What are the rules/logic behind persistent connection management when using PDO?
Web Server
DB Server
Non-persistent connections
<?php
// Open a new connection
// Session created in Oracle
$dbh = new PDO('DSN', 'webuser', 'password');
// webuser is active in v$session with a SID=1
$dbh = NULL;
// webuser removed from v$session
// Manually calling $dbh = NULL; will remove the session from v$session
// OR
// Wait for script EOL so a kill-session command is sent to Oracle?
?>
Persistent connections
<?php
// Open a new connection and make it persistent
// Session created in Oracle
// Is Apache maintaining some sort of keep-alive with Oracle here?
// because I thought php.exe is only alive for the duration of the script
$dbh = new PDO('DSN', 'webuser', 'password', array(PDO::ATTR_PERSISTENT => TRUE));
// webuser is active in v$session with a SID=1
$dbh = NULL;
// webuser is still active in v$session with a SID=1
$dbh = new PDO('DSN', 'webuser', 'password', array(PDO::ATTR_PERSISTENT => TRUE));
// webuser is still active in v$session with a SID=1
// Manually calling $dbh = NULL; does not kill session
// OR
// Script EOL does not kill session
// ^^ this is good, just as expected
?>
I visit the page and webuser
gets a SID=1
My colleague visits the page and webuser
gets an additional SID=2
<- rinse, repeat, and increment SID for new computers visiting this page
Shouldn't a new visitor be re-using SID=1
?
All answers, suggestions, requests for alternate testing, links to reading material are welcomed.
I have RTFM'ed for a while and Googling has only produced meager Advantages of Persistent vs. Non-persistent
blogs.
Apache has one parent process. This process creates child processes that will handle any requests coming to the web server.
The initial amount of child processes being started when the web server starts is configured by the StartServers
directive in the apache configuration. The number goes up as needed with a raising amount of requests hitting the web server until ServerLimit
is reached.
If PHP (ran as mod_php, as CGI all resources are freed at the end of script execution) is now being told to establish a persistent connection with a database for a request, this connection is hold even after the script finishes. The connection being now hold is a connection between the apache child process which the request was handled by and the database server and can be re-used by any request that is being handled by this exact child process.
If, for some reason (do not ask me exactly why), the child process is being occupied longer than the actual request and another request comes in, the parent apache process redirects this request to a (new) child process which may has not established a connection to the database up to this time. If it has to during the execution of the script, it raises the SID as you have observed. Now there are two connections be hold by two different child processes of apache.
It is important to know, that this can also cause a lot of trouble. If there is an endless loop or an aborted transaction or some other may be even unpredictable error during the script execution, the connection is blocked and can not be re-used. Also it could happen that all of the available connections of the database are used, but there is another child process of the apache server trying to access the database. This process is blocked for the time being until a connection is freed by the database or apache (timeout or voluntarily by termination). Any further information about this topic on this page: http://www.php.net/manual/en/features.persistent-connections.php
I hope I got all that we have discussed in our comment conversation summarized correctly and did not forget anything. If so, please, leave me a hint and I will add it. :)
Edit:
I just finished reading the article @MonkeyZeus mentioned in this comment. It describes the process I summarized above and provides useful information on how to optimize your apache server to work better together with persistent connections. It can be used with or without oracle database backends, though. You should give a look: http://www.oracle.com/technetwork/articles/coggeshall-persist-084844.html