I just learned of the preference argument which can be passed to search queries.
If we specified a custom string on all queries, I think it would force the nodes to specialize on particular data instead of trying to maintain the exact same shard search metadata in memory. Given our heavy data locality on the current date, I'm not sure it'd actually make much of a difference for most queries.
The other ones which looked intriguing were _shards and _prefer_node, but I haven't formulated a particular idea on that yet. It'd be cool if it supported targeting by node attributes.
I just learned of the
preference
argument which can be passed to search queries.If we specified a custom string on all queries, I think it would force the nodes to specialize on particular data instead of trying to maintain the exact same shard search metadata in memory. Given our heavy data locality on the current date, I'm not sure it'd actually make much of a difference for most queries.
The other ones which looked intriguing were
_shards
and_prefer_node
, but I haven't formulated a particular idea on that yet. It'd be cool if it supported targeting by node attributes.Going to think about this further...