Details
Description
The partition-failover workload was designed to demonstrate dataloss of comfirmed observe-based majority writes on pre-6.5.0 clusters, due to a failure to promote the replica with highest seqno. This was implemented by introducing a partition such that (for some vbuckets) the active node could not communicate with replica 1, then, after some time, failing over the active. Writes during the partition would succeed as active + replica 2 form a majority, but upon failover the mutations are lost when replica 1 is promoted.
This behaviour was completely broken by commit 4d2ecb1 which erroneously changes the behaviour of both the partition and the failover stage.
Note that this workload is also prevented from demonstrating the failure of observe based behaviour due to MB-35292.