Tuning memcached options for redundancy & fast failover

memcached-demo.php

<?php

$servers = [
    ['10.0.0.1', 11211],
    ['10.0.0.1', 11212],
];

$id = 1;
$timeout = 50;
$mc = new \Memcached($id);
$mc->setOptions([
    // Assure that dead servers are properly removed and ...
    Memcached::OPT_REMOVE_FAILED_SERVERS => true,

    // ... retried after a short while
    Memcached::OPT_RETRY_TIMEOUT         => 1,

    // KETAMA must be enabled so sthat replication can be used
    Memcached::OPT_LIBKETAMA_COMPATIBLE  => true,

    // Replicate the data, i.e. write it to both memcached servers
    Memcached::OPT_NUMBER_OF_REPLICAS    => 1,

    // Those values assure that a dead (due to increased latency or really unresponsive) memcached server increased
    // dropped fast and the other is used.
    Memcached::OPT_POLL_TIMEOUT          => $timeout,
    Memcached::OPT_SEND_TIMEOUT          => $timeout * 1000,
    Memcached::OPT_RECV_TIMEOUT          => $timeout * 1000,
    Memcached::OPT_CONNECT_TIMEOUT       => $timeout,

    // Those I am not sur. Bnary proto is supposed to be faster. Not blocking sounds legit.
    Memcached::OPT_BINARY_PROTOCOL       => true,
    Memcached::OPT_NO_BLOCK              => true,
]);

// init servers
if (!$mc->getServerList()) {
    foreach ($servers as $server) {
        $mc->addServer($server[0], $server[1]);
    }
}

// write & read
$k = "redundant-counter";
$c = $mc->get($k) ?: 1;
$mc->set($k, $c+1);
print "GET: $c\n";