MySQL Trigger Replication between Master and Slave

we are running a rather highly loaded MySQL Application replicating between Master and two Slave instances. We are using Statement-based replication.

Recently, we have upgraded our hardware and a very strange problem appeared.

On Master (and slave(s)), there is an after_insert and an after_update trigger defined on Table A. Every time the trigger is fired it inserts a "log" entry into a separate table. Below is the Trigger Code:

    DELIMITER |
CREATE TRIGGER orders_after_insert AFTER INSERT ON orders
FOR EACH ROW 
BEGIN   
    DECLARE EXIT HANDLER FOR SQLEXCEPTION
        RESIGNAL;
    DECLARE EXIT HANDLER FOR SQLWARNING
        RESIGNAL;
    DECLARE EXIT HANDLER FOR NOT FOUND
        RESIGNAL; 

    IF NEW.company_id = 1 THEN
        INSERT INTO changelog SET
           company_id  = NEW.company_id, 
           dbtable     = 'orders',
           row_id      = NEW.order_id,
           created     = NOW(),
           action      = 'INSERT';       
    END IF;
END;
|

DELIMITER ;

After our hardware upgrade we now face the following problem: every now and then, the trigger will NOT fire on Master, it fires okay on both slaves. As a result we have inconsistent data between Master and Slave.

• All of this is always happening inside a transaction!
• The changelog table the trigger is inserting into does NOT have a primary key. Adding an auto-increment column to the table is not an option, as it makes the trigger unsafe for replication (MySQL Warning)
• Both tables are InnoDB Tables

I am already out of ideas how to further debug this problem. It is reproducible on our production system only. Happens randomly and rarely. How can this be analyzed? How can I dig to the root of the problem?

I will be really grateful for any inputs!