Mojo::Pg and migrations (strage behavior)

[Andrey Khozov]

I have the following code:

use Mojo::Pg;

use Mojolicious::Lite;

use Test::Mojo;

use Test::More;

helper pg => sub { state $pg = Mojo::Pg->new('postgresql://user:te...@127.0.0.1/test') };

app->pg->migrations->migrate;

get '/test' => sub {

  my $c = shift;

  $c->render_later;

  $c->delay(

    sub {

      my $delay = shift;

      $c->pg->db->query('insert into test (ts) values (now())' => $delay->begin);

    },

    sub {

      my ($delay, $e, $res) = @_;

      return $c->render_exception("Error while insert: $e") if $e;

      $c->render(text => "Ok\n");

    }

  );

};

app->start;

my $t = Test::Mojo->new;

$t->get_ok('/test')->content_is("Ok\n");

$t->get_ok('/test')->content_is("Ok\n");

done_testing();

And I have strange behavior (timeout for curl and blocked ioloop for morbo) with second query to server.

If I remove the migration line, everything works well.​ Tests also work well.

I tried to use different postgres database (9.3 and 9.4RC), perl (5.18 and 5.20) and last version of Mojo::Pg.

Can someone confirm this behavior?

--

Andrey Khozov

[sri]

Can someone confirm this behavior?

https://github.com/kraih/mojo-pg/blob/master/t/pg_lite_app.t

--

sebastian

[Andrey Khozov]

I think that for reproduce a bug you need to make a new connection to the server and perform at least two request.

https://gist.github.com/avkhozov/220f081d68422a7629be

--

Andrey Khozov

[sri]

I think that for reproduce a bug you need to make a new connection to the server and perform at least two request.

https://gist.github.com/avkhozov/220f081d68422a7629be

I can confirm the problem, and i suspect it might be a DBD::Pg bug. I have traced the problem to a $dbh->ping call, which appears to just block forever.

    https://github.com/kraih/mojo-pg/blob/v1.02/lib/Mojo/Pg.pm#L71

--

sebastian 

[sri]

I have traced the problem to a $dbh->ping call, which appears to just block forever.

    https://github.com/kraih/mojo-pg/blob/v1.02/lib/Mojo/Pg.pm#L71

There appears to be no connection to migrations though, i can replicate the problem with and without them. I have no idea why $ua->max_connections(0) matters though, that part is very strange.

--

sebastian 

[sri]

That was a really fun bug to hunt! And it should be fixed now.

    https://github.com/kraih/mojo-pg/compare/b94da8531e98a9da9197dfb8cf29e4591ad9c8b4...4a3d2e43f6a20f216d93f5b2521715faf65e9bfd

Turns out DBD::Pg is innocent after all, and it was just a little bit of Perl magic. We wrap the DBD::Pg file descriptor into a IO::Handle object, to be able to use it with Mojo::IOLoop.

    https://metacpan.org/source/GBARR/IO-1.25/lib/IO/Handle.pm#L327

But when the last reference to this wrapper is lost, Perl automatically closes the underlying file descriptor. (Although $dbh->ping blocking forever might still be a bug, i would expect for it to be able to detect the closed file descriptor)

--

sebastian

[Andrey Khozov]

Great! Thanks!