Many hyperlinks are disabled.
Use anonymous login
to enable hyperlinks.
Overview
Comment: | Fix many issues with new code. |
---|---|
Downloads: | Tarball | ZIP archive |
Timelines: | family | ancestors | descendants | both | threads-experimental |
Files: | files | file ages | folders |
SHA1: |
62c406a042d7246f6df6b943421182a8 |
User & Date: | dan 2014-04-12 19:34:44.467 |
Context
2014-04-14
| ||
07:30 | Improve use of multiple threads in sqlite3VdbeSorterRewind(). (check-in: e1bdc4b810 user: dan tags: threads-experimental) | |
2014-04-12
| ||
19:34 | Fix many issues with new code. (check-in: 62c406a042 user: dan tags: threads-experimental) | |
2014-04-11
| ||
19:43 | Avoid having the sorter merge too many PMAs at a time when incrementally merging data following a SorterRewind(). (check-in: 98bf0307b1 user: dan tags: threads-experimental) | |
Changes
Changes to src/vdbesort.c.
︙ | ︙ | |||
84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 | ** algorithm might be performed in parallel by separate threads. Threads ** are only used when one or more PMA spill to disk. If the sort is small ** enough to fit entirely in memory, everything happens on the main thread. */ #include "sqliteInt.h" #include "vdbeInt.h" /* ** Private objects used by the sorter */ typedef struct MergeEngine MergeEngine; /* Merge PMAs together */ typedef struct PmaReader PmaReader; /* Incrementally read one PMA */ typedef struct PmaWriter PmaWriter; /* Incrementally write on PMA */ typedef struct SorterRecord SorterRecord; /* A record being sorted */ typedef struct SortSubtask SortSubtask; /* A sub-task in the sort process */ typedef struct SorterFile SorterFile; typedef struct IncrMerger IncrMerger; /* | > > > > > > > > > > > > > > > > > > > > > > > > > > > | > > > > > > > > > | < < > > > | < < > > > | 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 | ** algorithm might be performed in parallel by separate threads. Threads ** are only used when one or more PMA spill to disk. If the sort is small ** enough to fit entirely in memory, everything happens on the main thread. */ #include "sqliteInt.h" #include "vdbeInt.h" /* ** If SQLITE_DEBUG_SORTER_THREADS is defined, this module outputs various ** messages to stderr that may be helpful in understanding the performance ** characteristics of the sorter in multi-threaded mode. */ #if 0 # define SQLITE_DEBUG_SORTER_THREADS 1 #endif /* ** Private objects used by the sorter */ typedef struct MergeEngine MergeEngine; /* Merge PMAs together */ typedef struct PmaReader PmaReader; /* Incrementally read one PMA */ typedef struct PmaWriter PmaWriter; /* Incrementally write on PMA */ typedef struct SorterRecord SorterRecord; /* A record being sorted */ typedef struct SortSubtask SortSubtask; /* A sub-task in the sort process */ typedef struct SorterFile SorterFile; typedef struct SorterThread SorterThread; typedef struct SorterList SorterList; typedef struct IncrMerger IncrMerger; /* ** A container for a temp file handle and the current amount of data ** stored in the file. */ struct SorterFile { sqlite3_file *pFd; /* File handle */ i64 iEof; /* Bytes of data stored in pFd */ }; /* ** An object of this type is used to store the thread handle for each ** background thread launched by the sorter. Before the thread is launched, ** variable bDone is set to 0. Then, right before it exits, the thread ** itself sets bDone to 1. ** ** This is then used for two purposes: ** ** 1. When flushing the contents of memory to a level-0 PMA on disk, to ** attempt to select a SortSubtask for which there is not already an ** active background thread (since doing so causes the main thread ** to block until it finishes). ** ** 2. If SQLITE_DEBUG_SORTER_THREADS is defined, to determine if a call ** to sqlite3ThreadJoin() is likely to block. ** ** In both cases, the effects of the main thread seeing (bDone==0) even ** after the thread has finished are not dire. So we don't worry about ** memory barriers and such here. */ struct SorterThread { SQLiteThread *pThread; int bDone; }; struct SorterList { SorterRecord *pList; /* Linked list of records */ u8 *aMemory; /* If non-NULL, blob of memory for pList */ int szPMA; /* Size of pList as PMA in bytes */ }; /* ** Sorting is divided up into smaller subtasks. Each subtask is controlled ** by an instance of this object. A Subtask might run in either the main thread ** or in a background thread. ** |
︙ | ︙ | |||
144 145 146 147 148 149 150 | ** the temp file if it is not already open. ** ** SORT_SUBTASK_CONS: ** Merge existing PMAs until SortSubtask.nConsolidate or fewer ** remain in temp file SortSubtask.pTemp1. */ struct SortSubtask { | < < | | < < < < < < | 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 | ** the temp file if it is not already open. ** ** SORT_SUBTASK_CONS: ** Merge existing PMAs until SortSubtask.nConsolidate or fewer ** remain in temp file SortSubtask.pTemp1. */ struct SortSubtask { SorterThread thread; sqlite3 *db; /* Database connection */ VdbeSorter *pSorter; /* Sorter */ KeyInfo *pKeyInfo; /* How to compare records */ UnpackedRecord *pUnpacked; /* Space to unpack a record */ int pgsz; /* Main database page size */ SorterList list; /* List for thread to write to a PMA */ int nPMA; /* Number of PMAs currently in file */ SorterFile file; /* Temp file for level-0 PMAs */ SorterFile file2; /* Space for other PMAs */ }; /* |
︙ | ︙ | |||
244 245 246 247 248 249 250 | ** ** mxKeysize: ** As records are added to the sorter by calls to sqlite3VdbeSorterWrite(), ** this variable is updated so as to be set to the size on disk of the ** largest record in the sorter. */ struct VdbeSorter { | < < > > > > > | 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 | ** ** mxKeysize: ** As records are added to the sorter by calls to sqlite3VdbeSorterWrite(), ** this variable is updated so as to be set to the size on disk of the ** largest record in the sorter. */ struct VdbeSorter { int mnPmaSize; /* Minimum PMA size, in bytes */ int mxPmaSize; /* Maximum PMA size, in bytes. 0==no limit */ int bUsePMA; /* True if one or more PMAs created */ int bUseThreads; /* True if one or more PMAs created */ PmaReader *pReader; /* Read data from here after Rewind() */ int mxKeysize; /* Largest serialized key seen so far */ UnpackedRecord *pUnpacked; /* Used by VdbeSorterCompare() */ #if 0 int nInMemory; /* Current size of pRecord list as PMA */ SorterRecord *pRecord; /* Head of in-memory record list */ u8 *aMemory; /* Block of memory to alloc records from */ #endif SorterList list; int iMemory; /* Offset of first free byte in aMemory */ int nMemory; /* Size of aMemory allocation in bytes */ int iPrev; /* Previous thread used to flush PMA */ int nTask; /* Size of aTask[] array */ SortSubtask aTask[1]; /* One or more subtasks */ }; |
︙ | ︙ | |||
288 289 290 291 292 293 294 | ** Normally, a PmaReader object iterates through an existing PMA stored ** within a temp file. However, if the PmaReader.pIncr variable points to ** an object of the following type, it may be used to iterate/merge through ** multiple PMAs simultaneously. */ struct IncrMerger { SortSubtask *pTask; /* Task that owns this merger */ | | | 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 | ** Normally, a PmaReader object iterates through an existing PMA stored ** within a temp file. However, if the PmaReader.pIncr variable points to ** an object of the following type, it may be used to iterate/merge through ** multiple PMAs simultaneously. */ struct IncrMerger { SortSubtask *pTask; /* Task that owns this merger */ SorterThread thread; /* Thread for populating aFile[1] */ MergeEngine *pMerger; /* Merge engine thread reads data from */ i64 iStartOff; /* Offset to start writing file at */ int mxSz; /* Maximum bytes of data to store */ int bEof; /* Set to true when merge is finished */ int bUseThread; /* True to use a bg thread for this object */ SorterFile aFile[2]; /* aFile[0] for reading, [1] for writing */ }; |
︙ | ︙ | |||
783 784 785 786 787 788 789 | /* If the application is using memsys3 or memsys5, use a separate ** allocation for each sort-key in memory. Otherwise, use a single big ** allocation at pSorter->aMemory for all sort-keys. */ if( sqlite3GlobalConfig.pHeap==0 ){ assert( pSorter->iMemory==0 ); pSorter->nMemory = pgsz; | | | | 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 | /* If the application is using memsys3 or memsys5, use a separate ** allocation for each sort-key in memory. Otherwise, use a single big ** allocation at pSorter->aMemory for all sort-keys. */ if( sqlite3GlobalConfig.pHeap==0 ){ assert( pSorter->iMemory==0 ); pSorter->nMemory = pgsz; pSorter->list.aMemory = (u8*)sqlite3Malloc(pgsz); if( !pSorter->list.aMemory ) rc = SQLITE_NOMEM; } } } return rc; } |
︙ | ︙ | |||
811 812 813 814 815 816 817 | /* ** Free all resources owned by the object indicated by argument pTask. All ** fields of *pTask are zeroed before returning. */ static void vdbeSortSubtaskCleanup(sqlite3 *db, SortSubtask *pTask){ sqlite3DbFree(db, pTask->pUnpacked); pTask->pUnpacked = 0; | | | | | | > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > > | > > > > > > > > > > > > > > > > > | > > > > > > > > > > > > > > > > < < | < < | < < > | 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 | /* ** Free all resources owned by the object indicated by argument pTask. All ** fields of *pTask are zeroed before returning. */ static void vdbeSortSubtaskCleanup(sqlite3 *db, SortSubtask *pTask){ sqlite3DbFree(db, pTask->pUnpacked); pTask->pUnpacked = 0; if( pTask->list.aMemory==0 ){ vdbeSorterRecordFree(0, pTask->list.pList); }else{ sqlite3_free(pTask->list.aMemory); pTask->list.aMemory = 0; } pTask->list.pList = 0; if( pTask->file.pFd ){ sqlite3OsCloseFree(pTask->file.pFd); pTask->file.pFd = 0; pTask->file.iEof = 0; } if( pTask->file2.pFd ){ sqlite3OsCloseFree(pTask->file2.pFd); pTask->file2.pFd = 0; pTask->file2.iEof = 0; } } #ifdef SQLITE_DEBUG_SORTER_THREADS static void vdbeSorterWorkDebug(SortSubtask *pTask, const char *zEvent){ i64 t; int iTask = (pTask - pTask->pSorter->aTask); sqlite3OsCurrentTimeInt64(pTask->db->pVfs, &t); fprintf(stderr, "%lld:%d %s\n", t, iTask, zEvent); } static void vdbeSorterRewindDebug(sqlite3 *db, const char *zEvent){ i64 t; sqlite3OsCurrentTimeInt64(db->pVfs, &t); fprintf(stderr, "%lld:X %s\n", t, zEvent); } static void vdbeSorterPopulateDebug( SortSubtask *pTask, const char *zEvent ){ i64 t; int iTask = (pTask - pTask->pSorter->aTask); sqlite3OsCurrentTimeInt64(pTask->db->pVfs, &t); fprintf(stderr, "%lld:bg%d %s\n", t, iTask, zEvent); } static void vdbeSorterBlockDebug( SortSubtask *pTask, int bBlocked, const char *zEvent ){ if( bBlocked ){ i64 t; sqlite3OsCurrentTimeInt64(pTask->db->pVfs, &t); fprintf(stderr, "%lld:main %s\n", t, zEvent); } } #else # define vdbeSorterWorkDebug(x,y) # define vdbeSorterRewindDebug(x,y) # define vdbeSorterPopulateDebug(x,y) # define vdbeSorterBlockDebug(x,y,z) #endif #if SQLITE_MAX_WORKER_THREADS>0 /* ** Join thread p. */ static int vdbeSorterJoinThread(SortSubtask *pTask, SorterThread *p){ int rc = SQLITE_OK; if( p->pThread ){ #ifdef SQLITE_DEBUG_SORTER_THREADS int bDone = p->bDone; #endif void *pRet; vdbeSorterBlockDebug(pTask, !bDone, "enter"); rc = sqlite3ThreadJoin(p->pThread, &pRet); vdbeSorterBlockDebug(pTask, !bDone, "exit"); if( rc==SQLITE_OK ) rc = SQLITE_PTR_TO_INT(pRet); assert( p->bDone==1 ); p->bDone = 0; p->pThread = 0; } return rc; } /* ** Launch a background thread to run xTask(pIn). */ static int vdbeSorterCreateThread( SorterThread *p, /* Thread object to populate */ void *(*xTask)(void*), /* Routine to run in a separate thread */ void *pIn /* Argument passed into xTask() */ ){ assert( p->pThread==0 && p->bDone==0 ); return sqlite3ThreadCreate(&p->pThread, xTask, pIn); } /* ** Join all outstanding threads launched by SorterWrite() to create ** level-0 PMAs. */ static int vdbeSorterJoinAll(VdbeSorter *pSorter, int rcin){ int rc = rcin; int i; for(i=0; i<pSorter->nTask; i++){ SortSubtask *pTask = &pSorter->aTask[i]; int rc2 = vdbeSorterJoinThread(pTask, &pTask->thread); if( rc==SQLITE_OK ) rc = rc2; } return rc; } #else # define vdbeSorterJoinAll(x,rcin) (rcin) # define vdbeSorterJoinThread(pTask,p) SQLITE_OK #endif /* ** Allocate a new MergeEngine object with space for nIter iterators. */ static MergeEngine *vdbeMergeEngineNew(int nIter){ int N = 2; /* Smallest power of two >= nIter */ |
︙ | ︙ | |||
904 905 906 907 908 909 910 | sqlite3DbFree(db, pSorter->pReader); pSorter->pReader = 0; } for(i=0; i<pSorter->nTask; i++){ SortSubtask *pTask = &pSorter->aTask[i]; vdbeSortSubtaskCleanup(db, pTask); } | | | | | | | 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 | sqlite3DbFree(db, pSorter->pReader); pSorter->pReader = 0; } for(i=0; i<pSorter->nTask; i++){ SortSubtask *pTask = &pSorter->aTask[i]; vdbeSortSubtaskCleanup(db, pTask); } if( pSorter->list.aMemory==0 ){ vdbeSorterRecordFree(0, pSorter->list.pList); } pSorter->list.pList = 0; pSorter->list.szPMA = 0; pSorter->bUsePMA = 0; pSorter->iMemory = 0; pSorter->mxKeysize = 0; sqlite3DbFree(db, pSorter->pUnpacked); pSorter->pUnpacked = 0; } /* ** Free any cursor components allocated by sqlite3VdbeSorterXXX routines. */ void sqlite3VdbeSorterClose(sqlite3 *db, VdbeCursor *pCsr){ VdbeSorter *pSorter = pCsr->pSorter; if( pSorter ){ sqlite3VdbeSorterReset(db, pSorter); sqlite3_free(pSorter->list.aMemory); sqlite3DbFree(db, pSorter); pCsr->pSorter = 0; } } /* ** Allocate space for a file-handle and open a temporary file. If successful, |
︙ | ︙ | |||
947 948 949 950 951 952 953 954 955 956 957 958 959 960 | ); if( rc==SQLITE_OK ){ i64 max = SQLITE_MAX_MMAP_SIZE; sqlite3OsFileControlHint( *ppFile, SQLITE_FCNTL_MMAP_SIZE, (void*)&max); } return rc; } /* ** Merge the two sorted lists p1 and p2 into a single list. ** Set *ppOut to the head of the new list. */ static void vdbeSorterMerge( SortSubtask *pTask, /* Calling thread context */ | > > > > > > > > > > > > > > > | 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 | ); if( rc==SQLITE_OK ){ i64 max = SQLITE_MAX_MMAP_SIZE; sqlite3OsFileControlHint( *ppFile, SQLITE_FCNTL_MMAP_SIZE, (void*)&max); } return rc; } static int vdbeSortAllocUnpacked(SortSubtask *pTask){ if( pTask->pUnpacked==0 ){ char *pFree; pTask->pUnpacked = sqlite3VdbeAllocUnpackedRecord( pTask->pKeyInfo, 0, 0, &pFree ); assert( pTask->pUnpacked==(UnpackedRecord*)pFree ); if( pFree==0 ) return SQLITE_NOMEM; pTask->pUnpacked->nField = pTask->pKeyInfo->nField; pTask->pUnpacked->errCode = 0; } return SQLITE_OK; } /* ** Merge the two sorted lists p1 and p2 into a single list. ** Set *ppOut to the head of the new list. */ static void vdbeSorterMerge( SortSubtask *pTask, /* Calling thread context */ |
︙ | ︙ | |||
987 988 989 990 991 992 993 | } /* ** Sort the linked list of records headed at pTask->pList. Return ** SQLITE_OK if successful, or an SQLite error code (i.e. SQLITE_NOMEM) if ** an error occurs. */ | | > > > > | | | | | | > > > > | 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 | } /* ** Sort the linked list of records headed at pTask->pList. Return ** SQLITE_OK if successful, or an SQLite error code (i.e. SQLITE_NOMEM) if ** an error occurs. */ static int vdbeSorterSort(SortSubtask *pTask, SorterList *pList){ int i; SorterRecord **aSlot; SorterRecord *p; int rc; rc = vdbeSortAllocUnpacked(pTask); if( rc!=SQLITE_OK ) return rc; aSlot = (SorterRecord **)sqlite3MallocZero(64 * sizeof(SorterRecord *)); if( !aSlot ){ return SQLITE_NOMEM; } p = pList->pList; while( p ){ SorterRecord *pNext; if( pList->aMemory ){ if( (u8*)p==pList->aMemory ){ pNext = 0; }else{ assert( p->u.iNext<sqlite3MallocSize(pList->aMemory) ); pNext = (SorterRecord*)&pList->aMemory[p->u.iNext]; } }else{ pNext = p->u.pNext; } p->u.pNext = 0; for(i=0; aSlot[i]; i++){ vdbeSorterMerge(pTask, p, aSlot[i], &p); aSlot[i] = 0; } aSlot[i] = p; p = pNext; } p = 0; for(i=0; i<64; i++){ vdbeSorterMerge(pTask, p, aSlot[i], &p); } pList->pList = p; sqlite3_free(aSlot); if( pTask->pUnpacked->errCode ){ assert( pTask->pUnpacked->errCode==SQLITE_NOMEM ); return SQLITE_NOMEM; } return SQLITE_OK; } /* ** Initialize a PMA-writer object. */ static void vdbePmaWriterInit( |
︙ | ︙ | |||
1140 1141 1142 1143 1144 1145 1146 | } #else # define vdbeSorterExtendFile(x,y,z) SQLITE_OK #endif /* | | > | | > > > > > > > | | > > > > > | | | | > | > | 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 | } #else # define vdbeSorterExtendFile(x,y,z) SQLITE_OK #endif /* ** Write the current contents of in-memory linked-list pList to a level-0 ** PMA in the temp file belonging to sub-task pTask. Return SQLITE_OK if ** successful, or an SQLite error code otherwise. ** ** The format of a PMA is: ** ** * A varint. This varint contains the total number of bytes of content ** in the PMA (not including the varint itself). ** ** * One or more records packed end-to-end in order of ascending keys. ** Each record consists of a varint followed by a blob of data (the ** key). The varint is the number of bytes in the blob of data. */ static int vdbeSorterListToPMA(SortSubtask *pTask, SorterList *pList){ int rc = SQLITE_OK; /* Return code */ PmaWriter writer; /* Object used to write to the file */ #ifdef SQLITE_DEBUG /* Set iSz to the expected size of file pTask->file after writing the PMA. ** This is used by an assert() statement at the end of this function. */ i64 iSz = pList->szPMA + sqlite3VarintLen(pList->szPMA) + pTask->file.iEof; #endif vdbeSorterWorkDebug(pTask, "enter"); memset(&writer, 0, sizeof(PmaWriter)); assert( pList->szPMA>0 ); /* If the first temporary PMA file has not been opened, open it now. */ if( pTask->file.pFd==0 ){ rc = vdbeSorterOpenTempFile(pTask->db->pVfs, &pTask->file.pFd); assert( rc!=SQLITE_OK || pTask->file.pFd ); assert( pTask->file.iEof==0 ); assert( pTask->nPMA==0 ); } /* Try to get the file to memory map */ if( rc==SQLITE_OK ){ vdbeSorterExtendFile(pTask->db, pTask->file.pFd, pTask->file.iEof + pList->szPMA + 9 ); } /* Sort the list */ if( rc==SQLITE_OK ){ rc = vdbeSorterSort(pTask, pList); } if( rc==SQLITE_OK ){ SorterRecord *p; SorterRecord *pNext = 0; vdbePmaWriterInit(pTask->file.pFd, &writer, pTask->pgsz, pTask->file.iEof); pTask->nPMA++; vdbePmaWriteVarint(&writer, pList->szPMA); for(p=pList->pList; p; p=pNext){ pNext = p->u.pNext; vdbePmaWriteVarint(&writer, p->nVal); vdbePmaWriteBlob(&writer, SRVAL(p), p->nVal); if( pList->aMemory==0 ) sqlite3_free(p); } pList->pList = p; rc = vdbePmaWriterFinish(&writer, &pTask->file.iEof); } vdbeSorterWorkDebug(pTask, "exit"); assert( rc!=SQLITE_OK || pList->pList==0 ); assert( rc!=SQLITE_OK || pTask->file.iEof==iSz ); return rc; } /* ** Advance the MergeEngine iterator passed as the second argument to ** the next entry. Set *pbEof to true if this means the iterator has ** reached EOF. |
︙ | ︙ | |||
1271 1272 1273 1274 1275 1276 1277 | } *pbEof = (pMerger->aIter[pMerger->aTree[1]].pFile==0); } return rc; } | < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < | < | < < < < | < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < < | < < < < < | < < < < < < < < < < < < < < < < | < < | > | > > > > > > > > > > > > > < | < < < < | > | < | | > | > > > > | | | > | | | | < < < < | < | | | < < < | < < < < < < < < < | < | | > | | < < < < < | < < < < < | | | < > | 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 | } *pbEof = (pMerger->aIter[pMerger->aTree[1]].pFile==0); } return rc; } /* ** The main routine for sorter-thread operations. */ static void *vdbeSorterFlushThread(void *pCtx){ SortSubtask *pTask = (SortSubtask*)pCtx; int rc; /* Return code */ assert( pTask->thread.bDone==0 ); rc = vdbeSorterListToPMA(pTask, &pTask->list); pTask->thread.bDone = 1; return SQLITE_INT_TO_PTR(rc); } /* ** Flush the current contents of VdbeSorter.list to a new PMA, possibly ** using a background thread. */ static int vdbeSorterFlushPMA(VdbeSorter *pSorter){ #if SQLITE_MAX_WORKER_THREADS==0 pSorter->bUsePMA = 1; return vdbeSorterListToPMA(&pSorter->aTask[0], &pSorter->list); #else int rc = SQLITE_OK; int i; SortSubtask *pTask = 0; /* Thread context used to create new PMA */ int nWorker = (pSorter->nTask-1); /* Set the flag to indicate that at least one PMA has been written. ** Or will be, anyhow. */ pSorter->bUsePMA = 1; /* Select a sub-task to sort and flush the current list of in-memory ** records to disk. If the sorter is running in multi-threaded mode, ** round-robin between the first (pSorter->nTask-1) tasks. Except, if ** the background thread from a sub-tasks previous turn is still running, ** skip it. If the first (pSorter->nTask-1) sub-tasks are all still busy, ** fall back to using the final sub-task. The first (pSorter->nTask-1) ** sub-tasks are prefered as they use background threads - the final ** sub-task uses the main thread. */ for(i=0; i<nWorker; i++){ int iTest = (pSorter->iPrev + i + 1) % nWorker; pTask = &pSorter->aTask[iTest]; if( pTask->thread.bDone ){ rc = vdbeSorterJoinThread(pTask, &pTask->thread); } if( pTask->thread.pThread==0 || rc!=SQLITE_OK ) break; } if( rc==SQLITE_OK ){ if( i==nWorker ){ /* Use the foreground thread for this operation */ rc = vdbeSorterListToPMA(&pSorter->aTask[nWorker], &pSorter->list); }else{ /* Launch a background thread for this operation */ u8 *aMem = pTask->list.aMemory; void *pCtx = (void*)pTask; assert( pTask->thread.pThread==0 && pTask->thread.bDone==0 ); assert( pTask->list.pList==0 ); assert( pTask->list.aMemory==0 || pSorter->list.aMemory!=0 ); pSorter->iPrev = (pTask - pSorter->aTask); pTask->list = pSorter->list; pSorter->list.pList = 0; pSorter->list.szPMA = 0; if( aMem ){ pSorter->list.aMemory = aMem; pSorter->nMemory = sqlite3MallocSize(aMem); }else{ pSorter->list.aMemory = sqlite3Malloc(pSorter->nMemory); if( !pSorter->list.aMemory ) return SQLITE_NOMEM; } rc = vdbeSorterCreateThread(&pTask->thread, vdbeSorterFlushThread, pCtx); } } return rc; #endif } /* ** Add a record to the sorter. */ int sqlite3VdbeSorterWrite( sqlite3 *db, /* Database handle */ |
︙ | ︙ | |||
1553 1554 1555 1556 1557 1558 1559 | ** ** * The total memory allocated for the in-memory list is greater ** than (page-size * 10) and sqlite3HeapNearlyFull() returns true. */ nReq = pVal->n + sizeof(SorterRecord); nPMA = pVal->n + sqlite3VarintLen(pVal->n); if( pSorter->mxPmaSize ){ | | | | | | | | | | | | | | | | | < < < < < < < < < < < < | 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 | ** ** * The total memory allocated for the in-memory list is greater ** than (page-size * 10) and sqlite3HeapNearlyFull() returns true. */ nReq = pVal->n + sizeof(SorterRecord); nPMA = pVal->n + sqlite3VarintLen(pVal->n); if( pSorter->mxPmaSize ){ if( pSorter->list.aMemory ){ bFlush = pSorter->iMemory && (pSorter->iMemory+nReq) > pSorter->mxPmaSize; }else{ bFlush = ( (pSorter->list.szPMA > pSorter->mxPmaSize) || (pSorter->list.szPMA > pSorter->mnPmaSize && sqlite3HeapNearlyFull()) ); } if( bFlush ){ rc = vdbeSorterFlushPMA(pSorter); pSorter->list.szPMA = 0; pSorter->iMemory = 0; assert( rc!=SQLITE_OK || pSorter->list.pList==0 ); } } pSorter->list.szPMA += nPMA; if( nPMA>pSorter->mxKeysize ){ pSorter->mxKeysize = nPMA; } if( pSorter->list.aMemory ){ int nMin = pSorter->iMemory + nReq; if( nMin>pSorter->nMemory ){ u8 *aNew; int nNew = pSorter->nMemory * 2; while( nNew < nMin ) nNew = nNew*2; if( nNew > pSorter->mxPmaSize ) nNew = pSorter->mxPmaSize; if( nNew < nMin ) nNew = nMin; aNew = sqlite3Realloc(pSorter->list.aMemory, nNew); if( !aNew ) return SQLITE_NOMEM; pSorter->list.pList = (SorterRecord*)( aNew + ((u8*)pSorter->list.pList - pSorter->list.aMemory) ); pSorter->list.aMemory = aNew; pSorter->nMemory = nNew; } pNew = (SorterRecord*)&pSorter->list.aMemory[pSorter->iMemory]; pSorter->iMemory += ROUND8(nReq); pNew->u.iNext = (u8*)(pSorter->list.pList) - pSorter->list.aMemory; }else{ pNew = (SorterRecord *)sqlite3Malloc(nReq); if( pNew==0 ){ return SQLITE_NOMEM; } pNew->u.pNext = pSorter->list.pList; } memcpy(SRVAL(pNew), pVal->z, pVal->n); pNew->nVal = pVal->n; pSorter->list.pList = pNew; return rc; } /* ** Read keys from pIncr->pMerger and populate pIncr->aFile[1]. The format ** of the data stored in aFile[1] is the same as that used by regular PMAs, ** except that the number-of-bytes varint is omitted from the start. */ static int vdbeIncrPopulate(IncrMerger *pIncr){ int rc = SQLITE_OK; |
︙ | ︙ | |||
1663 1664 1665 1666 1667 1668 1669 | rc2 = vdbePmaWriterFinish(&writer, &pOut->iEof); if( rc==SQLITE_OK ) rc = rc2; vdbeSorterPopulateDebug(pIncr->pTask, "exit"); return rc; } | | | > > < < < < < < < > | > | | < < < < < | < < | < < | > > | < < < < < < < < < < < < < < < | 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 | rc2 = vdbePmaWriterFinish(&writer, &pOut->iEof); if( rc==SQLITE_OK ) rc = rc2; vdbeSorterPopulateDebug(pIncr->pTask, "exit"); return rc; } static void *vdbeIncrPopulateThread(void *pCtx){ IncrMerger *pIncr = (IncrMerger*)pCtx; void *pRet = SQLITE_INT_TO_PTR( vdbeIncrPopulate(pIncr) ); pIncr->thread.bDone = 1; return pRet; } #if SQLITE_MAX_WORKER_THREADS>0 static int vdbeIncrBgPopulate(IncrMerger *pIncr){ void *pCtx = (void*)pIncr; assert( pIncr->bUseThread ); return vdbeSorterCreateThread(&pIncr->thread, vdbeIncrPopulateThread, pCtx); } #endif static int vdbeIncrSwap(IncrMerger *pIncr){ int rc = SQLITE_OK; #if SQLITE_MAX_WORKER_THREADS>0 if( pIncr->bUseThread ){ rc = vdbeSorterJoinThread(pIncr->pTask, &pIncr->thread); if( rc==SQLITE_OK ){ SorterFile f0 = pIncr->aFile[0]; pIncr->aFile[0] = pIncr->aFile[1]; pIncr->aFile[1] = f0; } if( rc==SQLITE_OK ){ if( pIncr->aFile[0].iEof==pIncr->iStartOff ){ pIncr->bEof = 1; }else{ rc = vdbeIncrBgPopulate(pIncr); } } }else #endif { rc = vdbeIncrPopulate(pIncr); pIncr->aFile[0] = pIncr->aFile[1]; if( pIncr->aFile[0].iEof==pIncr->iStartOff ){ pIncr->bEof = 1; } } return rc; } static void vdbeIncrFree(IncrMerger *pIncr){ if( pIncr ){ #if SQLITE_MAX_WORKER_THREADS>0 vdbeSorterJoinThread(pIncr->pTask, &pIncr->thread); if( pIncr->bUseThread ){ if( pIncr->aFile[0].pFd ) sqlite3OsCloseFree(pIncr->aFile[0].pFd); if( pIncr->aFile[1].pFd ) sqlite3OsCloseFree(pIncr->aFile[1].pFd); } #endif vdbeMergeEngineFree(pIncr->pMerger); sqlite3_free(pIncr); } } static IncrMerger *vdbeIncrNew(SortSubtask *pTask, MergeEngine *pMerger){ IncrMerger *pIncr = sqlite3_malloc(sizeof(IncrMerger)); if( pIncr ){ memset(pIncr, 0, sizeof(IncrMerger)); pIncr->pMerger = pMerger; pIncr->pTask = pTask; pIncr->mxSz = MAX(pTask->pSorter->mxKeysize+9,pTask->pSorter->mxPmaSize/2); pTask->file2.iEof += pIncr->mxSz; } return pIncr; } static void vdbeIncrSetThreads(IncrMerger *pIncr, int bUseThread){ if( bUseThread ){ pIncr->bUseThread = 1; |
︙ | ︙ | |||
1780 1781 1782 1783 1784 1785 1786 | SortSubtask *pTask = pIncr->pTask; int i; MergeEngine *pMerger = pIncr->pMerger; for(i=0; rc==SQLITE_OK && i<pMerger->nTree; i++){ rc = vdbeIncrInit2(&pMerger->aIter[i]); } | < < < | 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 | SortSubtask *pTask = pIncr->pTask; int i; MergeEngine *pMerger = pIncr->pMerger; for(i=0; rc==SQLITE_OK && i<pMerger->nTree; i++){ rc = vdbeIncrInit2(&pMerger->aIter[i]); } /* Set up the required files for pIncr */ if( rc==SQLITE_OK ){ if( pIncr->bUseThread==0 ){ if( pTask->file2.pFd==0 ){ rc = vdbeSorterOpenTempFile(pTask->db->pVfs, &pTask->file2.pFd); assert( pTask->file2.iEof>0 ); |
︙ | ︙ | |||
1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 | }else{ rc = vdbeSorterOpenTempFile(pTask->db->pVfs, &pIncr->aFile[0].pFd); if( rc==SQLITE_OK ){ rc = vdbeSorterOpenTempFile(pTask->db->pVfs, &pIncr->aFile[1].pFd); } } } if( rc==SQLITE_OK && pIncr->bUseThread ){ rc = vdbeIncrBgPopulate(pIncr); } if( rc==SQLITE_OK ){ rc = vdbePmaReaderNext(pIter); | > > > > | 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 | }else{ rc = vdbeSorterOpenTempFile(pTask->db->pVfs, &pIncr->aFile[0].pFd); if( rc==SQLITE_OK ){ rc = vdbeSorterOpenTempFile(pTask->db->pVfs, &pIncr->aFile[1].pFd); } } } for(i=pMerger->nTree-1; rc==SQLITE_OK && i>0; i--){ rc = vdbeSorterDoCompare(pIncr->pTask, pMerger, i); } if( rc==SQLITE_OK && pIncr->bUseThread ){ rc = vdbeIncrBgPopulate(pIncr); } if( rc==SQLITE_OK ){ rc = vdbePmaReaderNext(pIter); |
︙ | ︙ | |||
1994 1995 1996 1997 1998 1999 2000 | vdbeIncrSetThreads(pIncr, pSorter->bUseThreads); assert( pIncr->pTask!=pLast ); } } } } } | | < < | 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 | vdbeIncrSetThreads(pIncr, pSorter->bUseThreads); assert( pIncr->pTask!=pLast ); } } } } } if( rc==SQLITE_OK ) rc = vdbeIncrInit2(pIter); sqlite3_free(aMerge); return rc; } /* |
︙ | ︙ | |||
2018 2019 2020 2021 2022 2023 2024 | assert( pSorter ); /* If no data has been written to disk, then do not do so now. Instead, ** sort the VdbeSorter.pRecord list. The vdbe layer will read data directly ** from the in-memory list. */ if( pSorter->bUsePMA==0 ){ | | < < < < < | < < < | | | 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 | assert( pSorter ); /* If no data has been written to disk, then do not do so now. Instead, ** sort the VdbeSorter.pRecord list. The vdbe layer will read data directly ** from the in-memory list. */ if( pSorter->bUsePMA==0 ){ if( pSorter->list.pList ){ *pbEof = 0; rc = vdbeSorterSort(&pSorter->aTask[0], &pSorter->list); }else{ *pbEof = 1; } return rc; } /* Write the current in-memory list to a PMA. */ if( pSorter->list.pList ){ rc = vdbeSorterFlushPMA(pSorter); } /* Join all threads */ rc = vdbeSorterJoinAll(pSorter, rc); vdbeSorterRewindDebug(db, "rewind"); |
︙ | ︙ | |||
2072 2073 2074 2075 2076 2077 2078 | VdbeSorter *pSorter = pCsr->pSorter; int rc; /* Return code */ if( pSorter->pReader ){ rc = vdbePmaReaderNext(pSorter->pReader); *pbEof = (pSorter->pReader->pFile==0); }else{ | | | | | | | | 2000 2001 2002 2003 2004 2005 2006 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 | VdbeSorter *pSorter = pCsr->pSorter; int rc; /* Return code */ if( pSorter->pReader ){ rc = vdbePmaReaderNext(pSorter->pReader); *pbEof = (pSorter->pReader->pFile==0); }else{ SorterRecord *pFree = pSorter->list.pList; pSorter->list.pList = pFree->u.pNext; pFree->u.pNext = 0; if( pSorter->list.aMemory==0 ) vdbeSorterRecordFree(db, pFree); *pbEof = !pSorter->list.pList; rc = SQLITE_OK; } return rc; } /* ** Return a pointer to a buffer owned by the sorter that contains the ** current key. */ static void *vdbeSorterRowkey( const VdbeSorter *pSorter, /* Sorter object */ int *pnKey /* OUT: Size of current key in bytes */ ){ void *pKey; if( pSorter->pReader ){ *pnKey = pSorter->pReader->nKey; pKey = pSorter->pReader->aKey; }else{ *pnKey = pSorter->list.pList->nVal; pKey = SRVAL(pSorter->list.pList); } return pKey; } /* ** Copy the current sorter key into the memory cell pOut. */ |
︙ | ︙ |