A few comments from a quick scan:
There is, as far as I can see, no mention of oversharding, and this is pretty much the #1 problem we see with users. I think it's worth a mention and maybe a link to a blog post like this one.
Start first by setting
index.translog.durabilitytoasync.
Please don't recommend this. It will cause less experienced users to experience data loss. The default durability setting is much safer, and normally performs just fine on good hardware.
In fact I think that whole paragraph on translog tuning is a little misleading for a "survival guide". There are lots of other things I'd look at for performance gains before turning to these settings.
Adapt
index.refresh_intervalto your needs.
It might be best to leave this setting unset too. In recent versions, if you're indexing but not searching then there will be no refreshes taking place. From the docs:
If this setting is not explicitly set, shards that haven’t seen search traffic for at least
index.search.idle.afterseconds will not receive background refreshes until they receive a search request.
Compare-and-swap over
_versionfield is poor man’s transactions
The preferred CAS operation uses _primary_term and _seq_no, since _version has known issues.