Effective PL/SQL

Thomas Kyte
http://asktom.oracle.com/
Agenda

• Why PL/SQL?
• Write as little as you can
• Use Packages
• Use Static SQL
• Bulk processing
• Implicit or Explicit?
• Beware of some features
• Things to definitely do
Why PL/SQL
Why Use PL/SQL

• It is a ‘real’ language
– It is not a scripting language
– It is not a ‘toy’, it is used to code ‘real’ things

• It is the API to the database

It is the most efficient language for data
manipulation

• If your goal is to procedurally process data (after

ensuring a single SQL statement cannot do your
work!) then PL/SQL is simply the most productive
language to do so
 It is the most efficient language for data
manipulation

Create or replace procedure

process_data( p_inputs in varchar2 )
As
Begin
For x in ( select * from emp
where ename like p_inputs )
Loop
Process( X );
End loop
End;
• SQL datatypes are PL/SQL datatypes
• Tight coupling between the two languages
• Code short cuts (implicit cursors)
• Protected from many database changes
 It is the most efficient language for data
manipulation
static PreparedStatement while( rset.next() )
pstmt = null; {
empno = rset.getInt(1);
public static void ename = rset.getString(2);
process_data job = rset.getString(3);
( Connection conn, String inputs ) mgr = rset.getInt(4);
throws Exception hiredate = rset.getString(5);
{ sal = rset.getInt(6);
int empno; comm = rset.getInt(7);
String ename; deptno = rset.getInt(8);
String job; process( empno, ename, job, mgr,
int mgr; hiredate, sal, comm, deptno );
String hiredate; }
int sal; rset.close();
int comm; Pstmt.close();
int deptno; }
• SQL datatypes are not Java types (consider
if ( pstmt == null )
pstmt = conn.prepareStatement number(38) issues…)
("select * from emp " +
"where ename like ? " ); • No coupling between the two languages,
pstmt.setString( 1, inputs ); entirely procedural (what about SQLJ?)
ResultSet rset =
pstmt.executeQuery(); • No code short cuts (statement caching)
……
• Not protected from many database changes
(and no dependencies either!)
PL/SQL epitomizes portability and reusability

• It is the most advanced portable language I’ve ever

seen
– It is callable from every other language out there
– Anything that can connect to the database can use and reuse
it
• Sure – there are things like SOA and Services that let
X call Y
– But these introduce their own level complexities
– And if your service is a database server, it would be best to
be written in the database
• If you can connect to the database – you can use and
reuse PL/SQL from anything
Many mistakes made in other languages using the
database are avoided

• Bind Variables
– If you use static SQL in PL/SQL it is impossible to not use bind
variables correctly.
– You have to use cumbersome dynamic SQL to do it wrong.
• Parse Once, Execute many
– PL/SQL does statement caching
– You have to either configure and enable its use in other languages
or
– Do it your self (refer back to java code)
• Schema Changes are safer
– Alter table t modify c1 varchar2(255);
– We can find all uses of T (I didn’t know you were using that, sorry..)
– We can make the change without having to change code
However

• As with any language you can

– Write really good code
– Write really average code
– Write really really really bad code
• You can make the same mistakes with PL/SQL that
you can with every other language
– By not understanding the language
– By not understanding some implementation details
– By not understanding SQL
– By not designing
– And so on…
Write as Little
as you can
Code…

• Write as much code:

– As you have to
– But as little as you can…
• Think in SETS
• Use (really use – not just ‘use’) SQL

Begin into table (c1,c2,…)

Insert
select
For x
c1,c2,….
in ( select
From *
table@remote_db
from table@remote_db )
LOGLoop
ERRORS ( some_variable )
REJECTInsert
LIMIT into
UNLIMITED;
table ( c1, c2, … )
values ( x.c1, x.c2,… );
… code
End loop;
to handle errors
End;
for tag some_variable …
Use PL/SQL constructs only when SQL cannot do it

• Another coding ‘technique’ I see frequently:

For a in ( select * from t1 )

Loop
For b in ( select * from t2
where t2.key = a.key )
Loop
For c in ( select * from t3
where t3.key = b.key )
Loop
…

• The developer did not want to “burden” the database

with a join
More Code = More Bugs
Less Code = Less Bugs

• This code speaks for itself.

• So does this.

• Always look at the procedural code and ask yourself

“is there a set based way to do this algorithm”
– For example …
 More Code = More Bugs
Less Code = Less Bugs For any set of records with
more than one EMPNO,
insert into t ( .... )
remove rows with the oldest
select EMPNO, STATUS_DATE, .... STATUS_DATE.
from t1, t2, t3, t4, ....
where ....; Additionally – If the last set of
EMPNO records all have the
loop same STATUS_DATE, remove
delete from t them all.
where (EMPNO,STATUS_DATE)
EMPNO STATUS_DATE
in ( select EMPNO, ------- ------------
min(STATUS_DATE) 1 01-jan-2009
from t 1 15-jun-2009
group by EMPNO 1 01-sep-2009
having count(1) > 1 );
exit when sql%rowcount = 0; …
end loop; 1000 01-feb-2009
1000 22-aug-2009
1000 10-oct-2009
(1,01-jan-2009,3)
(1,15-jun-2009,2) (1001,01-feb-2009,4)
(1001,22-aug-2009,3)
(1001,22-aug-2009,2) 1000 10-oct-2009
 More Code = More Bugs
Less Code = Less Bugs
insert /* APPEND */ into t ( .... )
select EMPNO, STATUS_DATE, ......
from ( select EMPNO, STATUS_DATE, .... ,
max(STATUS_DATE)
OVER ( partition by EMPNO ) max_sd,
count(EMPNO)
OVER ( partition by EMPNO,STATUS_DATE ) cnt
from t1, t2, t3, t4, …
where … )
where STATUS_DATE = max_sd
and cnt = 1;

• This was a data warehouse load (load 2-3-4 times the data you want,
then delete? Ouch)
• It was wrong – procedural code is no easier to understand than set based
code, documentation is key
 More Code = More Bugs
Less Code = Less Bugs
/* Load table t using history tables. History tables have
multiple records per employee. We need to keep the
history records for each employee that have the maximum
status date for that employee. We do that by computing
the max(status_date) for each employee (partition by EMPNO
finding max(status_date) and keeping only the records such
that the status_date for that record = max(status_date)
for all records with same empno */

insert /* APPEND */ into t ( .... )

select EMPNO, STATUS_DATE, ......
from ( select EMPNO, STATUS_DATE, .... ,
max(STATUS_DATE)
OVER ( partition by EMPNO ) max_sd
from t1, t2, t3, t4, …
where … )
where STATUS_DATE = max_sd;
 More Code = More Bugs
Less Code = Less Bugs
/* Load table t using history tables. History tables have
multiple records per employee. We need to keep the
history records for each employee that have the maximum
status date for that employee. We do that by numbering
each history record by empno, keeping only the records such
that it is the first record for a EMPNO after sorting by
status_date desc. REALIZE: this is not deterministic if
there are two status_dates that are the same for a given
employee! */
insert /* APPEND */ into t ( .... )
select EMPNO, STATUS_DATE, ......
from ( select EMPNO, STATUS_DATE, .... ,
row_number() OVER ( partition by EMPNO
order by STATUS_DATE desc ) rn
from t1, t2, t3, t4, …
where … )
where rn=1;
“Here is a last bit of advice on writing as little as possible: When you are
writing code, make sure your routines (methods, procedures, functions,
or whatever you want to call them) fit on a screen. You should be able to
see the logic from start to finish on your monitor. Buy the biggest
monitor you can, and make the routines fit on your screen. This rule
forces you to think modularly, so you break up the code into bite-sized
snippets that are more easily understood”
Use Packages
 They break the dependency chain

• Most relevant in Oracle Database 10g Release 2 and

before:

ops$tkyte%ORA10GR2> create or replace procedure p1 as begin null; end;

ops$tkyte%ORA10GR2> create or replace procedure p2 as begin p1; end;

OBJECT_NAME STATUS TO_CHAR(LAST_DD

------------------------------ ------- ---------------
P1 VALID 04-oct 12:15:54
P2 VALID 04-oct 12:15:54

ops$tkyte%ORA10GR2> create or replace procedure p1 as begin /* updated */ null; end;

OBJECT_NAME STATUS TO_CHAR(LAST_DD

------------------------------ ------- ---------------
P1 VALID 04-oct 12:15:58
P2 INVALID 04-oct 12:15:54
They increase your namespace

• You can have only one procedure P in a schema

– What about EBR?
– Killer Feature of 11g Release 2
– Bryn is talking about EBR Wed at 11am!
• With packages, you can have as many procedure P’s
as you need
– Less chance of developer X using the same ‘name’ as
developer Y since only package names would clash
• A single package has many procedures/functions
– Reduces dictionary “clutter”
– Organizes things, related code goes together
– Promotes modularity
They support overloading

• A feature which is viewed as

– Positive by some
– Negative by others
• Overloading can be very useful in API packages
– 259 out of 728 ‘SYS’ packages employ this technique
They support encapsulation

• Helps live up to the “fit on a screen” rule

– Many small subroutines that are no use outside of the
package
– Hide them in the package body, no one can see them
– Reduces clutter in the dictionary

• Allows you to group related functionality together

– Makes it obvious what pieces of code are to be used together

• They support elaboration code

– When package is first invoked, complex initialization code
may be executed
Use Static SQL
Static SQL is checked at compile time

• You know the SQL will (probably) execute

– It is syntactically correct
– It could still raise an error (divide by zero, conversion error,
etc)
– It might be semantically incorrect, but that is a bug in your
logic, not a criticism of static SQL
PL/SQL understands the dictionary

• It will create record types for you

• It will allow you to define variables based on the

database types

• The compiler does more work, so you don’t have to.

One word - dependencies

• All referenced objects – tables, views, other bits of

code, etc – are right there in the dictionary.

• No more “Oh sorry, I didn’t know you were using that”

• If something changes – we know right away if

something is broken
– Grants – lose one that you need, code will stay invalid
– Drop column – drop one that you reference, code will stay
invalid
– Modify length of column – if you reference that, code will
recompile with new size.
Static SQL makes parse once, execute many a
reality

• Dynamic SQL makes it easy to lose out on this benefit.

• With DBMS_SQL, you have to cache the ‘cursor’ yourself and

make sure you use it over and over (eg: do not call
dbms_sql.close() until you have to)

• With native dynamic SQL, you need to make sure you use the
same SQL text over and over to cache statements
– And if you are doing that, why did you use dynamic SQL again?
– Different in 9i and before than 10g and later

• Impossible to be SQL Injected with static SQL! Far too easy to

be SQL Injected with dynamic SQL.
Dynamic SQL – when to use then?

• Dynamic SQL is something you want to use when

static SQL is no longer practical—when you would be
writing hundreds or thousands of lines of code, and
can replace it with a very small bit of safe (sql
injection) dynamic SQL.

• When you’ve shown that using static SQL would not

be practical – that is, it is never your first choice.
Bulk Up
Bulk Processing Defined:

• A method to bother the database less often

• A method to reduce round trips (even from PL/SQL to
SQL – there is a ‘round trip’ involved
• A method to utilize fewer resources in general
• A method to maintain data structures in better shape

• You get some data (more than a row), process it, and
send it all back (to update/insert/delete).
Bulk Processing

• You need to do it when…

– You retrieve data from the database
– AND you send it back to the database

• You need NOT do it when…

– You retrieve data from the database
– <this space left intentionally blank>
– For example…
Bulk Processing

• You need to do it when… THIS IS BAD CODE

For x in ( select * from t where … )
Loop
process(x);
update t set … where …;
End loop;

• Implicit array fetch for select

• Not so for update… Details on next slide
• You need NOT do it when… THIS IS OK CODE
For x in (select * from t where …)
Loop
dbms_output.put_line( … t.x … );
End loop;

• Implicit array fetch for select

• No going back to database
 Bulk Processing

create or replace procedure bulk create or replace procedure slow_by_slow

as as
type ridArray is table of rowid; begin
type onameArray is table for x in (select rowid rid, object_name
of t.object_name%type; from t t_slow_by_slow)
loop
cursor c is select rowid rid, object_name x.object_name := substr(x.object_name,2)
from t t_bulk; ||substr(x.object_name,1,1);
l_rids ridArray; update t
l_onames onameArray; set object_name = x.object_name
N number := 100; where rowid = x.rid;
begin end loop;
open c; end;
loop
fetch c bulk collect
into l_rids, l_onames limit N;
for i in 1 .. l_rids.count
loop
l_onames(i) := substr(l_onames(i),2)
||substr(l_onames(i),1,1);
end loop;
forall i in 1 .. l_rids.count
update t
set object_name = l_onames(i)
where rowid = l_rids(i);
exit when c%notfound;
end loop;
close c;
end;
 Bulk Processing
SELECT ROWID RID, OBJECT_NAME FROM T T_BULK

call count cpu elapsed disk query current rows

------- ------ -------- ---------- ---------- ---------- ---------- ----------
total 721 0.17 0.17 0 22582 0 71825
********************************************************************************
UPDATE T SET OBJECT_NAME = :B1 WHERE ROWID = :B2

call count cpu elapsed disk query current rows

------- ------ -------- ---------- ---------- ---------- ---------- ----------
Parse 1 0.00 0.00 0 0 0 0
Execute 719 12.83 13.77 0 71853 74185 71825
Fetch 0 0.00 0.00 0 0 0 0
------- ------ -------- ---------- ---------- ---------- ---------- ----------
total 720 12.83 13.77 0 71853 74185 71825

SELECT ROWID RID, OBJECT_NAME FROM T T_SLOW_BY_SLOW

call count cpu elapsed disk query current rows

------- ------ -------- ---------- ---------- ---------- ---------- ----------
total 721 0.17 0.17 0 22582 0 71825
********************************************************************************
UPDATE T SET OBJECT_NAME = :B2 WHERE ROWID = :B1

call count cpu elapsed disk query current rows

------- ------ -------- ---------- ---------- ---------- ---------- ----------
Parse 1 0.00 0.00 0 0 0 0
Execute 71824 21.25 22.25 0 71836 73950 71824
Fetch 0 0.00 0.00 0 0 0 0
------- ------ -------- ---------- ---------- ---------- ---------- ----------
total 71825 21.25 22.25 0 71836 73950 71824
 But of course, the bulkier the better…
SELECT ROWID RID, OBJECT_NAME FROM T T_BULK

call count cpu elapsed disk query current rows

------- ------ -------- ---------- ---------- ---------- ---------- ----------
total 721 0.17 0.17 0 22582 0 71825
********************************************************************************
UPDATE T SET OBJECT_NAME = :B1 WHERE ROWID = :B2

call count cpu elapsed disk query current rows

------- ------ -------- ---------- ---------- ---------- ---------- ----------
Parse 1 0.00 0.00 0 0 0 0
Execute 719 12.83 13.77 0 71853 74185 71825
Fetch 0 0.00 0.00 0 0 0 0
------- ------ -------- ---------- ---------- ---------- ---------- ----------
total 720 12.83 13.77 0 71853 74185 71825

Lots less code too! (dml error logging if you need)

update t set object_name = substr(object_name,2) || substr(object_name,1,1)

call count cpu elapsed disk query current rows

------- ------ -------- ---------- ---------- ---------- ---------- ----------
Parse 1 0.00 0.00 0 0 0 0
Execute 1 1.30 1.44 0 2166 75736 71825
Fetch 0 0.00 0.00 0 0 0 0
------- ------ -------- ---------- ---------- ---------- ---------- ----------
total 2 1.30 1.44 0 2166 75736 71825
Returning Data
To return data to a client program

• Either
– Simple, formal OUT parameters
– Ref cursor for all result sets

• Do not run a query

– To populate a collection
– To return collection to client

• Just run the query (open CURSOR for SQL_STMT)

– Ease of programming, everything can handle a cursor
– Flexibility (client decides how many rows to deal with, less memory
intensive)
– Performance – client might never get to the last row (probably won’t)
Implicit versus
Explicit
Implicit versus Explicit

• Implicit
– With this type of cursor, PL/SQL does most of the work for
you. You don’t have to open close, declare, or fetch from an
implicit cursor. For x in ( select * from t where … )
Loop
…
• Explicit End loop;

– With this type of cursor, you do all of the work. You must
open, close, fetch, and control an explicit cursor completely.
Declare
cursor c is select * from t where …;
l_rec c%rowtype;
Open c;
Loop
fetch c into l_rec;
exit when c%notfound;
…
End loop;
Close c;
Implicit versus Explicit

• There is a myth that explicit cursors are superior in performance

and usability to implicit cursors.

• The opposite is generally true

– Implicit cursors have implicit array fetching, Explicit cursors do not
– Implicit cursors have many of their operations hidden in the PL/SQL
runtime (C code) as opposed to explicit cursors being coded by you
in PL/SQL code.
– Implicit cursors are safer than explicit cursors code-wise
• Select into checks (at least and at most one row)
• Cursors opened/closed for you – implicitly – no ‘leaks’
• Both implicit and explicit cursors however are cached by PL/SQL
– But ref cursors are not…
 Single Row Processing

• Implicit Select … INTO <plsql variables>

from …
where …;

• Explicit Declare
cursor c is select … from … where …;
Begin
open c;
fetch c into <plsql variables>;
if (c%notfound) then
• These two bits of raise no_data_found;
end if;
code do the same fetch c into <plsql variables>;
if (c%found) then
thing. Which is raise too_many_rows;
end if;
more efficient? close c;
 Single Row Processing

• This is a bug Function get_something return …

is
• This is a bug I see a lot cursor c is select … from … where …;
• It is bad, but at least it Begin
open c;
probably returns NULL fetch c into <plsql variables>;
close c;
return plsql variables

----------------------------------------
• This is a ‘worse’ bug
• This I see even more Begin
open c1;
• Combines the “do it loop
yourself join” with “do it the fetch c1 into l_data;
exit when c1%notfound;
hard way” to get the wrong open c2( l_data );
answer after a longer fetch c2 into l_something_else;
period of time with lots close c2;
more development time process( l_data, l_something_else);
end loop;
close c1;
 Multi-Row Processing

• Which is “easier”? create or replace procedure implicit

as
• Which is more “bug begin
for x in ( select * from dept )
free”? loop
null;
end loop;
end;
• Which one array ----------------------------------------
create or replace procedure explicit
fetches for us? as
l_rec dept%rowtype;
• The explicit code cursor c is select * from dept;
begin
would be a lot more open c;
loop
intense if we wanted to fetch c into l_rec;
array fetch. exit when c%notfound;
end loop;
close c;
end;
 Multi-Row Processing

• This is nice too – all of create or replace procedure implicit

As
the benefits of implicit, cursor c is select * from dept;
the factoring out of the begin
for x in C
SQL usually attributed loop
null;
to explicit cursors. end loop;
end;
----------------------------------------
• This code in the tiny create or replace procedure explicit
as
type array is table of dept%rowtype;
font (to make it fit on l_rec array;
n number := 2;
the screen) does the cursor c is select * from dept;
begin
same thing as the open c;
loop
above bit of code. fetch c bulk collect into l_rec limit N;
for i in 1 .. l_rec.count
Which is more loop
null;
efficient? Which is less end loop;
exit when c%notfound;

bug likely? end loop;

close c;
end;
 Multi-Row Processing
create or replace procedure bulk
• So, is there ever a time as
type ridArray is table of rowid;
type onameArray is table of t.object_name%type;
to use explicit cursors? cursor c is select rowid rid, object_name
from t t_bulk;
– Never say Never l_rids ridArray;
l_onames onameArray;
– Never say Always N number := 100;
begin
– I always say open c;
loop
fetch c bulk collect
into l_rids, l_onames limit N;
• Bulk forall processing for i in 1 .. l_rids.count
loop
probably mandates l_onames(i) := substr(l_onames(i),2)
||substr(l_onames(i),1,1);
explicit cursors end loop;
forall i in 1 .. l_rids.count
update t
set object_name = l_onames(i)
where rowid = l_rids(i);
• Ref Cursors mandate exit when c%notfound;
end loop;
explicit cursors close c;
end;
Beware of…
Beware – When others

• When others
• Autonomous Transactions
• Triggers
Things to do…
“Instrument your code. Make it debuggable. Make it so that tracing the
code is easy. Identify yourself with DBMS_SESSION. Now
DBMS_MONITOR can trace you. Add copious calls to
DBMS_APPLICATION_INFO, now we can see who you are, what you
are doing and how long you’ve been doing it. Add calls to a logging
package (for example http://log4plsql.sourceforge.net/) to enable remote
debugging of your code. Use conditional compilation to enable extra
intensive debug code to be ‘deployed’ to production.”
“Use the tools – SQL Developer, built in source code
debugging. Hierarchical profiling – for performance. Tracing
tools to see how your SQL is doing. ASH/AWR reports.
PL/SCOPE. Learn the tools, then use them.”
“Always test things out – especially advice. I used to advise to use
BULK COLLECT for everything. That changed in Oracle Database
10g when they started implicitly doing that for us. There is advice on
the ‘internet’ to never use implicit cursor – always use explicit. It is
wrong. If they suggest it is “faster” and you cannot see it being
“faster”, question the advice.”
“Question Authority, Ask Questions”