Corion, that is a great example for demonstrating an input iterator with MCE. A global dbh variable is not necessary. Below, the db_iter closure is called by the manager process each time a worker requests the next input item.

Pretty much everything is wrapped inside db_iter for db apps. The closure simply returns the next row.

use MCE::Loop chunk_size => 1, max_workers => 4;
use DBI;

sub db_iter {
   my $dbh = DBI->connect("dbi:SQLite:dbname=sample.db", "", "", {
      PrintError => 0, RaiseError => 1, AutoCommit => 1,
      FetchHashKeyName => 'NAME_lc',
   });

   my $sth = $dbh->prepare("select fname, lname from people");
   $sth->execute();

   return sub {
      if (my @row = $sth->fetchrow_array) {
         return @row;
      }
      return;
   }
}

mce_loop {
   my ($mce, $chunk_ref, $chunk_id) = @_;
   my ($fname, $lname) = @{ $chunk_ref };

   MCE->say("Hello, $fname $lname");

} db_iter();
[download]

See https://metacpan.org/pod/MCE::Core#SYNTAX-for-INPUT_DATA if chunking is desired. The db_iter example, on the page, is written using the core API. Below same example using MCE::Loop instead.

use MCE::Loop;
use DBI;

sub db_iter {
   my $dsn = "DBI:Oracle:host=db_server;port=db_port;sid=db_name";

   my $dbh = DBI->connect($dsn, 'db_user', 'db_passwd') ||
             die "Could not connect to database: $DBI::errstr";

   my $sth = $dbh->prepare('select color, desc from table');
   $sth->execute();

   return sub {
      my $chunk_size = shift;

      if (my $aref = $sth->fetchall_arrayref(undef, $chunk_size)) {
         return @{ $aref };
      }

      return;
   }
}

## Let's enumerate column indexes for easy column retrieval.
my ($i_color, $i_desc) = (0 .. 1);

MCE::Loop::init {
   max_workers => 3, chunk_size => 1000,
   input_data => db_iter(),
};

mce_loop {
   my ($mce, $chunk_ref, $chunk_id) = @_;
   my $ret = '';

   foreach my $row (@{ $chunk_ref }) {
      $ret .= $row->[$i_color] .": ". $row->[$i_desc] ."\n";
   }

   MCE->print($ret);
};

MCE::Loop::finish;
[download]

But one thing is missing from the 2 examples above. Perhaps, workers also need to communicate with the DB. The user_begin and user_end options are where workers obtain a db connection and disconnect. The dbh handle is stored in the MCE hash for later retrieval by the loop block. Thus, the db connection is obtained once.

MCE::Loop::init {
   max_workers => 3, chunk_size => 1000,
   input_data => db_iter(),

   user_begin => sub {
      my ($mce) = @_;
      my $dsn = "DBI:Oracle:host=db_server;port=db_port;sid=db_name";
      $mce->{dbh} = DBI->connect($dsn, 'db_user', 'db_passwd') ||
          die "Could not connect to database: $DBI::errstr";
   },

   user_end => sub {
      my ($mce) = @_;
      $mce->{dbh}->disconnect;
   },
};

mce_loop {
   my ($mce, $chunk_ref, $chunk_id) = @_;
   my $dbh = $mce->{dbh}; my $ret = '';

   foreach my $row (@{ $chunk_ref }) {
      $ret .= $row->[$i_color] .": ". $row->[$i_desc] ."\n";
   }

   MCE->print($ret);
};
[download]