Mariadb 10.1 galera cluster and Drupal crashing cluster

I need some help. I assume that since the last post I seen around the 2010 time frame more people have started using drupal with galera cluster, but I am having issues with drupal/drus

First things, the setup

I have a 3 node galera cluster setup behind haproxy and keepalived.
We have 1 prod server (about to cluster into 3) with a staging and 3 dev servers, one for each developer

The obvious. We originally had had drush setup to import data from the production db to the dev server dbs through the haproxy server. The issue is that as ddl statements are causing the TOI to initiate a shutdown. My belief is that by going through the haproxy server the commands are split between the servers, this means that the tables not created on some nodes before the column data is added and therefore TOI is detecting an inconsistency and causing a restart of the node. With only 3 nodes if 2 crashes the cluster starts to cycle to the point where it becomes unstable and 2 nodes completely crashes and leaving the last node in a live, but unworkable state. Shutting down that node does not change the grastate.dat file to have a 1 for safe_to_bootstrap. We used to think it was caused by the --no-ordered-dump but that has not been verified.

To fix this I have had the drush statement import to one "master" node. This seems to fix it though I have seen some crashes still.

This brings me to my current dilemma, with this implemented I still have one problem. Under normal operations the above would not be too much of a concern, as there are not that many major changes to tables that would cause inconsistency, however clearing the cache seems to still cause this crash. At least one table, semaphore, writes and deletes so fast that it seems to cause TOI to kick in again, there may be other tables with this issue but I am not sure. To fix this specific one I changed the engine to MEMORY, but I am not sure the impact this will have on drupal with it not being replicated across the nodes.

so the question is can anyone here tell me how to prevent these TOI issues from arising when using drupal on a galera cluster. My collegues are pusing me to go back to a database per machine setup which is in my opinion unworkable, or have one offs for drupal machines which I dont like.

Thank you in advance.