[Catalyst] dbic models and startup time server

Brandon Black blblack at gmail.com
Tue Aug 22 17:29:19 CEST 2006

On 8/22/06, Josef Chladek <j.chladek at wirtschaftsblatt.at> wrote:

> what we notice: in Catalyst/Model/DBIC/Schema.pm the time of the
> require of the schema
> $schema_class->require
> is fast for the first loaded schemas (0.05s) but takes about 0.6s for
> the last schemas. also time to connect
> $self->schema->connection(@{$self->{connect_info}});
> begins with 0.05s and goes up to 0.6s, so overall the startup process
> takes now about 15s for 19 models/schemas.

Is this timed under mod_perl, or standalone?  If you're timing it under
mod_perl, its probably just getting progressively slower as all of the
copies get started up.  If not, can the increasing times be explained by
increasing schema complexity?  At "$schema_class->require" is when
load_classes runs and you should be mostly seeing
Class::C3/Alg::C3/load_components/etc initialization delays (which is what
I'm still working on reducing, but it may be a little while).  I'm surprised
that ->connection takes any significant time in perl, are you sure this
isn't a case of the DB server responding slower and slower?

If you want to find a spot in one of the public modules to improve
performance, you should profile your models under Devel::Profile to point
out the hotspots.  Make a script which "uses" all of your model classes,
then sets up any config on them that was coming from YAML or other such
sources (connect_info and/or schema class), then calls ->new on each of them
and exits.  Run it w/ "perl -D:Profile testscript.pl", then look at the top
entries in the generated prof.out.  This example is typical of what I
usually see percentage-wise:

%Time    Sec.     #calls   sec/call  F  name
27.67    1.2513      259   0.004831     Algorithm::C3::merge
24.69    1.1163      161   0.006933
 4.52    0.2042      161   0.001269
 3.64    0.1646      161   0.001023
 2.97    0.1343       98   0.001370
 2.75    0.1243     5837   0.000021
 1.93    0.0871      482   0.000181
 1.38    0.0626      183   0.000342     DBIx::Class::Schema::register_source

It was based on this kind of feedback that I focused on Alg::C3 first (its
time is down substantially from what it used to be, and it might not be
possible to make it significantly faster without rewriting it in XS).
Obviously the *_method_dispatch_table methods in Class::C3 are another
target that needs looking at (and again, there's some work in progress on
this, which involves XS code, but it might not be ready until sometime after
YAPC::EU at the least).

-- Brandon
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.rawmode.org/pipermail/catalyst/attachments/20060822/d740c730/attachment.htm 

More information about the Catalyst mailing list