Hi,

I have setup a cluster of 3 machines with following tags:

mydb/1/master, mydb/2/master, mydb/3/master

On a fourth machine I have created a multiplex funnel as follows:

monetdb create -m monetdb+monetdb@*/mydb/1/master,monetdb+monetdb@*/mydb/2/master,monetdb+monetdb@*/mydb/3/master mydbmaster

My objective here is that I use this multiplex for only DDL/INSERTS/UPDATES.

This would ensure replication on all the 3 masters thus achieving a hacked form of master-master replication.

Then I could use any one of the masters for analytics querying.

However I am stuck with the following issues while handling fail-over.

1. I expect that if one of the boxes die, others keep getting the updates until I realize the failure. The I could lock the multiplex, sync the dbfarms to catch up the failing db and then release the multiplex.

2. If the box with tag mydb/3/master dies, then the updates go to 1 and 2. But if mydb/1/master dies the none of the dbs get any more updates. Looks like the multiplex returns on the first failure without trying rest of the dbs.

Any pointers as to where do I change this sequential failure logic?

Also what do you think about my replication approach?

Thanks and Regards,

Tapomay