This issue has come up before and after contemplating all the comments
surrounding the currval/nextval issue I still feel that getting the nextval
from a sequence before the insert statement is the best and most universal
approach. The nextval should work in any database environment. Did Herouth
actually create the code surrounding how currval works? Is this how other
databases deal with their equivalent of currval? It would important for me
to know whether this technique is valid for Postgres or all databases (esp
Oracle and SQL Server). Given the details of the explanation below and how
it interacts with multiple simultaneous users, I would like to know how
intimate Herouth is with the actual code.
Thanks, Michael
-----Original Message-----From: Herouth Maoz [SMTP:herouth@oumail.openu.ac.il]Sent: Wednesday, September 22, 1999
10:51AMTo: Hutton, Rob; steinbeck@ice.mpg.deCc: pgsql-sql@postgreSQL.orgSubject: RE: [SQL] Re: [INTERFACES]
JDBCand getting just
assigned serial number
At 18:21 +0200 on 22/09/1999, Hutton, Rob wrote:
> Except in an active database, or possibly not so active, where
another> record is inserted between the time yours is committed and you do
the> currval. It is much safer to do a select on nextval to get the
value, the> put it in as part of your update. It eliminates the chance that
you will> get the wrong record, and there is minimal overhead.
NOT TRUE!
Sorry to shout, but currval is the *safe* way to do this. It
absolutelyguarantees you get the correct number, it was designed exactly for
thispurpose. Using nextval and putting the value in the table will work,
but itrequires two operations for the insert, thus potentially wasting
morenumbers (someone else locks the row between the nextval and the
insert).
Currval works more or less like this: You have some part of memory
for thesession between you and the backend. When it draws a nextval for
you, itautomatically puts the number it has drawn in this session memory.
Thecurrval operation gets this stored number. It DOES NOT get the last
valuefrom the sequence table. That's a common misconception.
To make myself clear, let's suppose process A and process B run the
sameprogram, where the table has two fields, one serial named "ser", the
othera data field, named "dat".
INSERT INTO the_table( dat ) VALUES ('value');SELECT currval( 'the_table_ser_seq' );
This has an implicit nextval( 'the_table_ser_seq' ) within the
INSERT.
Process Does Memory stateA initially emptyB initially emptyA INSERT
'the_table_ser_seq' = 28B INSERT 'the_table_ser_seq' = 29A currval
'the_table_ser_seq'= 28 --- returns 28B currval 'the_table_ser_seq' = 29 --- returns 29A (again)
INSERT 'the_table_ser_seq' = 30
if B asks for currval, it still has 29. It will answer 29. You see?
Herouth
--Herouth Maoz, Internet developer.Open University of Israel - Telem projecthttp://telem.openu.ac.il/~herutma
************
