Implement create-before-destroy when migrating NVME instances
When triggering a cross-region migration Atlas immediately degrades a cluster by going to N-1 nodes prior to creating a replacement node in the new region.
When using NVME instances, this violates the agreed upon replication depth. Once the NVME instance in the source region is destroyed, there are only N-1 copies of the data left, as there is no EBS backing.
NVME target instances should be brought up as {hidden=true, priority=0} until leaving STARTUP2 and entering SECONDARY. Once this occurs, the target instance should be brought to {hidden=false, priority=P}, and a source instance hidden, removed after a quiescence period, and destroyed.
2
votes
Ian
shared this idea