We were discussing the design of an HA environment for Puppet, and what the potential impacts were of a "split-brain" situation. For the model that I was discussing...
...the "worst case" scenario is probably that the two Postgres databases become out-of-sync. This would require a network interruption that would cause:
- the two front-end Puppet masters to lose contact with each other, causing both assume the master role, with the result that pgpool would be running on both nodes and each
puppetdb
process would be talking to it's local pgpool
instance.
pgpool
on one of the nodes to lose contact with the current master database server, causing it to erroneously promote the slave database.
- some puppet clients in the environment to see one master and others to see the other master