Re: Re: Getting last record ID created from DB

markw@mohawksoft.com wrote:

markw@mohawksoft.com wrote:

markw@mohawksoft.com wrote:

On Sat, 2007-03-17 at 12:19 -0400, Mark wrote:

There in lies the biggest problem with LAMP, and that's MySQL. The architecture of your methodology *only* works with MySQL, and not more capable databases like Oracle, DB2, or even PostgreSQL.

I too thought the same thing, but was corrected when I starting working on a redesign for the billing system for the company that I am working for. They use PostgreSQL and we do have a auto incremented unique id field. It isn't call AUTO_INCREMENT, but rather a sequence value. It is a value that is controlled by the default value entry. It then runs this: nextval('customers_customer_id_seq'::regclass) and uses this as the value of the id field for the newly inserted row.

Yea, I've been using PG for over 10 years now. Sequences are awesome, but there have an "auto number" column type as well. It simply creates a sequence but it's there. I guess maybe most databases support the notion of "auto number" in some form. I haven't used MSSQL recently, but I think their solution is a GUID.

Then instead of running a command like SELECT LAST_INSERT_ID(); you have to do some other stuff like this: $SQL = "SELECT currval('{$table}_{$column}_seq');";

Again, this is absolutely wrong unless it is wrapped in a transaction. You will need a "begin" and a "commit" in an MVCC type system. If you don't wrap it in a transaction you don't know who got the last "nextval" of the sequence.

Check the documentation - currval returns the last one *for that session* - it does not return the last global change. http://www.postgresql.org/docs/current/static/functions-sequence.html It is perfectly safe to use this.

In theory that may be true, but can the application developer make any assumption about the underlying architecture? Might you be familiar with connection pooling? Where multiple threads or processes share a database connection or "session?"

If the developers of the database say it's safe, then I'm happy to take their word on it. If you can prove otherwise, please do so and post a bug report to them. In this case, tons of people are using connection pooling with postgres. If a problem had been discovered, then they would have complained and it would have been fixed. Do I need to evaluate every section of code that a 3rd party provides to make sure the code does what the docs say it should? I'd never get anything done. At some point I just need to know how to use it and trust it works as it should. If it's a super critical application (eg financial information) where integrity is 100% required, then I'd test it. If I found a problem then I'd complain *shrug*. As Richard mentioned in another reply, the developers (whether it's mysql or postgres or "other") have taken a lot of pain to make sure these things work properly. Why would I try and do exactly the same thing myself and most likely get it completely wrong? The rest of your rant I agree with in theory, but making something "flexible" is extremely difficult unless you understand what you're trying to accomplish from the beginning - and know each system well enough to know how each one works and the differences between them. Not even "LIMIT" is standard across databases let alone anything else like data types ('int(11)' compared to 'int', 'datetime' compared to 'timestamp' etc etc etc) or even date formats that they store. Taking all of that into account from the very beginning is pretty tough. -- Postgresql & php tutorials http://www.designmagick.com/