I have one index with 100 million docs, on 12 data nodes(with 12 shards), 16G ram, 500G storge each node; the query is filter on biz id(keyword) and date range and amt range; the biggest biz id have 20 million docs, and 95% biz id only have 1000 docs in average; so the problem is when query on the biggest id, the response time take 4+s,the small ids query only need 0.02s, after profile I find breakdown next_doc use the most time, how should I optimize this?
I can not split the biggest id to another index,cuz I split index on month, and I cannot predict future biggest id.