Over a 24-48 hour period the distribution of load between our 3 node cluster diverges.
After a fresh restart of maxscale everything lines up again.
When exhibiting the odd distribution, using dbadmin to examine the stats on the servers "show servers" shows a "Current no. of operations:" for the effected machine (s) significantly higher than the rest, and the number doesn't change over time. Examining process list on the backend server shows no long running processes to explain the stuck number of operations.