Assuming we have two data copies in each DC and failover not triggered. For a key has data changed in remote DC, but two nodes which hold both primary and replica for this key fail. If XDCR can still reach N-2 nodes, will it remember the checkpoint for 2 down nodes only and resume missing update of the key once two remote nodes up again?
Yes – XDCR has separate checkpoints for each vBucket. As long as the vBuckets and the documents are recovered, XDCR should resume accordingly.