARTITION BY EXTRACT(YEAR FROM ORDER_DATE)
ORDER BY ORDER_DATE, PO_NUMBER)
AS QTY_CUMUL_YEAR
FROM SALES
ORDER BY ORDER_DATE, PO_NUMBER

PLAN SORT (SALES NATURAL)

SALES 33 non indexed reads

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices

ORDER_DATE PO_NUMBER QTY_ORDERED QTY_CUMUL_DAY QTY_CUMUL_MONTH QTY_CUMUL_YEAR

=========== ========= =========== ============= =============== ==============
1991-03-04 V91E0210 10 10 10 10
1992-07-26 V92J1003 15 15 15 15
1992-10-15 V92E0340 7 7 7 22
1992-10-15 V92F3004 3 10 10 25
1993-02-03 V9333005 2 2 2 2
1993-03-22 V93C0120 1 1 1 3
1993-04-27 V9333006 5 5 5 8
1993-08-01 V93H3009 3 3 3 11
1993-08-09 V9324200 1000 1000 1003 1011
1993-08-09 V93C0990 40 1040 1043 1051
1993-08-16 V9324320 1 1 1044 1052
1993-08-20 V93J3100 16 16 1060 1068
1993-08-27 V93F3088 10 10 1070 1078

Philippe Makowski - IBPhoenix - 2016-09-29

 Firebird SQL best practices

Thank you !

Philippe Makowski - IBPhoenix - 2016-09-29