You are viewing a plain text version of this content. The canonical link for it is here.

Posted to notifications@iotdb.apache.org by "周沛辰 (Jira)" <ji...@apache.org> on 2023/01/10 07:46:00 UTC

[jira] [Created] (IOTDB-5392) Accelerate PointPriorityReader in fast compaction

周沛辰 created IOTDB-5392:
--------------------------

             Summary: Accelerate PointPriorityReader in fast compaction
                 Key: IOTDB-5392
                 URL: https://issues.apache.org/jira/browse/IOTDB-5392
             Project: Apache IoTDB
          Issue Type: Improvement
            Reporter: 周沛辰
            Assignee: 周沛辰
             Fix For: master branch


In the fast compaction, the overlapping pages will be put into the pointPriorityReader, and the points are spit out in order. Every time a point is spit out, an element must be popped out from the priority queue, and then added into the queue. When the points do not overlap, it will cause a lot of invalid heap sort, which will reduce the compaction rate. For example: there are two pages, one is a sparse page with a range of 100 to 500, which has only  two points (100 and 500). The other is a  dense page with a range of 0 to 400. the data of 0 to 99 and 101 to 400 can be spit out directly, instead of adding into the queue and popping them out every time.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)