Porting Oracle Applications to PostgreSQL

Porting the SQL Porting Tools PL/SQL vs. PL/pgSQL Interfaces Project Management

Porting Oracle Applications to PostgreSQL

Peter Eisentraut, credativ GmbH / credativ Ltd.

PGCon 2008

Porting Oracle Applications to PostgreSQL: 1 / 80


Disclaimers

This presentation was written by a PostgreSQL expert, not an
Oracle expert.
Both Oracle and PostgreSQL are diverse, complex, and
moving targets.
Both Oracle and PostgreSQL are (probably) Turing-complete,
so almost anything is “possible”, but we are looking for
reasonable options.



You Will See . . .

Porting projects are hard.
Compatibility and compatibility layers are an illusion.
It might be better not to do it.
But success can be very rewarding.



Outline

1 Porting the SQL

2 Porting Tools

3 PL/SQL vs. PL/pgSQL

4 Interfaces

5 Project Management



Outline

1 Porting the SQL

2 Porting Tools


4 Interfaces




Syntax

Identiﬁers Oracle case folds to upper case, PostgreSQL to lower
case. Big trouble if you mix quoted and unquoted
identiﬁers.
Column aliases SELECT foo [AS] bar — Most Oracle
applications omit the AS, but PostgreSQL requires it.
Fixed in PostgreSQL 8.4.
MINUS Change to EXCEPT.
SQL key words Usually not a big problem, but should be kept in
mind.
“FROM dual” Easy to work around (or use orafce).



Table Deﬁnition

The usual features are mostly the same: columns, constraints,
defaults.
Data types are more work; see below.
No fancy features like “table of type”.



Data Types: General

Both Oracle and PostgreSQL support plenty of
SQL-conforming data types.
But usually the nonconforming ones are in wider use.
Thin compatibility layers can usually help, but that will make
your PostgreSQL application unpretty.
A big search-and-replace is usually in order.



Data Types: Speciﬁcs

varchar2 → varchar or text
clob, long → varchar or text
nchar, nvarchar2, nclob → (varchar or text)
number → numeric or bigint or int or smallint or double
precision or real (bug potential)
binary ﬂoat/binary double → real/double precision
blob, raw, long raw → bytea (additional porting required)
date → date or timestamp



Null Values

Infamous Oracle behavior: NULL = ’’
Consequently, ’’ = ’’ is not true
Completely weird and inconsistent
Usually, your data will just disappear in PostgreSQL
transform_null_equals does not help here
If your application relies on any of this, you are in trouble.



Functions: General

Function compatibility is a bottomless pit.
PostgreSQL (+ orafce) supports many Oracle compatibility
functions.
It’s easy to write your own.
Only the special syntax is trouble.



Functions: Compatibility

For example, the following common functions are supported by
PostgreSQL as well:
substr
to char
nvl, nullif (orafce)



Functions: Speciﬁcs

Manual work required here:
sysdate → current_timestamp or localtimestamp



Functions: decode

DECODE(expr, search, expr[, search, expr...] [, default])

becomes

CASE WHEN expr THEN search .. ELSE default END



Default Parameters: Overview

PostgreSQL supports neither default values for parameters nor
named parameters in function calls.
Oracle applications make ample use of both.



Default Parameters: The Easy Case

CREATE FUNCTION foo (a int, b int, c int = 0) ...

becomes

CREATE FUNCTION foo (a int, b int, c int) ...

CREATE FUNCTION foo (a int, b int) ...
AS $$ SELECT foo(a, b, 0) $$;



Default Parameters: The Hard Case

CREATE FUNCTION foo (a int, b int = 5, c int = 0) ...

This is only callable with named parameters.
PostgreSQL doesn’t support this.
You will have to change your client application.
Your project time will double.



Default Parameters: Conclusion

Approx. 97% of applications to be ported contain issues like
this.
Client code must be reworked.
Adding this support in PostgreSQL would be a great feature.



Sequences: Creating

Sequences are somewhat compatible . . .
Change NOCACHE to CACHE 1 (or omit).
MAXVALUE 9999999999999999999999999 needs to be
reduced.
Don’t rely on the caching behavior.



Sequences: Using

Oracle syntax: sequence_name.nextval
PostgreSQL syntax: nextval(’sequence_name’)
Search-and-replace; but direct sequence calls are rare.



Outer Joins: Overview

PostgreSQL only supports the SQL-standard outer join syntax.
Oracle supports it since version 9.
Most Oracle code uses the old, Oracle-speciﬁc syntax.
Porting is usually straightforward, but requires manual work.
Set up test queries to catch porting mistakes.



Outer Joins: Simple Example

SELECT * FROM a, b WHERE a.x = b.y(+)

becomes

SELECT * FROM a LEFT JOIN b ON a.x = b.y



Outer Joins: Complex Example

SELECT ...
FROM A, B, C
WHERE A.A_ID (+) = B.A_ID
AND C.C_KEY(+) = B.C_KEY

becomes

SELECT ...
FROM A
RIGHT JOIN
B ON (A.A_ID = B.A_ID)
LEFT JOIN
C ON (C.C_KEY = B.C_KEY)



Outer Joins: Unclear Example

SELECT ...
FROM A, B, C, D, E
WHERE A.A_ID = B.A_ID
AND B.B_ID = C.A_ID(+)
AND B.B_KEY = C.B_KEY(+)
AND C.C_ID = D.C_ID(+)
AND B.A_ID = E.A_ID(+)
AND B.B_KEY = E.B_KEY(+)
AND ’CONSTANT’ = C.X_ID(+)

What’s that???



Locking

Transaction isolation, locking, SELECT FOR UPDATE behave
pretty much the same.
Oracle also defaults to read committed.
Usually, no one at the client has ever heard of concurrency
issues, so the code is likely buggy anyway.



Indexes

Basic syntax the same:
CREATE INDEX name ON table (a, b)
Primary keys and unique constraints are automatically
indexed.
Other features are implementation-speciﬁc.
You will have to re-tune the entire porting result anyway.



Optimizer Hints

Delete them
Or keep them for future investigation
Usually useless



Date Formatting

TO_CHAR is largely compatible.
Warning: PostgreSQL version is not very robust.
One-argument variant provided by orafce
NLS_DATE_FORMAT is replaced by locale settings.



Date Arithmetic

Usually, date arithmetic is easier in PostgreSQL, so consider a
small code rewrite.
orafce provides compatibility functions, such as last_day,
add_months.
Oracle code often does date + int . . .
In PostgreSQL, this may become timestamp + int.
This doesn’t work.
Write a custom operator or rewrite the code.



Encodings

Both Oracle and PostgreSQL support the same ideas.
But everything is named diﬀerently.
Might be a good time to review the encoding and locale
choices.



NLS * vs. LC *

Approximate analogies:

NLS CALENDAR —
NLS COMP lc collate = ’C’
NLS CURRENCY lc monetary
NLS DATE FORMAT DateStyle
NLS DATE LANGUAGE lc messages, lc time (8.4?)
NLS LANG, NLS LANGUAGE LANG, client encoding
NLS NCHAR —
NLS NUMERIC CHARACTERS lc numeric
NLS SORT lc collate
NLS TERRITORY LANG, lc *



ROWNUM and ROWID

ROWNUM:
Use generate_series, or
Rewrite and apply LIMIT, or
Just handle in the client
ROWID:
Analogous to ctid
Good code should usually not use this.
That does not prevent some from trying.



XML

(untested!)
xmltype → xml
extract → xpath
XMLELEMENT, XMLATTRIBUTES, etc. are the same.
Most functionality is diﬀerent or missing in PostgreSQL.



Triggers: Declarations

Oracle uses inline trigger actions:

CREATE TRIGGER foo AFTER action ON table
AS BEGIN ... END;

becomes

CREATE OR REPLACE FUNCTION foo_tg() RETURNS TRIGGER
LANGUAGE xxx
AS $$ ... $$;
EXECUTE PROCEDURE foo_tg();

Note: FOR EACH STATEMENT is the default in Oracle and
PostgreSQL.



Triggers: Column-Level Triggers
Oracle supports column-level triggers:
CREATE TRIGGER foo BEFORE UPDATE OF column ON table
AS BEGIN ... END;
becomes
CREATE OR REPLACE FUNCTION foo_tg() RETURNS TRIGGER
LANGUAGE xxx AS $$
BEGIN
IF NEW.column IS NOT DISTINCT FROM OLD.column THEN
RETURN NEW;
END IF;
... -- normal code
END;
$$;
EXECUTE PROCEDURE foo_tg();
But this doesn’t catch updates to the same value.
You will need to make a choice which behavior you need.


Things That Won’t Work Directly

CONNECT BY Try contrib/tablefunc.
Materialized views Write your own wrapper.
Snapshots Write your own wrapper.
Database links Use contrib/dblink plus views.
Autonomous transactions Try dblink.
Synonyms Try views or wrapper or schema path.
Partitioning Write your own system.



Outline

1 Porting the SQL

2 Porting Tools


4 Interfaces




orafce

http://orafce.projects.postgresql.org/
Large set of Oracle compatibility functions
“dual” table
Debian and RPM packages available
Invaluable



ora2pg

http://ora2pg.projects.postgresql.org/
Converts Oracle schema deﬁnitions
Extracts data from Oracle database for import into
PostgreSQL
Packages available
Invaluable



TOra

http://tora.sourceforge.net/
GUI for PostgreSQL and Oracle
Contains exploration and debugging facilities for Oracle
Packages available, but usually without Oracle support
Generally a bit outdated, but good for this purpose



DBD::Oracle

http://search.cpan.org/dist/DBD-Oracle/
Needed for ora2pg
Also helpful for test scripts etc.
Building it can be challenging
Debian and RPM packages available



Oracle Instant Client

Needed for DBD::Oracle and TOra
Also contains sqlplus
download from Oracle



Oracle Database Express Edition

Use this for testing if you have no other Oracle instance.
download from Oracle



Outline

1 Porting the SQL

2 Porting Tools


4 Interfaces




From PL/SQL to PL/pgSQL

Compatibility isn’t that great, but it’s obviously the best
choice.
The PL/pgSQL parser is DAAB.
See also http://www.postgresql.org/docs/current/
static/plpgsql-porting.html.



Function Creation

CREATE FUNCTION ... RETURN type becomes
CREATE FUNCTION ... RETURNS type
Function body must be quoted (dollar quoting).
Various other details are incompatible:
LANGUAGE
STRICT, STABLE, etc.
For variable declarations, DECLARE is needed in PostgreSQL.



Syntax Diﬀerences

FOR i IN REVERSE 1..10 LOOP — Order must be switched
for PostgreSQL.



Variables

PL/SQL can distinguish column names and variable names.
PL/pgSQL replaces all matching tokens by variables.
Find a namespacing mechanism to tell apart variables,
parameters, and columns.



Packages

Use schemas to group your functions.
Call syntax is about the same.
But there is no equivalent public/private mechanism.



Package Variables

Not supported by PostgreSQL
Write a wrapper based on (temporary) tables.



Cursors

Usually, you need less cursors in PostgreSQL.

CURSOR foo IS SELECT ...;

BEGIN
FOR x IN foo LOOP

can be simpliﬁed to

BEGIN
FOR x IN SELECT ... LOOP

Note: The x is deﬁned implicitly in Oracle. In PostgreSQL, you
need to declare it.



Cursors Variables
This doesn’t work in PostgreSQL:

CURSOR foo IS SELECT ...;
x foo%ROWTYPE;

BEGIN
FOR x IN foo LOOP

Use RECORD:

DECLARE
CURSOR foo IS SELECT ..;
x RECORD;

BEGIN
FOR x IN foo LOOP



PERFORM

In PostgreSQL, “procedure” calls must start with PERFORM. E. g.,

service.put_utl(’Error’);

becomes

PERFORM service.put_utl(’Error’);



EXECUTE

For DDL statements, EXECUTE might be necessary. E. g.,

EXECUTE ’CREATE TABLE ’ || quote_ident(foo) || ...



Subcommits

Code that does COMMIT or ROLLBACK needs major, client-side
changes. (Savepoints won’t usually do the job.)



Exceptions (1)

An exception rolls back all changes implicitly in PostgreSQL.
You can drop most savepoint-using code from the Oracle
version.
More complex behavior needs a redesign.



Exceptions (2)

Exception block syntax is the same.
Exception names are different.
Oracle supports user-defined exception names.
Error codes are different.
Variable SQLERRM is available.
Of course, error messages are also different.



Exceptions (3)

We use a scheme to encode exception information into the
message string:

RAISE name; ---> RAISE EXCEPTION ’name’;

Similar for error codes:

raise_application_error(12345, ’msg’);
---> RAISE EXCEPTION ’+12345:msg’;

Codes can be positive or negative. Write a wrapper function.

errcode := substr(SQLERRM, 1, 6)



No Data Found Exceptions

Oracle throws NO_DATA_FOUND exceptions for
SELECT
INSERT
UPDATE
DELETE
PostgreSQL only for:
SELECT INTO STRICT
Use IF NOT FOUND to deal with other cases.



Logging

dbms_output “package” is provided by orafce. E. g.

dbms_output.put_line(’WRONG PARAMETER: ’ || par);

Watch for interferences from null values!



Backtraces

orafce provides dbms_utility.format_call_stack()
impossible to implement
dbms_utility.format_error_stack() in PostgreSQL
(except by patching PL/pgSQL directly)



What About PL/Java?

Should be compatible with SQL/JRT and Oracle
Basic functionality should work without changes
Reality is more complex
There is little or no experience with this scenario.



Outline

1 Porting the SQL

2 Porting Tools


4 Interfaces




psql/sqlplus

psql is much nicer for interactive use. :-)
sqlplus is much nicer for scripting use. :-(
With use of variables and naming conventions, sqlplus scripts
can be converted anyway.
Consider a wholesale rewrite.



Backup, Recovery

Build a new system using transaction log archiving or SQL dumps.



Setup, initdb

Works completely diﬀerently.
Forget everything you get from Oracle.
Write new setup scripts that integrate well with the operating
system.
Forget about tablespaces, partitioning, OS tuning, etc. until
you have a porting result.



JDBC

Works great, aside from SQL syntax issues



Outline

1 Porting the SQL

2 Porting Tools


4 Interfaces




Testing

Have a test suite for:
functions
setup
tables/database contents



Dividing the Work

In PL/SQL-heavy applications, you can usually divide the
work by function or package.
Someone needs to drive and monitor integration work.
Have a test suite.



Long-Term Maintenance

Will the original application continue to be developed?
. . . while the porting project runs?!?
How is the merging going to work?
One time ports, maintainable ports, and mergeable ports are
all slightly diﬀerent.



Code Formatting

Create a code formatting standard.
(This applies to any development project.)
I tend to stick with the original layout.
This is important for later updates, merges, and maintenance.



Version Control

Use version control, even if the client doesn’t.
Prefer to use your own VCS; merge later.



Beyond the SQL

Applications also contain setup and maintenance scripts.
These were typically written by old-school Oracle
administrators.
Hence completely incomprehensible and written in ksh
Half the logic usually doesn’t make sense for PostgreSQL.
Half the logic required to make PostgreSQL work will not
have been written yet.
Reserve plenty of time for dealing with this.



Legacy Code and Legacy Environments

Applications to be ported are usually very old and crufty.
This multiplies the time required to deal with them.
A lot of the code won’t compile/run on newer operating
systems.
Half your tools won’t compile/run on the old operating
system.
Everything is locked down, so you can’t do anything about
this.
Evaluate this carefully before starting the project.



Client Participation

Almost always, the database clients will need to be adjusted.
Almost always, you need someone reachable who understands
the code.
Clients think SQL code is a black box with a clean and simple
interface.
In practice, a port is like a major new software release.
A port aﬀects your entire system.
The client must be willing, able, and available to participate in the
project.



“Josh’s Rules (of Database Contracting)”

http://blogs.ittoolbox.com/database/soup/archives/
joshs-rules-of-database-contracting-17253
Learn them by heart.
Print them out.
Post them at your oﬃce door.
Quote them to the sales people.



Contribute Your Improvements

Porting projects are a great source of ideas for features and
bug ﬁxes.
Record your experiences, e. g., in the wiki.
Contribute to orafce and ora2pg.
Contribute to PL/pgSQL and PL/Java.



Coincidence?

If you need help:
Oracle Ask Tom: http://asktom.oracle.com/
PostgreSQL Ask Tom: tgl@sss.pgh.pa.us



Fly Way to Conclude SQL-Themed Presentation

COMMIT;


Porting Oracle Applications to PostgreSQL

Recommended

More Related Content

What's hot (20)

Viewers also liked (7)

Similar to Porting Oracle Applications to PostgreSQL (20)

More from Peter Eisentraut (20)

Recently uploaded (20)

Porting Oracle Applications to PostgreSQL