Decreasing the probability of race condition in session lock

[tianhe1986]

When using redis or memcached session driver, if a web-client perform multiple requests at the same time, two or more requests may get the session lock, and finally some of them would generate "Session: Error while trying to free lock for" error log.

Using Redis::setnx and Memcached::add properly could help to decrease the probability of this race condition.

[narfbg]

It is unclear how or why this patch improves anything ... You'll need to elaborate a lot. :)

[tianhe1986]

Take redis driver for example.

When two requests with same cookie arrive at the same time, the two processes may also execute the code to here simultaneously, and get $ttl = -2.
Then they continue to execute $this->_redis->setex($lock_key, 300, time()) and would both succeed, resulting in they all get lock.

However, setnx() or set($key, $value, array('nx')) could be used as a lock. In this situation, only one process could succeed. the others would fail. Refer here

[narfbg]

Ok, that makes sense for Redis, although it can be simplified:

    $result = ($ttl === -2)
        ? $this->_redis->set($lock_key, time(), array('nx', 'ex' => 300))
        : $this->_redis->setex($lock_key, 300, time());

(yes, $result instead of $set_result :))

---

What about memcache though?

[tianhe1986]

memecache also could use add() for locking refer here

Although it also has problem, add() would be better than set() .

I'll modify the code according to your advice :)

[narfbg]

With the method params matching ...

$method = ($this->_memcached->getResultCode() === Memcached::RES_NOTFOUND) ? 'add' : 'set';
if ( ! $this->_memcached->$method($lock_key, time(), 300))

[narfbg]

Needs a single empty line prior to this.