QA » openQA Project

The insertion ends up being the following SQL:

    <- prepare_cached('INSERT INTO comments ( job_id, t_created, t_updated, text, user_id) VALUES ( ?, ?, ?, ?, ? ) RETURNING id', HASH(0x5632f072e6f0), ...)= ( DBI::st=HASH(0x5632f1dc41b0) ) [1 items] at DBI.pm line 1972
    <- bind_param(1, 2820, ...)= ( 1 ) [1 items] at DBI.pm line 2013
    <- bind_param(2, '2022-05-25 12:39:10', ...)= ( 1 ) [1 items] at DBI.pm line 2013
    <- bind_param(3, '2022-05-25 12:39:10', ...)= ( 1 ) [1 items] at DBI.pm line 2013
    <- bind_param(4, "bar", ...)= ( 1 ) [1 items] at DBI.pm line 2013
    <- bind_param(5, 2, ...)= ( 1 ) [1 items] at DBI.pm line 2013

So the timestamps are already computed on the Perl-side. The relevant Perl module is DBIx::Class::Timestamps (in our repository but utilizing https://metacpan.org/pod/DBIx::Class::InflateColumn::DateTime, https://metacpan.org/pod/DBIx::Class::TimeStamp and https://metacpan.org/pod/DBIx::Class::DynamicDefault). Judging by the code I'd say the race condition isn't a surprise:

sub add_timestamps {
    my $self = shift;

    $self->load_components(qw(InflateColumn::DateTime DynamicDefault));

    $self->add_columns(
        t_created => {
            data_type => 'timestamp',
            dynamic_default_on_create => 'now'
        },
        t_updated => {
            data_type => 'timestamp',
            dynamic_default_on_create => 'now',
            dynamic_default_on_update => 'now'
        },
    );
}

sub now {
    DateTime->now(time_zone => 'UTC');
}

The function specified via dynamic_default_on_create is most likely called individually per column and obviously sub now might return different values. So I suppose a real fix needed to be done in https://metacpan.org/dist/DBIx-Class-DynamicDefault/source/lib/DBIx/Class/DynamicDefault.pm.