cumulative indexing gets slower

Issue #147 resolved
Shinichiro Abe repo owner created an issue

it happens when using indexholder with small documents batch. because of realtime merging?

Comments (11)

  1. Shinichiro Abe reporter

    fix a deletion bug:

    before: SmallBatches 140 ms 1940:1950 SmallBatches 141 ms 1950:1960 SmallBatches 136 ms 1960:1970 SmallBatches 147 ms 1970:1980 SmallBatches 153 ms 1980:1990 ... SmallBatches 653 ms 9980:9990 SmallBatches 645 ms 9990:10000 indexed 332813 ms

    after: SmallBatches 68 ms 1960:1970 SmallBatches 71 ms 1970:1980 SmallBatches 61 ms 1980:1990 SmallBatches 62 ms 1990:2000 SmallBatches 81 ms 2000:2010 ... SmallBatches 318 ms 9980:9990 SmallBatches 291 ms 9990:10000 indexed 160401 ms

  2. Shinichiro Abe reporter

    call forceMerge() when segments > 10

    after:

    SmallBatches 25 ms 1960:1970

    SmallBatches 17 ms 1970:1980

    SmallBatches 18 ms 1980:1990

    SmallBatches 19 ms 1990:2000

    SmallBatches 17 ms 2000:2010

    ...

    SmallBatches 130 ms 9980:9990

    SmallBatches 129 ms 9990:10000

    indexed 103877 ms

  3. Shinichiro Abe reporter

    merge 6 ms

    resize 16 ms

    SmallBatches 25 ms 1960:1970

    merge 7 ms

    resize 9 ms

    SmallBatches 19 ms 1970:1980

    merge 6 ms

    resize 8 ms

    SmallBatches 17 ms 1980:1990

    merge 6 ms

    resize 8 ms

    SmallBatches 17 ms 1990:2000

  4. Shinichiro Abe reporter

    SmallBatches 2 ms 1960:1970

    SmallBatches 3 ms 1970:1980

    SmallBatches 3 ms 1980:1990

    SmallBatches 3 ms 1990:2000

    SmallBatches 5 ms 2000:2010

    ...

    SmallBatches 5 ms 9980:9990

    SmallBatches 4 ms 9990:10000

    indexed 80449 ms

  5. Shinichiro Abe reporter

    TODO:

    • index "a"

    • assert _openRealTimeIndexReader() has "a"

    • index "b"

    • assert _openRealTimeIndexReader() has "a" and "b"

    • reinit

    • index "c"

    • assert _openRealTimeIndexReader() has "a" and "b" and "c"?

  6. Log in to comment