ColinLeeo opened a new issue, #1075:
URL: https://github.com/apache/tsfile/issues/1075

   ## Problem
   
   A time range that covers only part of a chunk (e.g. a 1-day or 1-week query 
on a
   single-chunk-per-month file) marks every page BOUNDARY, so
   `AlignedChunkReader::get_next_page_multi_impl` falls into the row-by-row 
scatter
   loop: one `RowAppender::append` per value plus a per-timestamp filter 
re-check.
   Measured at 8.6% of a 1-week wide-table query, and it made queries reading 
less
   data as slow as a full-month query.
   
   ## Proposed changes
   
   A BOUNDARY page under a contiguous time filter (TimeBetween with
   `not_between=false`) has a fully matching `[row_begin, row_end]` range with 
no
   interior holes, and the page is decoded whole, so its values are row-indexed 
at
   `row_begin` and can take the same bulk memcpy path as FULL_PASS.
   
   - Add `Filter::is_contiguous_time_filter()` (default false; TimeBetween 
returns
     `!not_`).
   - Let `can_bulk` accept such BOUNDARY pages when no value column has nulls.
   - Discrete filters (TimeIn, NOT BETWEEN) keep the slow path.
   
   Implemented in #1072; 1-week wide-table query 1.63s -> ~1.05s.
   
   Related: #1072


-- 
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.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to