After setting index.translog.durability=true, the time to successfully respond to the client is whether to respond to the client as long as the primary shard is successful without shutting down the replicas, or to respond to the client after all replicas have responded successfully?
Regardless of index.translog.durability, Elasticsearch responds to a write request after every in-sync shard copy has accepted the write (both primaries and replicas). This setting does not accept the value true, only request or async, but these just determine whether “accepted the write” includes having fsynced it to disk or not.
1 Like