Re: Pear DB on mysql replication cluster

Michael Wallner wrote:

Hi Alberto M, you wrote: [...] Hi, I don't think it would make much sense bloating the DB class itself... class ClusterDB { function ClusterDB($master) { $this->master = &DB::connect($master); $slave = $this->master->getOne(' SELECT slave FROM slaves ORDER BY connections ASC LIMIT 1' ); $slavedsn = "mysql://user:pass@$slave/db"; $this->slave = &DB::connect($slavedsn); $this->query(" UPDATE slaves SET connections=connections+1 WHERE slave='$slave'" ); } function query($query) { if (DB::isManip($query)) { return $this->master->query($query); } else { return $this->slave->query($query); } } } What about this? Regards,

nice! i patched a little more class ClusterDB {

    var $master;
    var $slave;
    var $slavename;

    function ClusterDB($master)
    {
        $this->master = &DB::connect($master);
        $this->slavename = $this->master->getOne('
            SELECT slave
            FROM slaves
            WHERE connections<maxconnections
            ORDER BY connections ASC
            LIMIT 1'
        );
        $slavedsn = "mysql://user:pass@$this->slavename/db";
        $this->slave = &DB::connect($slavedsn);

if (DB::isError($this->slave)) $this->slave = $this->master;

        $this->query("
            UPDATE slaves
            SET connections=connections+1
            WHERE slave='$this->slavename'"
        );
    }

    function query($query)
    {
        if (DB::isManip($query)) {
            return $this->master->query($query);
        } else {
            return $this->slave->query($query);
        }
    }

    function close()
    {
        $this->query("
            UPDATE slaves
            SET connections=connections-1
            WHERE slave='$this->slavename' and connections>0"
        );
        $this->master->disconnect();
        $this->slave->disconnect();
    }

} let me try if works... but I don't like too much depending on master to be able to choose and do a connection to a slave, this is why i liked an array solution. on the other side, this solution has the connection number handling that is possible only with databases or other store methods. as I said, we can use a weight system instead of connection limit and defer to mysql server max connections the choice to enable a new connection or not. however we need to check if there are connections available... latest question: is it correct to do something like: ClusterDB extends DB or ClusterDB extends DB_mysql? since i would be able to combine the features of Pear DB, with a transparent master/slave handling. should we go on patching this way? i mean... i'm not so sure it can be interesting for other people. do you think could be interesting to put this one on PEAR package? bye -- "Imagination is more important than knowledge"