3 Catalyst::Manual::Monthly::2012::February::TestDBICWithBellsOn
5 =head2 Testing difficult-to-test database models
7 In this article we're going to describe a technique for testing database-heavy
8 web applications using either a temporary testing database, or using the DSN
9 defined in your Catalyst application depending on the presence of an
10 environment variable. If the latter, we will not delete the contents of the
11 database afterwards, because this workflow suggests that we will want to poke
12 around our application manually with the application in a known state.
14 Basically, providing automated testing of complex databases is a pain. For
15 generic type functions (e.g. the development of libraries rather than
16 applications), mock objects (objects that mimic the interface of a real
17 object) are useful for unit testing. But in the running on the
18 seat-of-your-pants development style that commercial web applications often
19 require, small changes to functionality can wreak havoc with your mock
20 objects, and they rapidly become more trouble than they're worth.
22 Which is where L<Test::DBIx::Class> comes in. The rest of this article
23 will explain how to achieve three different use-cases for Test::DBIx::Class
26 =head2 The easy use case - Inferring the Database from the DBIC result classes.
28 For a straightforward database schema where the L<DBIx::Class> (DBIC)
29 result classes can be used out of the box, one can simply use
30 L<Test::DBIx::Class> to infer the database schema into a temporary
31 database, using a temporary Postgresql instance via
32 L<Test::DBIx::Class::SchemaManager::Trait::Testpostgresql>:
38 use Test::WWW::Mechanize::Catalyst 'MyApp';
40 use Test::DBIx::Class {
41 schema_class => MyApp->model('DB')->schema_class,
42 traits => ['Testpostgresql']
44 'User', 'Adverts'; # only create the tables for the User and Advert Result
47 As an alternative to naming specific tables in the last part of the C<use
48 Test::DBIx::Class> declaration, you can use L<qw/:resultsets/> instead of
49 C<'User', 'Adverts'> in the example avove, to import all Result classes
50 defined in the C<schema_class>.
52 The next thing to produce the appropriate L<Moose> meta-object incantation
53 to swap out the configured C<< MyApp->model('DB')->schema >> with the
54 temporary one we want to use instead (note, this works even when we start
55 doing the clever override things in the next two sections):
57 # make TDBIC schema and app schema the same
58 my $db_meta = MyApp::Model::DB->meta;
59 $db_meta->make_mutable;
60 $db_meta->add_around_method_modifier( schema => sub { Schema() } );
61 $db_meta->make_immutable;
63 Now that we've done this we can start making requests:
65 my $mech = Test::WWW::Mechanize::Catalyst->new;
66 $mech->get('whatever');
69 And the database operations should all really happen, but to a temporary
70 database that gets deleted at the end of the run. This is especially
71 useful if you have lots of tests that all need a pristine copy of the
72 database with their own fixtures, as it means you can speed things up by
73 running in parallel (e.g. to run 3 tests in parallel run C< prove -l -j 3
76 =head2 OK Good. This time let's optionally override the temporary database
77 with the developer's DSN
79 One development style which works fairly well is to write tests to run on
80 the development database, and then have a play around at the end of the
81 test run either with the Perl debugger or using the built in development
82 server. However this means that one can't always rely on having a
83 temporary testing database for running tests.
85 So in this case we use the application's configured database instead. Note
86 this requires a bit more trickery than when we're just using a temporary TDBIC
93 use Test::WWW::Mechanize::Catalyst qw/MyApp/;
97 my %tdbic_args = ( schema_class => MyApp->model('DB')->schema_class,
98 traits => [qw/Testpostgresql/],
103 MyApp->model('DB')->schema_class->storage->connect_info,
104 force_drop_table => 1,
105 keep_db => 1, # assume we want to keep the test db at
106 # the end of the test run
111 # this pattern because we're messing with instantiation in BEGIN
112 require Test::DBIx::Class;
113 Test::DBIx::Class->import(\%tdbic_args, qw/:resultsets/);
115 =head2 Fine, that's the simple cases, what about the harder cases?
117 In many situations it's not desirable to infer the database directly from
118 the DBIC schema classes. While it is possible to put all the metadata (for
119 example including stuff that requires custom database engine extensions)
120 into the DBIC schema, this is not necessarily desirable. For example if you
121 have a process whereby your database schemas are signed off (and likely
122 modified) by a DBA you're likely going to want the master copy of your
123 database in SQL rather than DBIC files. Likewise if you have evil business
124 logic that's best encapsulated in a database trigger, you'll likely hit the
125 same type of problems.
127 Given we're using a Postgres database in this instance, we need some
128 Pg-specific code to spin up either a temporary database or to repopulate the
129 development database. So to complement
130 L<Test::DBIx::Class::SchemaManager::Trait::Testpostgresql>, we've written our
131 own internal C<Test::DBIx::Class::SchemaManager::Trait::DeploySQL> class that
132 should be kept in C< t/lib/Test/DBIx/Class/SchemaManager/Trait/DeploySQL.pm >
133 in your app's directory tree. It's possible this could be released as a CPAN
134 module one day, but at this stage we suspect that every development situation
135 is sufficiently different that it's probably better just to leave these
136 particular bits of wheel lying around for other people to adapt, rather than
137 offering an explicit canned solution that's supposed to work for everybody.
139 Meanwhile here's what we have for our Postgres database populated by SQL statements:
144 before prepare_schema_class => sub {
145 my ($self, $schema_class) = @_;
146 { no warnings 'redefine';
148 *{$schema_class.'::deploy'} = sub { $self->_deploy_sql(@_) };
153 my($self, $schema) = @_;
154 my $port = $self->postgresqlobj->port;
155 my $args = $self->postgresqlobj->postmaster_args;
156 my $storage = $schema->storage;
157 my $app_root = MyApp->path_to();
158 my ($db_name) = $storage->connect_info->[0]->{dsn} =~ /dbname=(.*?)(;|$)/;
159 my ($db_user) = $storage->connect_info->[0]->{user};
160 my @sql_files = qw/list of sql files here/;
162 unless ($ENV{DEV_DB}) {
163 $psql_cmd = "/usr/bin/psql $args -p $port";
164 $storage->dbh_do(sub {
165 system qq{$psql_cmd -U$db_user $db_name -q -c "create language plpgsql"}});
168 $psql_cmd = "/usr/bin/psql";
170 $storage->dbh_do(sub {
171 system "$psql_cmd -U$db_user $db_name -f $app_root/misc/db/$_"})
177 The main thing to note here is that wrapping the C<system> calls in a
178 C<< $storage->dbh_do > call ensures that the database handle from DBI is
179 connected to the database using the failsafe mechanisms in
180 L<DBIx::Class::Schema::Storage>.
182 So finally, deploying to our temporary database using L<Test::DBIx::Class>
183 and either a temporary or a development database from SQL files is done.
184 Now to start up the test file we change the traits in C<%tdbic_args> to
187 traits => [qw/Testpostgresql DeploySQL/],
189 Somewhat intricate, but for complicated development situations definitely
194 Our development team is still working out the best way to use this system,
195 but so far it's been really very handy indeed. What would be good next is
196 to work out how to modularise a lot of the boilerplate above so it can be
197 C<use>d or C<require>d as a single line in each test file. We'll get there
200 =head3 AUTHORS AND COPYRIGHT
202 Words and a little bit of code:
203 Kieren Diment <zarquon@cpan.org>
206 Eden Cardim <edencardim@gmail.com>
210 This documentation can be redistributed it and/or modified under the same terms as Perl itself.