Conversation

-- Quick question that I didn't find in your documentation. When documents are requested directly by their _id, is that operation an O(1) operation?

Replying to

and

not sure the O notation makes much sense in a distributed system. you'll have a mix of network hops (coordinating node, primary / replica shard), memory and potentially disk access the question might make more sense for lucene. blog.mikemccandless.com/2014/05/choosi is a great starting point

Replying to

and

That's a good point but I was curious from the standpoint of having one node and searching on that node. It's probably a better question to direct at Lucene. I was going to check with the Lucene team anyway on this: discuss.elastic.co/t/sorting-by-t The sort index feature is awesome, but

Replying to

and

it seems odd that it is so much slower if the sort is in the opposite direction. Thanks for your help!

5:32 PM · Oct 12, 2018Twitter Web Client

Replying to

and

Generally there are so many optimizations in Lucene that an O notation might depend on a lot of factors (like the structure of the ID). Index sorting is another special case but I'm surprised by your outcome. I'd have expected it to be similar (without knowing the code)

Replying to

and

I agree with you on the Lucene sort issue. It might be worth me going through Lucene's tickets and checking if there is anything already flagged but it's probably at least worth bringing up -- even if they have an explanation why (it would be nice to know why it is). Thanks!!

Show replies