Uploaded image for project: 'HBase'
  1. HBase
  2. HBASE-20525 Refactoring the code of read path
  3. HBASE-19034

Implement "optimize SEEK to SKIP" in storefile scanner

    XMLWordPrintableJSON

Details

    • Sub-task
    • Status: Closed
    • Major
    • Resolution: Won't Fix
    • None
    • None
    • None
    • None

    Description

        protected boolean trySkipToNextRow(Cell cell) throws IOException {
          Cell nextCell = null;
          do { 
            Cell nextIndexedKey = getNextIndexedKey();
            if (nextIndexedKey != null && nextIndexedKey != KeyValueScanner.NO_NEXT_INDEXED_KEY
                && matcher.compareKeyForNextRow(nextIndexedKey, cell) >= 0) { 
              this.heap.next();
              ++kvsScanned;
            } else {
              return false;
            }    
          } while ((nextCell = this.heap.peek()) != null && CellUtil.matchingRows(cell, nextCell));
          return true;
        }
      

      When SQM return a SEEK_NEXT_ROW, the store scanner will seek to the cell from next row. HBASE-13109 optimized the SEEK to SKIP when we can read the cell in current loaded block. So it will skip by call heap.next to the cell from next row. But the problem is it compare too many times with the nextIndexedKey in the while loop. We plan move the compare outside the loop to reduce compare times. One problem is the nextIndexedKey maybe changed when call heap.peek, because the current storefile scanner was changed. So my proposal is to move the "optimize SEEK to SKIP" to storefile scanner. When we call seek for storefile scanner, it may real seek or implement seek by several times skip.

      Any suggestions are welcomed. Thanks.

      Attachments

        Issue Links

          Activity

            People

              Unassigned Unassigned
              zghao Guanghao Zhang
              Votes:
              0 Vote for this issue
              Watchers:
              10 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved: