The remaining failure seems genuine (PG93MultiTests.test_failover, which victimizes test_replication). Given 3 nodes, dropping the master and adding a new node at the same time can cause a situation where there is no master. I have not been able to reproduce this locally. I need to trawl the logs from the CI system and try to reproduce manually on a cloud.
The remaining failure seems genuine (PG93MultiTests .test_failover, which victimizes test_replication). Given 3 nodes, dropping the master and adding a new node at the same time can cause a situation where there is no master. I have not been able to reproduce this locally. I need to trawl the logs from the CI system and try to reproduce manually on a cloud.