Apache1 only resets the write timer, which defaults to 300 seconds, on a successful write. That is, if the client has gone away and Apache attempts a write which fails it will set the conn->aborted flag but not reset the timeout. Assuming the PHP script is running in ignore_user_abort mode we ignore the aborted flag, but we'll still get blown out of the water 300 seconds after the failed write unless we periodically reset the timer. With set_time_limit(0), ignore_user_abort(true) and periodic apache_reset_timeout() calls we can theoretically run forever which is why I disabled this call in safe mode.