Anonymous Monk has asked for the wisdom of the Perl Monks concerning the following question:

wize monks!

my scenario:
server#1 runs N instances of application.pl (just executeable perl files, not via apache/mod_perl)

during runtime application.pl connects to a mysql-database at server#2 and executes several INSERT statements.

this works well - as long as there are not many instances of application.pl running. but the load increases from time to time, which leads to a bottleneck because of the many connections.

upon my quest to reduce connection traffic i came up with the idea of sharing only one mysql connection between the applications (like Apache::DBI). But because i don't start the application via apache, httpd is not the mother-instance...

as far i found ResourcePool::Command::DBI::Execute, which i would like to serve to the application.pl instances in some way.

or do i need something like a deamon.pl to which the several application.pl can connect and write their insert-statements?
but how to keep the several requests seperated so that only one INSERT takes place at a time. like a buffer for several insert statements...

...hmmm...???...


can you enlighten an unworthy perl student with knowledge, please?!
  • Comment on how can N client.pl connect to one deamon serving mysql

Replies are listed 'Best First'.
Re: how can N client.pl connect to one deamon serving mysql
by duckyd (Hermit) on Feb 22, 2006 at 00:33 UTC
    MySQL offers locking - each application.pl can lock the table, insert, and then release their lock to allow another application.pl to insert. Note that only 1 application.pl can be inserting at any time. This will not, however, help with your desire to share connections to the database.
      right. locking doesn't help in this case. ;-)

      however - isn't there a possibility to
      a) generate something like "a socket for each local connection attempt" (ResourcePool::Factory ??) within the deamon.pl, which collects the different application.pl's data and
      b) pump the data from the different socket connections to a "general/shared queue" in deamon.pl

      like this pseudo-code deamon.pl:
      $queue = generate_global_ressource(); create_socket_pool(max_sockets = 200); while(1=1){ if($input=connection_attempt) { if($input ne 'finish') { push($fetched_data_from_socket, queue); } else { close_socket(); create_new_socket_in_socket_pool(); } }
      (where generate_global_ressource() is something like ResourcePool::Command::DBI::Execute)

      because all i need that deamon for are INSERT statements, the applications don't need to wait for return values. they should only "dump" the sql statements to a socket without caring what happens. (also there is a quite generous error-tolerance here: i don't care for lost data, as long as it doesn't exceed 5-10%...)

      i still don't get:
      a) how to keep a number of opened sockets in stock for the different application.pl which all use THE SAME port; and
      b) how to use deamon.pl's general $queue AS a general queue for the several connection instances...

      jdporter fixed code formatting and CPAN links

Re: how can N client.pl connect to one deamon serving mysql
by saxman25 (Initiate) on Feb 22, 2006 at 15:17 UTC

    may be sth like this will work?!

    #!/usr/bin/perl

    use IO::Select;
    use IO::Socket::INET;

    use ResourcePool;
    use ResourcePool::Factory::DBI;
    use ResourcePool::Command::DBI::Execute;


    # Initialize parameter ------------------------------------------------

    $debug= 1;
    $main = new IO::Socket::INET (
          LocalHost => '127.0.0.1',
          LocalPort => 6000,
          Listen => 5,
          Proto => 'tcp',
          Reuse => 1 ) || die $!;
    $zero = chr(0);
    $/ = $zero;
    $\ = $zero;
    $| = 1;
    my $dsn = "DBI:mysql:DBNAME;your.server#2;3306";
    my $username = "xxx";
    my $passwd = "zzz";

    # Initialize Ressource ------------------------------------------------

    my $factory = ResourcePool::Factory::DBI->new(
          $dsn,
          $username,
          $passwd);
    my $pool = ResourcePool->new($factory, MaxTry => 3);


    # Initialize IO::Select ------------------------------------------------

    $handles = new IO::Select();
    $handles->add($main);

    print "Starting listening cycle\n" if($debug > 0);

    LISTEN: while (1)
       {
          ($pending) = IO::Select->select($handles, undef, undef, 60);
          foreach $sock (@$pending)
          {
             if ($sock == $main)
             {
                $num++;             print "Got new connection: $num from ".$sock->sockhost()."\n" if($debug > 0);
                my $newsock = $sock->accept();
                $newsock->autoflush();
                $number{$newsock} = $num;
                $handles->add($newsock);
             } else {
                my $buf = <$sock>;
                if ($buf)
                {
                   print "Existing socket $number{$sock} is pending: " if($debug > 0);
                   chomp $buf;
                   print "$buf\n" if($debug > 0);
                   my $cmd = ResourcePool::Command::DBI::Execute->new();
                   $pool->execute($cmd, $buf);
                } else {
                   print "Socket $number{$sock} is gone.\n" if($debug > 0);
                   $handles->remove($sock);
                }
             }
          }
       }

      i tried this with

      my $port = 6000;
      my $host = 'localhost';
      my $proto= getprotobyname('tcp');
      my $iaddr = inet_aton($host);
      my $paddr = sockaddr_in($port, $iaddr);

      my $code = 'INSERT INTO test (r1, r2, r3) VALUES (1,2,3);\n';

      socket(SOCK, PF_INET, SOCK_STREAM, $proto) || die "socket: $!";
      connect(SOCK, $paddr) || die "connect: $!";
      print SOCK $code;
      close (SOCK) || die "close: $!";


      but after one connection attempt succeeded, the second one kills the deamon.pl:

      linux:/srv/www/system # ./deamon.pl &
      1 7336
      linux:/srv/www/system # ./client.pl
      linux:/srv/www/system # ./client.pl
      1+ Stopped ./deamon.pl
      linux:/srv/www/system #


      ...strange... :-( ...i use a logfile for the data the deamon.pl receives, and the first connections data was logged; but not the second ones....
Re: how can N client.pl connect to one deamon serving mysql
by NiJo (Friar) on Feb 22, 2006 at 17:52 UTC
      it does, indeed.

      in fact - the INSERTS i use are DELAYED ones anyway...

Re: how can N client.pl connect to one deamon serving mysql
by NiJo (Friar) on Feb 23, 2006 at 20:17 UTC
    This is the guide to speed. I'm afraid you did not use placeholders, prepare() and execute.

    On the other hand: Is DBI the right tool for the job? Mass inserts (as you seem to be doing) are best left to the LOAD DATA command. Maybe you ftp the files over and have the daemon insert them locally.

      hmm.. problem is - the next stage will be a bidirectional communication, because i need the lastinsertid.

      i already switched back from INSERT DELAYED to plain INSERT. like i just posted, it works in a way :-)

      if i can handle the local connections in a multithreaded style, a "usual" db_connect should also be possible... i think...
Re: how can N client.pl connect to one deamon serving mysql
by Anonymous Monk on Feb 23, 2006 at 20:24 UTC
    thanx to saxman!
    i modified the deamon.pl to:

    #!/usr/bin/perl -w

    use strict;
    use warnings;

    use IO::Socket;
    use ResourcePool;
    use ResourcePool::Factory::DBI;
    use ResourcePool::Command::DBI::Execute;

    # Initialize parameter ------------------------------------------------
    my $port = 6000;
    my $dsn = "DBI:mysql:DBname;DBserver;3306";
    my $username = "xxx";
    my $passwd = "yyy";

    # Initialize Ressource ------------------------------------------------
    my $factory = ResourcePool::Factory::DBI->new(
          $dsn,
          $username,
          $passwd);
    my $pool = ResourcePool->new($factory, MaxTry => 3);

    # Initialize Socket ------------------------------------------------
    my $sock = new IO::Socket::INET(
          LocalHost => 'localhost',
          LocalPort => $port,
          Proto => 'tcp',
          Listen => SOMAXCONN,
          Reuse => 1);
    $sock or die "no socket :$!";

    my($new_sock, $c_addr, $buf);
    while(1==1)
    {
       while (($new_sock, $c_addr) = $sock->accept())
       {
          my ($client_port, $c_ip) =sockaddr_in($c_addr);
          my $client_ipnum = inet_ntoa($c_ip);
          my $client_host =gethostbyaddr($c_ip, AF_INET);
          my $code = '';
          while (defined ($buf = <$new_sock>))
          {
             $code .= $buf;
          }
          my $cmd = ResourcePool::Command::DBI::Execute->new();
          $pool->execute($cmd, $code);
       }
    }
    exit;

    __END__


    Also i used a skript which forked 100 instances of the client.pl each pushing a ~7k INSERT statement to the deamon.pl; this worked fine - all 7000 entries arrived at the DB server and were processed.

    it took around 5 secs for the database to execute all statements. that brought me to the following problem:

    if there are a lot of client scripts running, the queue will fill up and the clients will keep waiting for the deamon if its queue is full. so the traffic-problem i tried to solve just shiftet to a local connection bottleneck... :-(

    is there a possibility to do the socket stuff multithreaded?? each thread with its own socket but using the same sql-queue???
      if you
      - just send 1 statement at a time and
      - then disconnect and reconnect for the next one,
      - each transmission terminate by $EOL
      you could try this deamon, which seems to work for several hundred parallel client attempts. (i brute forced it up to 700..800x8kB - took 1.5 min, but the first 500 went through within 20..30sec or so - but with just 100 at a time youre done in about 10 sec. so depending on your average and peek load you have to monitor the runtime of the clients. just do several scenarios for your use and see if there is still a bottleneck...) :


      #!/usr/bin/perl -Tw

      use strict;
      use warnings;
      use IO::Socket;
      use Net::hostent;
      use ResourcePool;
      use ResourcePool::Factory::DBI;
      use ResourcePool::Command::DBI::Execute;

      my $EOL = "\015\012";
      sub spawn;
      my $port = 6000;
      my $proto = getprotobyname('tcp');
      my $dsn = "DBI:mysql:DB;HOST;3306";
      my $username = "xxx";
      my $passwd = "yyy";

      my $factory = ResourcePool::Factory::DBI->new($dsn,$username,$passwd);
      my $pool = ResourcePool->new($factory, MaxTry => 3);

      socket(Server, PF_INET, SOCK_STREAM, $proto) || die "socket: $!";
      setsockopt(Server, SOL_SOCKET, SO_REUSEADDR, pack("l", 1)) || die "setsockopt: $!";
      bind(Server, sockaddr_in($port, INADDR_ANY)) || die "bind: $!";
      listen(Server,SOMAXCONN) || die "listen: $!";

      my $waitedpid = 0;
      my $paddr;
      $SIG{CHLD} = 'IGNORE';

      for ($waitedpid = 0; ($paddr = accept(CLIENT,Server)) || $waitedpid; $waitedpid = 0, close CLIENT)
      {
         next if $waitedpid and not $paddr;
         my($port,$iaddr) = sockaddr_in($paddr);
         my $name = gethostbyaddr($iaddr,AF_INET);

         spawn sub {
            $|=1;
            my $line; my $clientdata = '';
            for (;;)
            {
               undef $!;
               unless (defined($line= <> ))
               {
                  die $! if $!;
                  last; # reached EOF
               }
               $clientdata .= $line;
               last if eof;
            }
            my $cmd = ResourcePool::Command::DBI::Execute->new();
         $pool->execute($cmd, $clientdata);
         # and here may be a possibility to send data back to the client (f.i. lastinsertid)
         # <> = "place return data here $EOL";
         };
      }

      sub spawn
      {
         my $coderef = shift;
         unless (@_ == 0 && $coderef && ref($coderef) eq 'CODE')
         {confess "usage: spawn CODEREF";}
         my $pid;
         if (!defined($pid = fork))
         {return;} elsif ($pid) {return; # I'm the parent}
         # else I'm the child -- go spawn
         open(STDIN, "<&CLIENT") || die "can't dup client to stdin";
         open(STDOUT, ">&CLIENT") || die "can't dup client to stdout";
         exit &$coderef();
      }
        cool! thanx a lot!!

        now i go and try to implement the bidirectional communication. seems to work so far... *smile*