[development] PDO and database limitations

Wed Apr 4 16:52:17 UTC 2007

>
> > Which distro?
> >
> > I use Ubuntu 6.10 and if you install PHP 5, 5.1.6 is the default you
> get.
> > This makes
> > it harder to use 5.2 (custom compile, ...etc.)
> >
> > If you have working code, I can verify whether I have the same problem
> or
> > not on my test server.
>
> I also run Ubuntu Edgy (6.10), which is how I ran into this
> problem.  After spending 4 hours trying to figure out why I was segfaulting
> only sometimes on a session-load query, I checked with the php-general
> development list and was told that 5.1.6 had a lot of segfault issues.  I
> found a 3rd party repository and installed 5.2.1 and the segfaults all
> went away.

Interesting. Out of three Ubuntu installs, all of them 5.1.6, only one
segfaults, but
that is because of eAccelerator. I  have a client who uses Fedora Core 4,
and their
PHP segfaulted every hour or two when we put in eAccelerator. We compiled
PHP 5.2.1 and eAccelerator, and all the problems went away.

> My concern here is that all db_num_rows() cases become two queries for
> > people using MySQL (the vast majority) as well as PostgreSQL (it has
> > pg_num_rows()).
>
> Not all, actually.  At least one (in sess_write()) really should have been
> a count query in the first place as it's never iterated.  I'm not sure how
> many cases will simply become 2 queries and how many will be able to stay as
> one with some logic refactoring.  I don't think I'll know that until I start
> doing so.

I guess the safest way is to benchmark selected page loads, and see what the
impact is. But from the looks of it, it seems two queries is slower than
just one.

> If these are index only queries, then it may not be a big deal, but
> > anything more (which we have a lot of), would incur a significant
> performance penalty.
> > It also affects large sites using InnoDB which is notorious for very
> slow SELECT
> > COUNT(*).
> >
> > See a benchmark here:
> >
> >
> http://2bits.com/articles/mysql-innodb-performance-gains-as-well-as-some-pitfalls.html
>
> Wouldn't InnoDB be just as slow running a full select statement and then
> doing a num_rows() on it as it would be doing a COUNT() query in the first
> place?  It still has all the WHERE-ing to do.  More specifically, MyISAM is
> a special case where count(*) is faster than it otherwise would be in any
> other situation.  It's not that InnoDB is slow, it's that MyISAM is fast.
> :-)

But at present,  db_num_rows() is a wrapper around mysql_num_rows() which
does not
do a COUNT(*) query, so it is fast on both MyISAM and InnoDB.

Still, though, I suspect that at least some cases will not get hosed if we
> refactor properly; some don't need to be full normal queries in the first
> place, and I am not opposed to adding additional indexes where appropriate.

Agreed.

> Not sure what else can be done here. Is there a way to sneek in a pass
> > thru
> > function
> > (mysql_num_rows() in this case) like ODBC for example allows?
>
> I'm not sure what ODBC does.  Can you elaborate?

It has been a while, but in ODBC, there is a pass-thru query mode, where
instead
of writing portable queries that work with any ODBC supported database (and
write
to the lowest common denominator), you can tell ODBC to just pass the
db-engine
specific query through and get back the results. This allows queries that
are
optimized for a certain db engine to be executed, and/or queries that use
features
in the engine your are using.

I'm pretty sure one can't call mysql_* functions on a PDOStatement object.

Too bad.

> Option D) : File a feature request against PDO to include a numRows()
> > method,
> > but that is not a quick solution.
>
> That's the same as C for the forseeable future.  Such an addition wouldn't
> be added until at least PHP 5.3, I imagine, which is who knows when.  And
> even at that, it would require a lot of C-level implementation to emulate
> num_rows on all databases, whether they natively support it or not.

So file an issue anyway, but we all won't hold our breath on it.

> I am hoping I can get this into Drupal 6, but time will tell if I can get
> > it ready in time.
> >
> >
> > Ubuntu Feisty (due later in April) will have PHP 5.2.1.
> > http://packages.ubuntu.com/feisty/web/php5
> >
> > But, what about other distros? If they are on PHP4 or 5.1, then this
> will
> > be a show stopper.
>
> I'm not deprecating the existing mysql and pgsql drivers at this
> point.  This is strictly a "if you're using a stable version of PHP 5, then
> you get this extra, faster option" feature.

Good.

Eventually I hope to deprecate the legacy database drivers and just use PDO
> with thin wrappers (for db_query_range() and such) for all database stuff,
> but that's not something we can even consider until 2008 at the
> earliest.  Several web hosts already offer 5.2, though, as do some distros
> (Debian Etch, Ubuntu Feisty, etc.), so for those people the PDO option
> should offer a performance boost on prepared statements.

the problem is that Red Hat and Fedora seem to be more common on hosts than
the Debian/Ubuntu variants (which is my favorite).

As you said, if we are not relying on it solely, then why not include it for
those who
can use it.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.drupal.org/pipermail/development/attachments/20070404/9bd5d98c/attachment.htm 