/ Check-in [bacfe8cb]
Login

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

Overview
Comment:Avoid constantly freeing and reallocing small internal buffers associated with LSM cursors. Instead, allow them to persist for the lifetime of the cursor if they are LSM_SEGMENTPTR_FREE_THRESHOLD (default 1024) bytes or smaller. This change is based on research by Martijn Blaauw.
Downloads: Tarball | ZIP archive | SQL archive
Timelines: family | ancestors | descendants | both | trunk
Files: files | file ages | folders
SHA3-256: bacfe8cb3e4d3be736c6789b2ce55ef5603e5c7a289b05b37cae2203cd7f2290
User & Date: dan 2017-06-30 11:44:11
Original Comment: Avoid constantly freeing and reallocing small internal buffers associated with LSM cursors. Instead, allow them to persist for the lifetime of the cursor if they are LSM_SEGMENTPTR_FREE_THRESHOLD (default 1024) bytes or smaller.
Context
2017-06-30
18:12
Fix some minor typos in lsm1. check-in: 0ef777d7 user: mistachkin tags: trunk
11:44
Avoid constantly freeing and reallocing small internal buffers associated with LSM cursors. Instead, allow them to persist for the lifetime of the cursor if they are LSM_SEGMENTPTR_FREE_THRESHOLD (default 1024) bytes or smaller. This change is based on research by Martijn Blaauw. check-in: bacfe8cb user: dan tags: trunk
2017-06-29
21:33
In the command-line shell, add the -quote option to start up in quote mode. Enhance the ".mode" command so that it reports the current output mode if given no arguments. check-in: 5e3f9ea5 user: drh tags: trunk
Changes
Hide Diffs Unified Diffs Ignore Whitespace Patch

Changes to ext/lsm1/lsm_sorted.c.

94
95
96
97
98
99
100





101
102
103
104
105
106
107
....
1144
1145
1146
1147
1148
1149
1150




1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
....
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
....
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
....
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
....
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
....
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
....
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
....
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
....
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
....
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
....
6075
6076
6077
6078
6079
6080
6081
6082
6083
6084
6085
6086
6087
6088
6089
6090

#define SEGMENT_EOF(pgsz, nEntry) SEGMENT_CELLPTR_OFFSET(pgsz, nEntry)

#define SEGMENT_BTREE_FLAG     0x0001
#define PGFTR_SKIP_NEXT_FLAG   0x0002
#define PGFTR_SKIP_THIS_FLAG   0x0004






typedef struct SegmentPtr SegmentPtr;
typedef struct Blob Blob;

struct Blob {
  lsm_env *pEnv;
  void *pData;
  int nData;
................................................................................
    pLevel->nSplitKey = blob.nData;
    lsmFsPageRelease(pPg);
  }

  *pRc = rc;
}





static void segmentPtrReset(SegmentPtr *pPtr){
  lsmFsPageRelease(pPtr->pPg);
  pPtr->pPg = 0;
  pPtr->nCell = 0;
  pPtr->pKey = 0;
  pPtr->nKey = 0;
  pPtr->pVal = 0;
  pPtr->nVal = 0;
  pPtr->eType = 0;
  pPtr->iCell = 0;
  sortedBlobFree(&pPtr->blob1);
  sortedBlobFree(&pPtr->blob2);
}

static int segmentPtrIgnoreSeparators(MultiCursor *pCsr, SegmentPtr *pPtr){
  return (pCsr->flags & CURSOR_READ_SEPARATORS)==0
      || (pPtr!=&pCsr->aPtr[pCsr->nPtr-1]);
}

................................................................................
    if( rc!=LSM_OK ) return rc;

    if( svFlags && pPtr->pPg ){
      int res = sortedKeyCompare(pCsr->pDb->xCmp,
          rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey,
          pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey
      );
      if( res<0 ) segmentPtrReset(pPtr);
    }

    if( pPtr->pPg==0 && (svFlags & LSM_END_DELETE) ){
      Segment *pSeg = pPtr->pSeg;
      rc = lsmFsDbPageGet(pCsr->pDb->pFS, pSeg, pSeg->iFirst, &pPtr->pPg);
      if( rc!=LSM_OK ) return rc;
      pPtr->eType = LSM_START_DELETE | LSM_POINT_DELETE;
................................................................................
  if( rc==LSM_OK && pPtr->pPg ){
    rc = segmentPtrLoadCell(pPtr, bLast ? (pPtr->nCell-1) : 0);
    if( rc==LSM_OK && bLast && pPtr->pSeg!=&pLvl->lhs ){
      int res = sortedKeyCompare(pCsr->pDb->xCmp,
          rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey,
          pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey
      );
      if( res<0 ) segmentPtrReset(pPtr);
    }
  }
  
  bIgnore = segmentPtrIgnoreSeparators(pCsr, pPtr);
  if( rc==LSM_OK && pPtr->pPg && bIgnore && rtIsSeparator(pPtr->eType) ){
    rc = segmentPtrAdvance(pCsr, pPtr, bLast);
  }
................................................................................
    ptr.pLevel = pPtr->pLevel;
    ptr.pSeg = &ptr.pLevel->aRhs[ptr.pLevel->nRight-1];
    rc = sortedRhsFirst(pCsr, ptr.pLevel, &ptr);
    if( rc==LSM_OK ){
      rc = ptrFwdPointer(ptr.pPg, ptr.iCell, ptr.pSeg, &iOut, &bFound);
      ptr.pPg = 0;
    }
    segmentPtrReset(&ptr);
  }

  *piPtr = iOut;
  return rc;
}

static int segmentPtrSeek(
................................................................................
#endif

  assert( pPtr->nCell>0 
       || pPtr->pSeg->nSize==1 
       || lsmFsDbPageIsLast(pPtr->pSeg, pPtr->pPg)
  );
  if( pPtr->nCell==0 ){
    segmentPtrReset(pPtr);
  }else{
    iMin = 0;
    iMax = pPtr->nCell-1;

    while( 1 ){
      int iTry = (iMin+iMax)/2;
      void *pKeyT; int nKeyT;       /* Key for cell iTry */
................................................................................
              pCsr->eType = pPtr->eType;
              rc = sortedBlobSet(pEnv, &pCsr->key, pPtr->pKey, pPtr->nKey);
              if( rc==LSM_OK ){
                rc = sortedBlobSet(pEnv, &pCsr->val, pPtr->pVal, pPtr->nVal);
              }
              pCsr->flags |= CURSOR_SEEK_EQ;
            }
            segmentPtrReset(pPtr);
            break;
          }
          case LSM_SEEK_LE:
            if( res>0 ) rc = segmentPtrAdvance(pCsr, pPtr, 1);
            break;
          case LSM_SEEK_GE: {
            /* Figure out if we need to 'skip' the pointer forward or not */
................................................................................
      /* If the segment-pointer has settled on a key that is smaller than
      ** the splitkey, invalidate the segment-pointer.  */
      if( pPtr->pPg ){
        res = sortedKeyCompare(pCsr->pDb->xCmp, 
            rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey, 
            pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey
        );
        if( res<0 ) segmentPtrReset(pPtr);
      }

      if( aPtr[i].pKey ) bHit = 1;
    }

    if( rc==LSM_OK && eSeek==LSM_SEEK_LE && bHit==0 ){
      rc = segmentPtrEnd(pCsr, &aPtr[0], 1);
................................................................................

  /* Close the tree cursor, if any. */
  lsmTreeCursorDestroy(pCsr->apTreeCsr[0]);
  lsmTreeCursorDestroy(pCsr->apTreeCsr[1]);

  /* Reset the segment pointers */
  for(i=0; i<pCsr->nPtr; i++){
    segmentPtrReset(&pCsr->aPtr[i]);
  }

  /* And the b-tree cursor, if any */
  btreeCursorFree(pCsr->pBtCsr);

  /* Free allocations */
  lsmFree(pEnv, pCsr->aPtr);
................................................................................
      for(iRhs=0; iRhs<pLvl->nRight && rc==LSM_OK; iRhs++){
        rc = segmentPtrEnd(pCsr, &pPtr[iRhs+1], 1);
        if( pPtr[iRhs+1].pPg ) bHit = 1;
      }
      if( bHit==0 && rc==LSM_OK ){
        rc = segmentPtrEnd(pCsr, pPtr, 1);
      }else{
        segmentPtrReset(pPtr);
      }
    }else{
      int bLhs = (pPtr->pSeg==&pLvl->lhs);
      assert( pPtr->pSeg==&pLvl->lhs || pPtr->pSeg==&pLvl->aRhs[0] );

      if( bLhs ){
        rc = segmentPtrEnd(pCsr, pPtr, 0);
        if( pPtr->pKey ) bHit = 1;
      }
      for(iRhs=0; iRhs<pLvl->nRight && rc==LSM_OK; iRhs++){
        if( bHit ){
          segmentPtrReset(&pPtr[iRhs+1]);
        }else{
          rc = sortedRhsFirst(pCsr, pLvl, &pPtr[iRhs+bLhs]);
        }
      }
    }
    i += pLvl->nRight;
  }
................................................................................
}

void lsmMCursorReset(MultiCursor *pCsr){
  int i;
  lsmTreeCursorReset(pCsr->apTreeCsr[0]);
  lsmTreeCursorReset(pCsr->apTreeCsr[1]);
  for(i=0; i<pCsr->nPtr; i++){
    segmentPtrReset(&pCsr->aPtr[i]);
  }
  pCsr->key.nData = 0;
}

static int treeCursorSeek(
  MultiCursor *pCsr,
  TreeCursor *pTreeCsr, 
................................................................................
        if( ptr1.pPg==0 ){
          assert( 0 );
        }
      }
    }
  }

  segmentPtrReset(&ptr1);
  segmentPtrReset(&ptr2);
  return LSM_OK;
}

/*
** This function is only included in the build if LSM_DEBUG_EXPENSIVE is 
** defined. Its only purpose is to evaluate various assert() statements to 
** verify that the database is well formed in certain respects.







>
>
>
>
>







 







>
>
>
>
|









|
|







 







|







 







|







 







|







 







|







 







|







 







|







 







|







 







|











|







 







|







 







|
|







94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
....
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
....
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
....
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
....
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
....
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
....
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
....
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
....
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
....
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
....
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
....
6084
6085
6086
6087
6088
6089
6090
6091
6092
6093
6094
6095
6096
6097
6098
6099

#define SEGMENT_EOF(pgsz, nEntry) SEGMENT_CELLPTR_OFFSET(pgsz, nEntry)

#define SEGMENT_BTREE_FLAG     0x0001
#define PGFTR_SKIP_NEXT_FLAG   0x0002
#define PGFTR_SKIP_THIS_FLAG   0x0004


#ifndef LSM_SEGMENTPTR_FREE_THRESHOLD
# define LSM_SEGMENTPTR_FREE_THRESHOLD 1024
#endif

typedef struct SegmentPtr SegmentPtr;
typedef struct Blob Blob;

struct Blob {
  lsm_env *pEnv;
  void *pData;
  int nData;
................................................................................
    pLevel->nSplitKey = blob.nData;
    lsmFsPageRelease(pPg);
  }

  *pRc = rc;
}

/*
** Reset a segment cursor. Also free its buffers if they are nThreshold
** bytes or larger in size.
*/
static void segmentPtrReset(SegmentPtr *pPtr, int nThreshold){
  lsmFsPageRelease(pPtr->pPg);
  pPtr->pPg = 0;
  pPtr->nCell = 0;
  pPtr->pKey = 0;
  pPtr->nKey = 0;
  pPtr->pVal = 0;
  pPtr->nVal = 0;
  pPtr->eType = 0;
  pPtr->iCell = 0;
  if( pPtr->blob1.nAlloc>=nThreshold ) sortedBlobFree(&pPtr->blob1);
  if( pPtr->blob2.nAlloc>=nThreshold ) sortedBlobFree(&pPtr->blob2);
}

static int segmentPtrIgnoreSeparators(MultiCursor *pCsr, SegmentPtr *pPtr){
  return (pCsr->flags & CURSOR_READ_SEPARATORS)==0
      || (pPtr!=&pCsr->aPtr[pCsr->nPtr-1]);
}

................................................................................
    if( rc!=LSM_OK ) return rc;

    if( svFlags && pPtr->pPg ){
      int res = sortedKeyCompare(pCsr->pDb->xCmp,
          rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey,
          pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey
      );
      if( res<0 ) segmentPtrReset(pPtr, LSM_SEGMENTPTR_FREE_THRESHOLD);
    }

    if( pPtr->pPg==0 && (svFlags & LSM_END_DELETE) ){
      Segment *pSeg = pPtr->pSeg;
      rc = lsmFsDbPageGet(pCsr->pDb->pFS, pSeg, pSeg->iFirst, &pPtr->pPg);
      if( rc!=LSM_OK ) return rc;
      pPtr->eType = LSM_START_DELETE | LSM_POINT_DELETE;
................................................................................
  if( rc==LSM_OK && pPtr->pPg ){
    rc = segmentPtrLoadCell(pPtr, bLast ? (pPtr->nCell-1) : 0);
    if( rc==LSM_OK && bLast && pPtr->pSeg!=&pLvl->lhs ){
      int res = sortedKeyCompare(pCsr->pDb->xCmp,
          rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey,
          pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey
      );
      if( res<0 ) segmentPtrReset(pPtr, LSM_SEGMENTPTR_FREE_THRESHOLD);
    }
  }
  
  bIgnore = segmentPtrIgnoreSeparators(pCsr, pPtr);
  if( rc==LSM_OK && pPtr->pPg && bIgnore && rtIsSeparator(pPtr->eType) ){
    rc = segmentPtrAdvance(pCsr, pPtr, bLast);
  }
................................................................................
    ptr.pLevel = pPtr->pLevel;
    ptr.pSeg = &ptr.pLevel->aRhs[ptr.pLevel->nRight-1];
    rc = sortedRhsFirst(pCsr, ptr.pLevel, &ptr);
    if( rc==LSM_OK ){
      rc = ptrFwdPointer(ptr.pPg, ptr.iCell, ptr.pSeg, &iOut, &bFound);
      ptr.pPg = 0;
    }
    segmentPtrReset(&ptr, 0);
  }

  *piPtr = iOut;
  return rc;
}

static int segmentPtrSeek(
................................................................................
#endif

  assert( pPtr->nCell>0 
       || pPtr->pSeg->nSize==1 
       || lsmFsDbPageIsLast(pPtr->pSeg, pPtr->pPg)
  );
  if( pPtr->nCell==0 ){
    segmentPtrReset(pPtr, LSM_SEGMENTPTR_FREE_THRESHOLD);
  }else{
    iMin = 0;
    iMax = pPtr->nCell-1;

    while( 1 ){
      int iTry = (iMin+iMax)/2;
      void *pKeyT; int nKeyT;       /* Key for cell iTry */
................................................................................
              pCsr->eType = pPtr->eType;
              rc = sortedBlobSet(pEnv, &pCsr->key, pPtr->pKey, pPtr->nKey);
              if( rc==LSM_OK ){
                rc = sortedBlobSet(pEnv, &pCsr->val, pPtr->pVal, pPtr->nVal);
              }
              pCsr->flags |= CURSOR_SEEK_EQ;
            }
            segmentPtrReset(pPtr, LSM_SEGMENTPTR_FREE_THRESHOLD);
            break;
          }
          case LSM_SEEK_LE:
            if( res>0 ) rc = segmentPtrAdvance(pCsr, pPtr, 1);
            break;
          case LSM_SEEK_GE: {
            /* Figure out if we need to 'skip' the pointer forward or not */
................................................................................
      /* If the segment-pointer has settled on a key that is smaller than
      ** the splitkey, invalidate the segment-pointer.  */
      if( pPtr->pPg ){
        res = sortedKeyCompare(pCsr->pDb->xCmp, 
            rtTopic(pPtr->eType), pPtr->pKey, pPtr->nKey, 
            pLvl->iSplitTopic, pLvl->pSplitKey, pLvl->nSplitKey
        );
        if( res<0 ) segmentPtrReset(pPtr, LSM_SEGMENTPTR_FREE_THRESHOLD);
      }

      if( aPtr[i].pKey ) bHit = 1;
    }

    if( rc==LSM_OK && eSeek==LSM_SEEK_LE && bHit==0 ){
      rc = segmentPtrEnd(pCsr, &aPtr[0], 1);
................................................................................

  /* Close the tree cursor, if any. */
  lsmTreeCursorDestroy(pCsr->apTreeCsr[0]);
  lsmTreeCursorDestroy(pCsr->apTreeCsr[1]);

  /* Reset the segment pointers */
  for(i=0; i<pCsr->nPtr; i++){
    segmentPtrReset(&pCsr->aPtr[i], 0);
  }

  /* And the b-tree cursor, if any */
  btreeCursorFree(pCsr->pBtCsr);

  /* Free allocations */
  lsmFree(pEnv, pCsr->aPtr);
................................................................................
      for(iRhs=0; iRhs<pLvl->nRight && rc==LSM_OK; iRhs++){
        rc = segmentPtrEnd(pCsr, &pPtr[iRhs+1], 1);
        if( pPtr[iRhs+1].pPg ) bHit = 1;
      }
      if( bHit==0 && rc==LSM_OK ){
        rc = segmentPtrEnd(pCsr, pPtr, 1);
      }else{
        segmentPtrReset(pPtr, LSM_SEGMENTPTR_FREE_THRESHOLD);
      }
    }else{
      int bLhs = (pPtr->pSeg==&pLvl->lhs);
      assert( pPtr->pSeg==&pLvl->lhs || pPtr->pSeg==&pLvl->aRhs[0] );

      if( bLhs ){
        rc = segmentPtrEnd(pCsr, pPtr, 0);
        if( pPtr->pKey ) bHit = 1;
      }
      for(iRhs=0; iRhs<pLvl->nRight && rc==LSM_OK; iRhs++){
        if( bHit ){
          segmentPtrReset(&pPtr[iRhs+1], LSM_SEGMENTPTR_FREE_THRESHOLD);
        }else{
          rc = sortedRhsFirst(pCsr, pLvl, &pPtr[iRhs+bLhs]);
        }
      }
    }
    i += pLvl->nRight;
  }
................................................................................
}

void lsmMCursorReset(MultiCursor *pCsr){
  int i;
  lsmTreeCursorReset(pCsr->apTreeCsr[0]);
  lsmTreeCursorReset(pCsr->apTreeCsr[1]);
  for(i=0; i<pCsr->nPtr; i++){
    segmentPtrReset(&pCsr->aPtr[i], LSM_SEGMENTPTR_FREE_THRESHOLD);
  }
  pCsr->key.nData = 0;
}

static int treeCursorSeek(
  MultiCursor *pCsr,
  TreeCursor *pTreeCsr, 
................................................................................
        if( ptr1.pPg==0 ){
          assert( 0 );
        }
      }
    }
  }

  segmentPtrReset(&ptr1, 0);
  segmentPtrReset(&ptr2, 0);
  return LSM_OK;
}

/*
** This function is only included in the build if LSM_DEBUG_EXPENSIVE is 
** defined. Its only purpose is to evaluate various assert() statements to 
** verify that the database is well formed in certain respects.