Looks like there are two separate relevant configs: index.translog.interval - How often to check if a flush is needed, randomized between the interval value an

Looks like there are two separate relevant configs: index.translog.interval - How often to check if a flush is needed, randomized between the interval value and 2x the interval value. Defaults to 5s. index.gateway.local.sync - How often the translog is fsynced to disk. Defaults to 5s. http://www.elastic.co/guide/en/elasticsearch/reference/current/index-modules-translog.html http://www.elastic.co/guide/en/elasticsearch/reference/curre... It's probably the flush interval that also defaults to 5 seconds that is responsible for losing data from simple a kill -9 to the process. Before that the data is in the process memory but not the OS buffer.