Re: PostgreSQL vs. MySQL: fight

On Sat, 4 Aug 2007, Florian Weimer wrote:

> I think for fairness, you should mention that PostgreSQL's MVCC locking
> does not properly implement the SERIALIZABLE semantics.  For some table
> types (InnoDB, IIRC), MySQL implements phantom key logging, so it gets
> more cases right.

I assume you mean phantom key locking, not logging.

I've been trying to follow up on this, but I still don't understand
exactly what you're describing--certainly not well enough to explain it.
There's a section in the PostgreSQL documentation describing "Serializable
Isolation versus True Serializability" at
http://www.postgresql.org/docs/current/static/transaction-iso.html ; does
that cover what you describe as "not properly implement the SERIALIZABLE
semantics" or is there something else you're alluding to here?

I also can't find anything definitive on why MySQL's phantom key
implementation is a better solution.  The two most relevant documents seem
to be

http://dev.mysql.com/doc/refman/5.0/en/innodb-next-key-locking.html
http://www.greatlinux.com/mysql/books/mysqlpress/mysql-tutorial/ch10.html

but I don't see how that "gets more cases right".  Can you comment more
about this?

As a side-note, it's hard for me to feel too compelled to point out a
theoretical advantage for MySQL here when I find stuff like
http://bugs.mysql.com/bug.php?id=27197 floating around (that's just the
worse of several such bugs I came across when researching this topic).

--
* Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD