Hvassing.com

Not long ago, I uploaded latest version of a project I'm working on to my production server at (mt) Media Temple, only to discover that users were kicked out of the system no less than five seconds after logging in.

At first I didn't know what was causing this, so I went through most of my code, searching every tidbit of it in an attempt to find out what was causing this unexpected behaviour. I searched for possible reasons to such an "time out", but couldn't find anything that even seemed remotely plausible. After staring at my code for quite some time, it dawned on me; because of the sheer number of sites hosted at (mt) Media Temple and their (gs) Grid-Server cluster, even with the default settings the sessions initiated would be cleared out and deleted before the users would be able to do anything.

I tried to set the session.gc_probability higher, which didn't work. The same with the lifetime of each session, using the session.gc_maxlifetime, which controls when a session is to be deleted.

The solution

After some more research, I found some rather massive scripts that kept track of sessions using tables and some other really paranoid measures. The solution, however, is actually quite simple, and is stated in the manual as well:

Note: If different scripts have different values of session.gc_maxlifetime but share the same place for storing the session data then the script with the minimum value will be cleaning the data. In this case, use this directive together with session.save_path.¹

Here is what I did

1. Created a directory called sessions one level above my html-catalogue
2. Added the following lines to my scripts
   < ?php
ini_set('session.gc_maxlifetime', 3600); // Valid for an hour
session_save_path('/sessions/'); // Relative to file system, not URI
session_start();
?>

Anyone know of a better way to solve this issue?