The elastic translog configuration can be set to request (default) or async.
The default translog mode request generates lots of disk IO on elastic nodes.
Before 5.6 this setting was configurable at server level now it needs to be explicitly set at index setting level.
The translog is already configurable for the nuxeo repository index using:
But this is not the case for the 2 others indexes: audit and uid.
This explains why elasticsearch disks are heavily stressed on mass audit write operation.
The optimization elasticsearch.index.translog.durability=async should apply to all Nuxeo indexes (nuxeo, audit and uid).
This async option should not be activated by default.
Note that translog can be changed anytime using curl: