(placeholder for further description as we'll release note it in MadHatter and fix it in the future, per discussion on 2019-01-07)
When a synchronous durability request is made, owing to
MB-10291, unrelated operations pipelined on the same socket may be serviced with higher latency. The higher operational complexity of a synchronous durability request will necessarily take longer to coordinate and complete compared to current memory or disk backfill serviced operations.
In failure cases, it can lead to the entire system being effectively unavailable until some kind of recovery, like failover/rebalance.
(more to write...)