/ Check-in [8cf02090]
Login

Many hyperlinks are disabled.
Use anonymous login to enable hyperlinks.

Overview
Comment:Optimize seek operations on fts5 b-trees.
Downloads: Tarball | ZIP archive | SQL archive
Timelines: family | ancestors | descendants | both | trunk
Files: files | file ages | folders
SHA1: 8cf02090ce53ec150492d77d9e5e5f27665bd34f
User & Date: dan 2015-07-04 18:44:07
Context
2015-07-05
22:15
Do not allow recursive CTEs that use aggregate queries in the recursive part. check-in: 6d2999af user: drh tags: trunk
2015-07-04
18:44
Optimize seek operations on fts5 b-trees. check-in: 8cf02090 user: dan tags: trunk
18:15
Preserve the number of requested PAGECACHE pages even if the memory pointer or size is zero. Enhance the pcache1.c header comment to explain the memory layout of a page cache line. check-in: dacb2a61 user: drh tags: trunk
Changes
Hide Diffs Unified Diffs Ignore Whitespace Patch

Changes to ext/fts5/fts5_index.c.

2007
2008
2009
2010
2011
2012
2013
















































































































2014
2015
2016
2017
2018
2019
2020
....
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
      iOff += fts5GetPoslistSize(&pLeaf->p[iOff], &nPos, &bDummy);
      iOff += nPos;
    }
  }

  pIter->pDlidx = fts5DlidxIterInit(p, bRev, iSeg, pIter->iTermLeafPgno);
}

















































































































/*
** Initialize the object pIter to point to term pTerm/nTerm within segment
** pSeg. If there is no such term in the index, the iterator is set to EOF.
**
** If an error occurs, Fts5Index.rc is set to an appropriate error code. If 
** an error has already occurred when this function is called, it is a no-op.
................................................................................
  assert( pTerm && nTerm );
  memset(pIter, 0, sizeof(*pIter));
  pIter->pSeg = pSeg;

  /* This block sets stack variable iPg to the leaf page number that may
  ** contain term (pTerm/nTerm), if it is present in the segment. */
  for(h=pSeg->nHeight-1; h>0; h--){
    Fts5NodeIter node;              /* For iterating through internal nodes */
    i64 iRowid = FTS5_SEGMENT_ROWID(pSeg->iSegid, h, iPg);
    Fts5Data *pNode = fts5DataRead(p, iRowid);
    if( pNode==0 ) break;

    fts5NodeIterInit(pNode->p, pNode->n, &node);
    assert( node.term.n==0 );

    iPg = node.iChild;
    bDlidx = node.bDlidx;
    for(fts5NodeIterNext(&p->rc, &node);
        node.aData && fts5BufferCompareBlob(&node.term, pTerm, nTerm)<=0;
        fts5NodeIterNext(&p->rc, &node)
    ){
      iPg = node.iChild;
      bDlidx = node.bDlidx;
    }
    fts5NodeIterFree(&node);
    fts5DataRelease(pNode);
  }

  if( iPg<pSeg->pgnoFirst ){
    iPg = pSeg->pgnoFirst;
    bDlidx = 0;
  }







>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>







 







<




|
<
<
<
<
<
<
<
<
<
<
<
<







2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
....
2148
2149
2150
2151
2152
2153
2154

2155
2156
2157
2158
2159












2160
2161
2162
2163
2164
2165
2166
      iOff += fts5GetPoslistSize(&pLeaf->p[iOff], &nPos, &bDummy);
      iOff += nPos;
    }
  }

  pIter->pDlidx = fts5DlidxIterInit(p, bRev, iSeg, pIter->iTermLeafPgno);
}

#ifdef SQLITE_DEBUG
static void fts5AssertNodeSeekOk(
  Fts5Data *pNode,
  const u8 *pTerm, int nTerm,     /* Term to search for */
  int iExpectPg,
  int bExpectDlidx
){
  int bDlidx;
  int iPg;
  int rc = SQLITE_OK;
  Fts5NodeIter node;

  fts5NodeIterInit(pNode->p, pNode->n, &node);
  assert( node.term.n==0 );
  iPg = node.iChild;
  bDlidx = node.bDlidx;
  for(fts5NodeIterNext(&rc, &node);
      node.aData && fts5BufferCompareBlob(&node.term, pTerm, nTerm)<=0;
      fts5NodeIterNext(&rc, &node)
  ){
    iPg = node.iChild;
    bDlidx = node.bDlidx;
  }
  fts5NodeIterFree(&node);

  assert( rc!=SQLITE_OK || iPg==iExpectPg );
  assert( rc!=SQLITE_OK || bDlidx==bExpectDlidx );
}
#else
#define fts5AssertNodeSeekOk(v,w,x,y,z)
#endif

/*
** Argument pNode is an internal b-tree node. This function searches
** within the node for the largest term that is smaller than or equal
** to (pTerm/nTerm).
**
** It returns the associated page number. Or, if (pTerm/nTerm) is smaller
** than all terms within the node, the leftmost child page number. 
**
** Before returning, (*pbDlidx) is set to true if the last term on the
** returned child page number has a doclist-index. Or left as is otherwise.
*/
static int fts5NodeSeek(
  Fts5Data *pNode,                /* Node to search */
  const u8 *pTerm, int nTerm,     /* Term to search for */
  int *pbDlidx                    /* OUT: True if dlidx flag is set */
){
  int iPg;
  u8 *pPtr = pNode->p;
  u8 *pEnd = &pPtr[pNode->n];
  int nMatch = 0;                 /* Number of bytes of pTerm already matched */
  
  assert( *pbDlidx==0 );

  pPtr += fts5GetVarint32(pPtr, iPg);
  while( pPtr<pEnd ){
    int nEmpty = 0;
    int nKeep;
    int nNew;

    /* If there is a "no terms" record at pPtr, read it now. Store the
    ** number of termless pages in nEmpty. If it indicates a doclist-index, 
    ** set (*pbDlidx) to true.*/
    if( *pPtr<2 ){
      *pbDlidx = (*pPtr==0x01);
      pPtr++;
      pPtr += fts5GetVarint32(pPtr, nEmpty);
    }

    /* Read the next "term" pointer. Set nKeep to the number of bytes to
    ** keep from the previous term, and nNew to the number of bytes of
    ** new data that will be appended to it. */
    nKeep = (int)*pPtr++;
    nNew = (int)*pPtr++;
    if( (nKeep | nNew) & 0x0080 ){
      pPtr -= 2;
      pPtr += fts5GetVarint32(pPtr, nKeep);
      pPtr += fts5GetVarint32(pPtr, nNew);
    }
    nKeep -= 2;

    /* Compare (pTerm/nTerm) to the current term on the node (the one described
    ** by nKeep/nNew). If the node term is larger, break out of the while()
    ** loop. 
    **
    ** Otherwise, if (pTerm/nTerm) is larger or the two terms are equal, 
    ** leave variable nMatch set to the size of the largest prefix common to
    ** both terms in bytes.  */
    if( nKeep==nMatch ){
      int nTst = MIN(nNew, nTerm-nMatch);
      int i;
      for(i=0; i<nTst; i++){
        if( pTerm[nKeep+i]!=pPtr[i] ) break;
      }
      nMatch += i;
      assert( nMatch<=nTerm );

      if( i<nNew && (nMatch==nTerm || pPtr[i] > pTerm[nMatch]) ) break;
    }else if( nKeep<nMatch ){
      break;
    }

    iPg += 1 + nEmpty;
    *pbDlidx = 0;
    pPtr += nNew;
  }

  fts5AssertNodeSeekOk(pNode, pTerm, nTerm, iPg, *pbDlidx);
  return iPg;
}

/*
** Initialize the object pIter to point to term pTerm/nTerm within segment
** pSeg. If there is no such term in the index, the iterator is set to EOF.
**
** If an error occurs, Fts5Index.rc is set to an appropriate error code. If 
** an error has already occurred when this function is called, it is a no-op.
................................................................................
  assert( pTerm && nTerm );
  memset(pIter, 0, sizeof(*pIter));
  pIter->pSeg = pSeg;

  /* This block sets stack variable iPg to the leaf page number that may
  ** contain term (pTerm/nTerm), if it is present in the segment. */
  for(h=pSeg->nHeight-1; h>0; h--){

    i64 iRowid = FTS5_SEGMENT_ROWID(pSeg->iSegid, h, iPg);
    Fts5Data *pNode = fts5DataRead(p, iRowid);
    if( pNode==0 ) break;

    iPg = fts5NodeSeek(pNode, pTerm, nTerm, &bDlidx);












    fts5DataRelease(pNode);
  }

  if( iPg<pSeg->pgnoFirst ){
    iPg = pSeg->pgnoFirst;
    bDlidx = 0;
  }