I’m wondering if I really loose data of at most `index.translog.sync_interval`

when a node with `index.translog.durability=ASYNC`

crashes **and** there’s at least one replica.

My understanding is that Elasticsearch only answer with HTTP 2XX once the data is *stored* on all replicas. In case of a catastrophic failure on a node, I expect that the missing data can be recovered from the remianing replica(s). A data loss will only occur (assuming there is exactly 1 replica) when two nodes fail within `index.translog.sync_interval`

. Is this correct?