By the way, the script runs without warning if not using persistent connection.

Also, when using persistent connection, the script works fine (as intended) even though there is a warning in the log file.

No it is not a bug, 98% quite sure about that.

COM_CHANGE_USER is a re-authentication command send from mysqlnd to the MySQL server. COM_CHANGE_USER is there to check that the user who wants to use the persistent connection is a) still allowed to connect to MySQL and b) it resets all session related data (reset user variables, close open transactions and so on - the MySQL manual has details on it).

ext/mysqli @ mysqlnd does send a COM_CHANGE_USER unlike ext/mysql. ext/mysql does not properly reset persistent connections, ext/mysqli does. All ext/mysql does is send a ping to check if the connection is still alive. If you want the ext/mysql behaviour, recompile PHP with -DMYSQLI_NO_CHANGE_USER_ON_PCONNECT .

I assume that every couple of minutes you hit a closed (timeout, whatever...) connection in the pool connection. mysqlnd tries to send a COM_CHANGE_USER, fails and bails. ext/mysqli recognizes the failure and continues as it should. Its the same procedure as with ext/mysql and a failing ping - just a bit more verbose...

What MySQL version are you running? Very old and very early 5.1 (<5.1.10 or something like that) servers had issues with COM_CHANGE_USER.

I'm using MySQL 5.1.37.

I have 2 variables set in mysqld that might have affected this:
wait_timeout = 300
interactive_timeout = 300

Are these too low and do they have any connection with the PHP warning?

5.1.37 should be fine to my knowledge.

The timeouts should be described in the MySQL manual. Their values appear "normal" to me. Please ask on one of the discussion lists what value may be best - in the end this is a bug system not a discussion list. 

As said, I believe that you get the warning because you occasionally hit a connection that has timed out or similar - nothing to worry about: PHP simply picks the next connection from the pool or opens a new one. The same can happen with ext/mysql. The only difference is that ext/mysqli @ mysqlnd is a bit more verbose...

The only question here is if a) libmysql also barks and b) if we want that warning or not. Maybe the warning should be suppressed because its handled internally.

Assigning to myself to discuss with Andrey and Johannes.

I have comittet a change to silence CHANGE_USER when reusing pooled persistent connections. The warning should no longer appear. 

The change should either make it into 5.3.1 or 5.3.2 - that is up to the release manager to decide. It is not a critical change.