[GitHub] [incubator-pinot] jasperjiaguo opened a new issue #4461: Proposal for Automated Inverted Index Recommendation

[GitBox <gi...@apache.org>]

jasperjiaguo opened a new issue #4461: Proposal for Automated Inverted Index Recommendation 
URL: https://github.com/apache/incubator-pinot/issues/4461
 
 
   Adding inverted indices to frequently used dimensions is very helpful for cutting down query latency. Currently, whether or not applying an inverted index is purely the decision of system admin. A lot of manual work can go into that and there are several drawbacks:
     1. The problem of choosing a proper set of dimensions to apply inverted indices can be very time consuming. A system admin has to pull up the logs and scan through the queries for frequent used dimensions, which will be impossible at high QPS or with multiple use cases onboard. 
     2. Furthermore, this approach does guarantee the inverted indices used are optimal or targeting the right queries (say, one wants to decrease the latency slow running queries with >90th% latency)
    3. There is an on average 100% storage penalty for adding inverted index to a specific column. And there is a non-zero cost of applying inverted index in the filtering phase, meaning there is a "sweet spot" for the number of inverted index.
    4. For a lot of use cases with and-connected predicates, it is important to pick the one with best selectivity, which has the best effect in cutting down the time spent in lookup phase.
   
   Based on these points, we want to develop an automated process for inverted index recommendation.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services

---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@pinot.apache.org
For additional commands, e-mail: commits-help@pinot.apache.org