Cache\Adapter\Libmemcached::set() is not working under special circumstances

[zsilbi]

$cache->set() is not working when Libmemcached is used at the same time for cache and modelsMetadata.

class Robots extends \Phalcon\Mvc\Model {}

$di->setShared(
    'cache',
    function () {
        return new \Phalcon\Cache\Cache(
            new \Phalcon\Cache\Adapter\Libmemcached(
                new \Phalcon\Storage\SerializerFactory(),
                [
                    'servers' => [
                        [
                            'host' => 'my-host',
                            'port' => 11211
                        ]
                    ],
                    'prefix'  => 'my-cache-prefix'
                ]
            )
        );
    }
);

$di->setShared(
    'modelsMetadata',
    function () {
        return new \Phalcon\Mvc\Model\MetaData\Libmemcached(
            new \Phalcon\Cache\AdapterFactory(
                new \Phalcon\Storage\SerializerFactory()
            ),
            [
                'servers' => [
                    [
                        'host' => 'my-host',
                        'port' => 11211
                    ]
                ],
                'prefix'  => 'my-metadata-prefix'
            ]
        );
    }
);

Provide minimal script to reproduce the issue

$cache = $di->getShared('cache');

var_dump($cache->get('my-test-key')); // NULL

$testData = ['my-test-data'];

Robots::findFirst();

var_dump($cache->set('my-test-key', $testData)); // bool(true)

This script should run more than once, of course.

$cache->set() always returns true, but fails to save the test data.

It works:

• When MetaData cache is not used
• If Robots::findFirst() is called before $cache->get()
• Without Robots::findFirst()
• Without $cache->get()

Details

• Phalcon version: 4.0.0-alpha.5
• PHP Version: 7.3.4
• Operating System: Ubuntu 16.04
• Installation type: Compiling from source
• Zephir version: Version 0.12.0 (from the development branch)
• Server: nginx

[zsilbi]

The problem is, that two Memached instances are running at the same time using "persistent" connections with same persistentId option set. I am sending a PR to correct this.

[niden]

Resolved in #14106