Suspecting this problem has occurred for "a few months" is interesting. To try to be a little more precise, would you say that means 2 or 3 months, 5 or 6 months, or perhaps longer?
I think what you are experiencing is a problem with exclusive file access locks on the session file when multiple server hosts are involved. Since an flock function is built into PHP I assume session_start() uses that for setting its locks.
For a while I've had a similar problem, but with flock in Perl. In my case the lock is used to prevent two or more independent visitors to a page causing a file to be updated at the same time (and overwriting each others updates). In my case I think each visitor is likely to be served from a different host in the cluster.
At the moment Plusnet and Hostopia are playing ticket tennis with the support query. Hostopia failed to find any system change around the time the problem first occurred. Moreover they claim flock works with myself supplying hard evidence (even using their test case) that it doesn't. The problem manifests itself as 30 second delays in my tests for something that should complete in milliseconds. Necessarily Plusnet support are the intermediary between me and Hostopia but I must emphasise that Plusnet support has been exemplary through this saga.
My problem is with the Perl implementation so having a similar one with PHP is very useful. To this end if would be most helpful if you could create a simple test page showing test date and time, page loading times and the server hostnames involved.
David
David
