提交 8329284a 编写于 作者: H Hongze Cheng

refact for further dev

上级 7fe743d0
...@@ -586,6 +586,12 @@ struct SDelFWriter { ...@@ -586,6 +586,12 @@ struct SDelFWriter {
uint8_t *aBuf[1]; uint8_t *aBuf[1];
}; };
struct STsdbReadSnap {
SMemTable *pMem;
SMemTable *pIMem;
STsdbFS fs;
};
struct SDataFWriter { struct SDataFWriter {
STsdb *pTsdb; STsdb *pTsdb;
SDFileSet wSet; SDFileSet wSet;
...@@ -603,12 +609,6 @@ struct SDataFWriter { ...@@ -603,12 +609,6 @@ struct SDataFWriter {
uint8_t *aBuf[4]; uint8_t *aBuf[4];
}; };
struct STsdbReadSnap {
SMemTable *pMem;
SMemTable *pIMem;
STsdbFS fs;
};
struct SDataFReader { struct SDataFReader {
STsdb *pTsdb; STsdb *pTsdb;
SDFileSet *pSet; SDFileSet *pSet;
......
...@@ -23,7 +23,7 @@ typedef struct { ...@@ -23,7 +23,7 @@ typedef struct {
int64_t pgno; int64_t pgno;
} STsdbFD; } STsdbFD;
int32_t tsdbOpenFile(const char *path, int32_t opt, STsdbFD *pFD) { static int32_t tsdbOpenFile(const char *path, int32_t opt, STsdbFD *pFD) {
int32_t code = 0; int32_t code = 0;
pFD->pFD = taosOpenFile(path, opt); pFD->pFD = taosOpenFile(path, opt);
...@@ -45,12 +45,12 @@ _exit: ...@@ -45,12 +45,12 @@ _exit:
return code; return code;
} }
void tsdbCloseFile(STsdbFD *pFD) { static void tsdbCloseFile(STsdbFD *pFD) {
taosMemoryFree(pFD->pBuf); taosMemoryFree(pFD->pBuf);
taosCloseFile(&pFD->pFD); taosCloseFile(&pFD->pFD);
} }
int32_t tsdbSyncFile(STsdbFD *pFD) { static int32_t tsdbSyncFile(STsdbFD *pFD) {
int32_t code = 0; int32_t code = 0;
if (taosFsyncFile(pFD->pFD) < 0) { if (taosFsyncFile(pFD->pFD) < 0) {
...@@ -62,7 +62,7 @@ _exit: ...@@ -62,7 +62,7 @@ _exit:
return code; return code;
} }
int32_t tsdbWriteFile(STsdbFD *pFD, uint8_t *pBuf, int32_t nBuf, int64_t *offset) { static int32_t tsdbWriteFile(STsdbFD *pFD, uint8_t *pBuf, int32_t nBuf, int64_t *offset) {
int32_t code = 0; int32_t code = 0;
int32_t n = 0; int32_t n = 0;
...@@ -120,7 +120,7 @@ _exit: ...@@ -120,7 +120,7 @@ _exit:
return code; return code;
} }
int64_t tsdbReadFile(STsdbFD *pFD, int64_t offset, uint8_t *pBuf, int64_t count) { static int64_t tsdbReadFile(STsdbFD *pFD, int64_t offset, uint8_t *pBuf, int64_t count) {
int32_t code = 0; int32_t code = 0;
int64_t pgno = offset / pFD->szPage; int64_t pgno = offset / pFD->szPage;
...@@ -147,1219 +147,1195 @@ _exit: ...@@ -147,1219 +147,1195 @@ _exit:
return code; return code;
} }
// SDataFReader ==================================================== // SDataFWriter ====================================================
int32_t tsdbDataFReaderOpen(SDataFReader **ppReader, STsdb *pTsdb, SDFileSet *pSet) { int32_t tsdbDataFWriterOpen(SDataFWriter **ppWriter, STsdb *pTsdb, SDFileSet *pSet) {
int32_t code = 0; int32_t code = 0;
SDataFReader *pReader; int32_t flag;
int64_t n;
SDataFWriter *pWriter = NULL;
char fname[TSDB_FILENAME_LEN]; char fname[TSDB_FILENAME_LEN];
char hdr[TSDB_FHDR_SIZE] = {0};
// alloc // alloc
pReader = (SDataFReader *)taosMemoryCalloc(1, sizeof(*pReader)); pWriter = taosMemoryCalloc(1, sizeof(*pWriter));
if (pReader == NULL) { if (pWriter == NULL) {
code = TSDB_CODE_OUT_OF_MEMORY; code = TSDB_CODE_OUT_OF_MEMORY;
goto _err; goto _err;
} }
pReader->pTsdb = pTsdb; pWriter->pTsdb = pTsdb;
pReader->pSet = pSet; pWriter->wSet = (SDFileSet){
.diskId = pSet->diskId,
.fid = pSet->fid,
.pHeadF = &pWriter->fHead,
.pDataF = &pWriter->fData,
.pSmaF = &pWriter->fSma,
.nSstF = pSet->nSstF //
};
pWriter->fHead = *pSet->pHeadF;
pWriter->fData = *pSet->pDataF;
pWriter->fSma = *pSet->pSmaF;
for (int8_t iSst = 0; iSst < pSet->nSstF; iSst++) {
pWriter->wSet.aSstF[iSst] = &pWriter->fSst[iSst];
pWriter->fSst[iSst] = *pSet->aSstF[iSst];
}
// open impl
// head // head
tsdbHeadFileName(pTsdb, pSet->diskId, pSet->fid, pSet->pHeadF, fname); flag = TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC;
pReader->pHeadFD = taosOpenFile(fname, TD_FILE_READ); tsdbHeadFileName(pTsdb, pWriter->wSet.diskId, pWriter->wSet.fid, &pWriter->fHead, fname);
if (pReader->pHeadFD == NULL) { pWriter->pHeadFD = taosOpenFile(fname, flag);
if (pWriter->pHeadFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
// data n = taosWriteFile(pWriter->pHeadFD, hdr, TSDB_FHDR_SIZE);
tsdbDataFileName(pTsdb, pSet->diskId, pSet->fid, pSet->pDataF, fname); if (n < 0) {
pReader->pDataFD = taosOpenFile(fname, TD_FILE_READ);
if (pReader->pDataFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
// sma ASSERT(n == TSDB_FHDR_SIZE);
tsdbSmaFileName(pTsdb, pSet->diskId, pSet->fid, pSet->pSmaF, fname);
pReader->pSmaFD = taosOpenFile(fname, TD_FILE_READ); pWriter->fHead.size += TSDB_FHDR_SIZE;
if (pReader->pSmaFD == NULL) {
// data
if (pWriter->fData.size == 0) {
flag = TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC;
} else {
flag = TD_FILE_WRITE;
}
tsdbDataFileName(pTsdb, pWriter->wSet.diskId, pWriter->wSet.fid, &pWriter->fData, fname);
pWriter->pDataFD = taosOpenFile(fname, flag);
if (pWriter->pDataFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
if (pWriter->fData.size == 0) {
// sst n = taosWriteFile(pWriter->pDataFD, hdr, TSDB_FHDR_SIZE);
for (int32_t iSst = 0; iSst < pSet->nSstF; iSst++) { if (n < 0) {
tsdbSstFileName(pTsdb, pSet->diskId, pSet->fid, pSet->aSstF[iSst], fname);
pReader->aLastFD[iSst] = taosOpenFile(fname, TD_FILE_READ);
if (pReader->aLastFD[iSst] == NULL) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
}
*ppReader = pReader; pWriter->fData.size += TSDB_FHDR_SIZE;
return code; } else {
n = taosLSeekFile(pWriter->pDataFD, 0, SEEK_END);
_err: if (n < 0) {
tsdbError("vgId:%d, tsdb data file reader open failed since %s", TD_VID(pTsdb->pVnode), tstrerror(code)); code = TAOS_SYSTEM_ERROR(errno);
*ppReader = NULL; goto _err;
return code; }
}
int32_t tsdbDataFReaderClose(SDataFReader **ppReader) { ASSERT(n == pWriter->fData.size);
int32_t code = 0; }
if (*ppReader == NULL) goto _exit;
// head // sma
if (taosCloseFile(&(*ppReader)->pHeadFD) < 0) { if (pWriter->fSma.size == 0) {
flag = TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC;
} else {
flag = TD_FILE_WRITE;
}
tsdbSmaFileName(pTsdb, pWriter->wSet.diskId, pWriter->wSet.fid, &pWriter->fSma, fname);
pWriter->pSmaFD = taosOpenFile(fname, flag);
if (pWriter->pSmaFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
if (pWriter->fSma.size == 0) {
// data n = taosWriteFile(pWriter->pSmaFD, hdr, TSDB_FHDR_SIZE);
if (taosCloseFile(&(*ppReader)->pDataFD) < 0) { if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
// sma pWriter->fSma.size += TSDB_FHDR_SIZE;
if (taosCloseFile(&(*ppReader)->pSmaFD) < 0) { } else {
n = taosLSeekFile(pWriter->pSmaFD, 0, SEEK_END);
if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
ASSERT(n == pWriter->fSma.size);
}
// sst // sst
for (int32_t iSst = 0; iSst < (*ppReader)->pSet->nSstF; iSst++) { ASSERT(pWriter->fSst[pSet->nSstF - 1].size == 0);
if (taosCloseFile(&(*ppReader)->aLastFD[iSst]) < 0) { flag = TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC;
tsdbSstFileName(pTsdb, pWriter->wSet.diskId, pWriter->wSet.fid, &pWriter->fSst[pSet->nSstF - 1], fname);
pWriter->pLastFD = taosOpenFile(fname, flag);
if (pWriter->pLastFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
n = taosWriteFile(pWriter->pLastFD, hdr, TSDB_FHDR_SIZE);
if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
} }
pWriter->fSst[pWriter->wSet.nSstF - 1].size += TSDB_FHDR_SIZE;
for (int32_t iBuf = 0; iBuf < sizeof((*ppReader)->aBuf) / sizeof(uint8_t *); iBuf++) { *ppWriter = pWriter;
tFree((*ppReader)->aBuf[iBuf]);
}
taosMemoryFree(*ppReader);
_exit:
*ppReader = NULL;
return code; return code;
_err: _err:
tsdbError("vgId:%d, data file reader close failed since %s", TD_VID((*ppReader)->pTsdb->pVnode), tstrerror(code)); tsdbError("vgId:%d, tsdb data file writer open failed since %s", TD_VID(pTsdb->pVnode), tstrerror(code));
*ppWriter = NULL;
return code; return code;
} }
int32_t tsdbReadBlockIdx(SDataFReader *pReader, SArray *aBlockIdx) { int32_t tsdbDataFWriterClose(SDataFWriter **ppWriter, int8_t sync) {
int32_t code = 0; int32_t code = 0;
int64_t offset = pReader->pSet->pHeadF->offset; STsdb *pTsdb = NULL;
int64_t size = pReader->pSet->pHeadF->size - offset;
int64_t n;
uint32_t delimiter;
taosArrayClear(aBlockIdx);
if (size == 0) {
goto _exit;
}
// alloc if (*ppWriter == NULL) goto _exit;
code = tRealloc(&pReader->aBuf[0], size);
if (code) goto _err;
// seek pTsdb = (*ppWriter)->pTsdb;
if (taosLSeekFile(pReader->pHeadFD, offset, SEEK_SET) < 0) { if (sync) {
if (taosFsyncFile((*ppWriter)->pHeadFD) < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
// read if (taosFsyncFile((*ppWriter)->pDataFD) < 0) {
n = taosReadFile(pReader->pHeadFD, pReader->aBuf[0], size);
if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} else if (n < size) {
code = TSDB_CODE_FILE_CORRUPTED;
goto _err;
} }
// check if (taosFsyncFile((*ppWriter)->pSmaFD) < 0) {
if (!taosCheckChecksumWhole(pReader->aBuf[0], size)) { code = TAOS_SYSTEM_ERROR(errno);
code = TSDB_CODE_FILE_CORRUPTED;
goto _err; goto _err;
} }
// decode if (taosFsyncFile((*ppWriter)->pLastFD) < 0) {
n = 0; code = TAOS_SYSTEM_ERROR(errno);
n = tGetU32(pReader->aBuf[0] + n, &delimiter); goto _err;
ASSERT(delimiter == TSDB_FILE_DLMT); }
}
while (n < size - sizeof(TSCKSUM)) { if (taosCloseFile(&(*ppWriter)->pHeadFD) < 0) {
SBlockIdx blockIdx; code = TAOS_SYSTEM_ERROR(errno);
n += tGetBlockIdx(pReader->aBuf[0] + n, &blockIdx); goto _err;
}
if (taosArrayPush(aBlockIdx, &blockIdx) == NULL) { if (taosCloseFile(&(*ppWriter)->pDataFD) < 0) {
code = TSDB_CODE_OUT_OF_MEMORY; code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
if (taosCloseFile(&(*ppWriter)->pSmaFD) < 0) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
} }
ASSERT(n + sizeof(TSCKSUM) == size); if (taosCloseFile(&(*ppWriter)->pLastFD) < 0) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
}
for (int32_t iBuf = 0; iBuf < sizeof((*ppWriter)->aBuf) / sizeof(uint8_t *); iBuf++) {
tFree((*ppWriter)->aBuf[iBuf]);
}
taosMemoryFree(*ppWriter);
_exit: _exit:
*ppWriter = NULL;
return code; return code;
_err: _err:
tsdbError("vgId:%d, read block idx failed since %s", TD_VID(pReader->pTsdb->pVnode), tstrerror(code)); tsdbError("vgId:%d, data file writer close failed since %s", TD_VID(pTsdb->pVnode), tstrerror(code));
return code; return code;
} }
int32_t tsdbReadSstBlk(SDataFReader *pReader, int32_t iSst, SArray *aSstBlk) { int32_t tsdbUpdateDFileSetHeader(SDataFWriter *pWriter) {
int32_t code = 0; int32_t code = 0;
int64_t offset = pReader->pSet->aSstF[iSst]->offset;
int64_t size = pReader->pSet->aSstF[iSst]->size - offset;
int64_t n; int64_t n;
uint32_t delimiter; char hdr[TSDB_FHDR_SIZE];
taosArrayClear(aSstBlk);
if (size == 0) {
goto _exit;
}
// alloc // head ==============
code = tRealloc(&pReader->aBuf[0], size); memset(hdr, 0, TSDB_FHDR_SIZE);
if (code) goto _err; tPutHeadFile(hdr, &pWriter->fHead);
taosCalcChecksumAppend(0, hdr, TSDB_FHDR_SIZE);
// seek n = taosLSeekFile(pWriter->pHeadFD, 0, SEEK_SET);
if (taosLSeekFile(pReader->aLastFD[iSst], offset, SEEK_SET) < 0) { if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
// read n = taosWriteFile(pWriter->pHeadFD, hdr, TSDB_FHDR_SIZE);
n = taosReadFile(pReader->aLastFD[iSst], pReader->aBuf[0], size);
if (n < 0) { if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} else if (n < size) { }
code = TSDB_CODE_FILE_CORRUPTED;
// data ==============
memset(hdr, 0, TSDB_FHDR_SIZE);
tPutDataFile(hdr, &pWriter->fData);
taosCalcChecksumAppend(0, hdr, TSDB_FHDR_SIZE);
n = taosLSeekFile(pWriter->pDataFD, 0, SEEK_SET);
if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
// check n = taosWriteFile(pWriter->pDataFD, hdr, TSDB_FHDR_SIZE);
if (!taosCheckChecksumWhole(pReader->aBuf[0], size)) { if (n < 0) {
code = TSDB_CODE_FILE_CORRUPTED; code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
// decode // sma ==============
n = 0; memset(hdr, 0, TSDB_FHDR_SIZE);
n = tGetU32(pReader->aBuf[0] + n, &delimiter); tPutSmaFile(hdr, &pWriter->fSma);
ASSERT(delimiter == TSDB_FILE_DLMT); taosCalcChecksumAppend(0, hdr, TSDB_FHDR_SIZE);
while (n < size - sizeof(TSCKSUM)) { n = taosLSeekFile(pWriter->pSmaFD, 0, SEEK_SET);
SSstBlk blockl; if (n < 0) {
n += tGetSstBlk(pReader->aBuf[0] + n, &blockl); code = TAOS_SYSTEM_ERROR(errno);
goto _err;
}
if (taosArrayPush(aSstBlk, &blockl) == NULL) { n = taosWriteFile(pWriter->pSmaFD, hdr, TSDB_FHDR_SIZE);
code = TSDB_CODE_OUT_OF_MEMORY; if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
// sst ==============
memset(hdr, 0, TSDB_FHDR_SIZE);
tPutSstFile(hdr, &pWriter->fSst[pWriter->wSet.nSstF - 1]);
taosCalcChecksumAppend(0, hdr, TSDB_FHDR_SIZE);
n = taosLSeekFile(pWriter->pLastFD, 0, SEEK_SET);
if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
} }
ASSERT(n + sizeof(TSCKSUM) == size); n = taosWriteFile(pWriter->pLastFD, hdr, TSDB_FHDR_SIZE);
if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
}
_exit:
return code; return code;
_err: _err:
tsdbError("vgId:%d read blockl failed since %s", TD_VID(pReader->pTsdb->pVnode), tstrerror(code)); tsdbError("vgId:%d, update DFileSet header failed since %s", TD_VID(pWriter->pTsdb->pVnode), tstrerror(code));
return code; return code;
} }
int32_t tsdbReadBlock(SDataFReader *pReader, SBlockIdx *pBlockIdx, SMapData *mBlock) { int32_t tsdbWriteBlockIdx(SDataFWriter *pWriter, SArray *aBlockIdx) {
int32_t code = 0; int32_t code = 0;
int64_t offset = pBlockIdx->offset; SHeadFile *pHeadFile = &pWriter->fHead;
int64_t size = pBlockIdx->size; int64_t size = 0;
int64_t n; int64_t n;
int64_t tn;
// alloc // check
code = tRealloc(&pReader->aBuf[0], size); if (taosArrayGetSize(aBlockIdx) == 0) {
if (code) goto _err; pHeadFile->offset = pHeadFile->size;
goto _exit;
// seek
if (taosLSeekFile(pReader->pHeadFD, offset, SEEK_SET) < 0) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
} }
// read // prepare
n = taosReadFile(pReader->pHeadFD, pReader->aBuf[0], size); size = sizeof(uint32_t);
if (n < 0) { for (int32_t iBlockIdx = 0; iBlockIdx < taosArrayGetSize(aBlockIdx); iBlockIdx++) {
code = TAOS_SYSTEM_ERROR(errno); size += tPutBlockIdx(NULL, taosArrayGet(aBlockIdx, iBlockIdx));
goto _err;
} else if (n < size) {
code = TSDB_CODE_FILE_CORRUPTED;
goto _err;
} }
size += sizeof(TSCKSUM);
// check // alloc
if (!taosCheckChecksumWhole(pReader->aBuf[0], size)) { code = tRealloc(&pWriter->aBuf[0], size);
code = TSDB_CODE_FILE_CORRUPTED; if (code) goto _err;
goto _err;
}
// decode // build
n = 0; n = 0;
n = tPutU32(pWriter->aBuf[0] + n, TSDB_FILE_DLMT);
for (int32_t iBlockIdx = 0; iBlockIdx < taosArrayGetSize(aBlockIdx); iBlockIdx++) {
n += tPutBlockIdx(pWriter->aBuf[0] + n, taosArrayGet(aBlockIdx, iBlockIdx));
}
taosCalcChecksumAppend(0, pWriter->aBuf[0], size);
uint32_t delimiter; ASSERT(n + sizeof(TSCKSUM) == size);
n += tGetU32(pReader->aBuf[0] + n, &delimiter);
ASSERT(delimiter == TSDB_FILE_DLMT);
tn = tGetMapData(pReader->aBuf[0] + n, mBlock); // write
if (tn < 0) { n = taosWriteFile(pWriter->pHeadFD, pWriter->aBuf[0], size);
code = TSDB_CODE_OUT_OF_MEMORY; if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
n += tn;
ASSERT(n + sizeof(TSCKSUM) == size);
// update
pHeadFile->offset = pHeadFile->size;
pHeadFile->size += size;
_exit:
tsdbTrace("vgId:%d write block idx, offset:%" PRId64 " size:%" PRId64 " nBlockIdx:%d", TD_VID(pWriter->pTsdb->pVnode),
pHeadFile->offset, size, taosArrayGetSize(aBlockIdx));
return code; return code;
_err: _err:
tsdbError("vgId:%d, read block failed since %s", TD_VID(pReader->pTsdb->pVnode), tstrerror(code)); tsdbError("vgId:%d, write block idx failed since %s", TD_VID(pWriter->pTsdb->pVnode), tstrerror(code));
return code; return code;
} }
int32_t tsdbReadBlockSma(SDataFReader *pReader, SDataBlk *pDataBlk, SArray *aColumnDataAgg) { int32_t tsdbWriteBlock(SDataFWriter *pWriter, SMapData *mBlock, SBlockIdx *pBlockIdx) {
int32_t code = 0; int32_t code = 0;
SSmaInfo *pSmaInfo = &pDataBlk->smaInfo; SHeadFile *pHeadFile = &pWriter->fHead;
int64_t size;
ASSERT(pSmaInfo->size > 0); int64_t n;
taosArrayClear(aColumnDataAgg); ASSERT(mBlock->nItem > 0);
// alloc // alloc
int32_t size = pSmaInfo->size + sizeof(TSCKSUM); size = sizeof(uint32_t) + tPutMapData(NULL, mBlock) + sizeof(TSCKSUM);
code = tRealloc(&pReader->aBuf[0], size); code = tRealloc(&pWriter->aBuf[0], size);
if (code) goto _err; if (code) goto _err;
// seek // build
int64_t n = taosLSeekFile(pReader->pSmaFD, pSmaInfo->offset, SEEK_SET); n = 0;
if (n < 0) { n += tPutU32(pWriter->aBuf[0] + n, TSDB_FILE_DLMT);
code = TAOS_SYSTEM_ERROR(errno); n += tPutMapData(pWriter->aBuf[0] + n, mBlock);
goto _err; taosCalcChecksumAppend(0, pWriter->aBuf[0], size);
} else if (n < pSmaInfo->offset) {
code = TSDB_CODE_FILE_CORRUPTED;
goto _err;
}
// read ASSERT(n + sizeof(TSCKSUM) == size);
n = taosReadFile(pReader->pSmaFD, pReader->aBuf[0], size);
// write
n = taosWriteFile(pWriter->pHeadFD, pWriter->aBuf[0], size);
if (n < 0) { if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} else if (n < size) {
code = TSDB_CODE_FILE_CORRUPTED;
goto _err;
}
// check
if (!taosCheckChecksumWhole(pReader->aBuf[0], size)) {
code = TSDB_CODE_FILE_CORRUPTED;
goto _err;
} }
// decode // update
n = 0; pBlockIdx->offset = pHeadFile->size;
while (n < pSmaInfo->size) { pBlockIdx->size = size;
SColumnDataAgg sma; pHeadFile->size += size;
n += tGetColumnDataAgg(pReader->aBuf[0] + n, &sma);
if (taosArrayPush(aColumnDataAgg, &sma) == NULL) {
code = TSDB_CODE_OUT_OF_MEMORY;
goto _err;
}
}
tsdbTrace("vgId:%d, write block, file ID:%d commit ID:%d suid:%" PRId64 " uid:%" PRId64 " offset:%" PRId64
" size:%" PRId64 " nItem:%d",
TD_VID(pWriter->pTsdb->pVnode), pWriter->wSet.fid, pHeadFile->commitID, pBlockIdx->suid, pBlockIdx->uid,
pBlockIdx->offset, pBlockIdx->size, mBlock->nItem);
return code; return code;
_err: _err:
tsdbError("vgId:%d tsdb read block sma failed since %s", TD_VID(pReader->pTsdb->pVnode), tstrerror(code)); tsdbError("vgId:%d, write block failed since %s", TD_VID(pWriter->pTsdb->pVnode), tstrerror(code));
return code; return code;
} }
static int32_t tsdbReadBlockDataImpl(SDataFReader *pReader, SBlockInfo *pBlkInfo, int8_t fromLast, int32_t tsdbWriteSstBlk(SDataFWriter *pWriter, SArray *aSstBlk) {
SBlockData *pBlockData) {
int32_t code = 0; int32_t code = 0;
SSstFile *pSstFile = &pWriter->fSst[pWriter->wSet.nSstF - 1];
int64_t size;
int64_t n;
tBlockDataClear(pBlockData); // check
if (taosArrayGetSize(aSstBlk) == 0) {
pSstFile->offset = pSstFile->size;
goto _exit;
}
TdFilePtr pFD = fromLast ? pReader->aLastFD[0] : pReader->pDataFD; // (todo) // size
size = sizeof(uint32_t); // TSDB_FILE_DLMT
for (int32_t iBlockL = 0; iBlockL < taosArrayGetSize(aSstBlk); iBlockL++) {
size += tPutSstBlk(NULL, taosArrayGet(aSstBlk, iBlockL));
}
size += sizeof(TSCKSUM);
// uid + version + tskey // alloc
code = tsdbReadAndCheck(pFD, pBlkInfo->offset, &pReader->aBuf[0], pBlkInfo->szKey, 1); code = tRealloc(&pWriter->aBuf[0], size);
if (code) goto _err; if (code) goto _err;
SDiskDataHdr hdr;
uint8_t *p = pReader->aBuf[0] + tGetDiskDataHdr(pReader->aBuf[0], &hdr);
ASSERT(hdr.delimiter == TSDB_FILE_DLMT); // encode
ASSERT(pBlockData->suid == hdr.suid); n = 0;
ASSERT(pBlockData->uid == hdr.uid); n += tPutU32(pWriter->aBuf[0] + n, TSDB_FILE_DLMT);
for (int32_t iBlockL = 0; iBlockL < taosArrayGetSize(aSstBlk); iBlockL++) {
n += tPutSstBlk(pWriter->aBuf[0] + n, taosArrayGet(aSstBlk, iBlockL));
}
taosCalcChecksumAppend(0, pWriter->aBuf[0], size);
pBlockData->nRow = hdr.nRow; ASSERT(n + sizeof(TSCKSUM) == size);
// uid // write
if (hdr.uid == 0) { n = taosWriteFile(pWriter->pLastFD, pWriter->aBuf[0], size);
ASSERT(hdr.szUid); if (n < 0) {
code = tsdbDecmprData(p, hdr.szUid, TSDB_DATA_TYPE_BIGINT, hdr.cmprAlg, (uint8_t **)&pBlockData->aUid, code = TAOS_SYSTEM_ERROR(errno);
sizeof(int64_t) * hdr.nRow, &pReader->aBuf[1]); goto _err;
if (code) goto _err;
} else {
ASSERT(!hdr.szUid);
} }
p += hdr.szUid;
// version // update
code = tsdbDecmprData(p, hdr.szVer, TSDB_DATA_TYPE_BIGINT, hdr.cmprAlg, (uint8_t **)&pBlockData->aVersion, pSstFile->offset = pSstFile->size;
sizeof(int64_t) * hdr.nRow, &pReader->aBuf[1]); pSstFile->size += size;
if (code) goto _err;
p += hdr.szVer;
// TSKEY
code = tsdbDecmprData(p, hdr.szKey, TSDB_DATA_TYPE_TIMESTAMP, hdr.cmprAlg, (uint8_t **)&pBlockData->aTSKEY,
sizeof(TSKEY) * hdr.nRow, &pReader->aBuf[1]);
if (code) goto _err;
p += hdr.szKey;
ASSERT(p - pReader->aBuf[0] == pBlkInfo->szKey - sizeof(TSCKSUM)); _exit:
tsdbTrace("vgId:%d tsdb write blockl, loffset:%" PRId64 " size:%" PRId64, TD_VID(pWriter->pTsdb->pVnode),
pSstFile->offset, size);
return code;
// read and decode columns _err:
if (taosArrayGetSize(pBlockData->aIdx) == 0) goto _exit; tsdbError("vgId:%d tsdb write blockl failed since %s", TD_VID(pWriter->pTsdb->pVnode), tstrerror(code));
return code;
}
if (hdr.szBlkCol > 0) { static int32_t tsdbWriteBlockSma(SDataFWriter *pWriter, SBlockData *pBlockData, SSmaInfo *pSmaInfo) {
int64_t offset = pBlkInfo->offset + pBlkInfo->szKey; int32_t code = 0;
code = tsdbReadAndCheck(pFD, offset, &pReader->aBuf[0], hdr.szBlkCol + sizeof(TSCKSUM), 1);
if (code) goto _err;
}
SBlockCol blockCol = {.cid = 0}; pSmaInfo->offset = 0;
SBlockCol *pBlockCol = &blockCol; pSmaInfo->size = 0;
int32_t n = 0;
// encode
for (int32_t iColData = 0; iColData < taosArrayGetSize(pBlockData->aIdx); iColData++) { for (int32_t iColData = 0; iColData < taosArrayGetSize(pBlockData->aIdx); iColData++) {
SColData *pColData = tBlockDataGetColDataByIdx(pBlockData, iColData); SColData *pColData = tBlockDataGetColDataByIdx(pBlockData, iColData);
while (pBlockCol && pBlockCol->cid < pColData->cid) { if ((!pColData->smaOn) || IS_VAR_DATA_TYPE(pColData->type)) continue;
if (n < hdr.szBlkCol) {
n += tGetBlockCol(pReader->aBuf[0] + n, pBlockCol);
} else {
ASSERT(n == hdr.szBlkCol);
pBlockCol = NULL;
}
}
if (pBlockCol == NULL || pBlockCol->cid > pColData->cid) { SColumnDataAgg sma;
// add a lot of NONE tsdbCalcColDataSMA(pColData, &sma);
for (int32_t iRow = 0; iRow < hdr.nRow; iRow++) {
code = tColDataAppendValue(pColData, &COL_VAL_NONE(pColData->cid, pColData->type));
if (code) goto _err;
}
} else {
ASSERT(pBlockCol->type == pColData->type);
ASSERT(pBlockCol->flag && pBlockCol->flag != HAS_NONE);
if (pBlockCol->flag == HAS_NULL) { code = tRealloc(&pWriter->aBuf[0], pSmaInfo->size + tPutColumnDataAgg(NULL, &sma));
// add a lot of NULL
for (int32_t iRow = 0; iRow < hdr.nRow; iRow++) {
code = tColDataAppendValue(pColData, &COL_VAL_NULL(pBlockCol->cid, pBlockCol->type));
if (code) goto _err; if (code) goto _err;
pSmaInfo->size += tPutColumnDataAgg(pWriter->aBuf[0] + pSmaInfo->size, &sma);
} }
} else {
// decode from binary
int64_t offset = pBlkInfo->offset + pBlkInfo->szKey + hdr.szBlkCol + sizeof(TSCKSUM) + pBlockCol->offset;
int32_t size = pBlockCol->szBitmap + pBlockCol->szOffset + pBlockCol->szValue + sizeof(TSCKSUM);
code = tsdbReadAndCheck(pFD, offset, &pReader->aBuf[1], size, 0); // write
if (code) goto _err; if (pSmaInfo->size) {
int32_t size = pSmaInfo->size + sizeof(TSCKSUM);
code = tsdbDecmprColData(pReader->aBuf[1], pBlockCol, hdr.cmprAlg, hdr.nRow, pColData, &pReader->aBuf[2]); code = tRealloc(&pWriter->aBuf[0], size);
if (code) goto _err; if (code) goto _err;
taosCalcChecksumAppend(0, pWriter->aBuf[0], size);
int64_t n = taosWriteFile(pWriter->pSmaFD, pWriter->aBuf[0], size);
if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
} }
}
pSmaInfo->offset = pWriter->fSma.size;
pWriter->fSma.size += size;
} }
_exit:
return code; return code;
_err: _err:
tsdbError("vgId:%d tsdb read block data impl failed since %s", TD_VID(pReader->pTsdb->pVnode), tstrerror(code)); tsdbError("vgId:%d tsdb write block sma failed since %s", TD_VID(pWriter->pTsdb->pVnode), tstrerror(code));
return code; return code;
} }
int32_t tsdbReadDataBlock(SDataFReader *pReader, SDataBlk *pDataBlk, SBlockData *pBlockData) { int32_t tsdbWriteBlockData(SDataFWriter *pWriter, SBlockData *pBlockData, SBlockInfo *pBlkInfo, SSmaInfo *pSmaInfo,
int8_t cmprAlg, int8_t toLast) {
int32_t code = 0; int32_t code = 0;
code = tsdbReadBlockDataImpl(pReader, &pDataBlk->aSubBlock[0], 0, pBlockData); ASSERT(pBlockData->nRow > 0);
if (code) goto _err;
if (pDataBlk->nSubBlock > 1) { pBlkInfo->offset = toLast ? pWriter->fSst[pWriter->wSet.nSstF - 1].size : pWriter->fData.size;
SBlockData bData1; pBlkInfo->szBlock = 0;
SBlockData bData2; pBlkInfo->szKey = 0;
// create int32_t aBufN[4] = {0};
code = tBlockDataCreate(&bData1); code = tCmprBlockData(pBlockData, cmprAlg, NULL, NULL, pWriter->aBuf, aBufN);
if (code) goto _err;
code = tBlockDataCreate(&bData2);
if (code) goto _err; if (code) goto _err;
// init // write =================
tBlockDataInitEx(&bData1, pBlockData); TdFilePtr pFD = toLast ? pWriter->pLastFD : pWriter->pDataFD;
tBlockDataInitEx(&bData2, pBlockData);
for (int32_t iSubBlock = 1; iSubBlock < pDataBlk->nSubBlock; iSubBlock++) { pBlkInfo->szKey = aBufN[3] + aBufN[2];
code = tsdbReadBlockDataImpl(pReader, &pDataBlk->aSubBlock[iSubBlock], 0, &bData1); pBlkInfo->szBlock = aBufN[0] + aBufN[1] + aBufN[2] + aBufN[3];
if (code) {
tBlockDataDestroy(&bData1, 1); int64_t n = taosWriteFile(pFD, pWriter->aBuf[3], aBufN[3]);
tBlockDataDestroy(&bData2, 1); if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
code = tBlockDataCopy(pBlockData, &bData2); n = taosWriteFile(pFD, pWriter->aBuf[2], aBufN[2]);
if (code) { if (n < 0) {
tBlockDataDestroy(&bData1, 1); code = TAOS_SYSTEM_ERROR(errno);
tBlockDataDestroy(&bData2, 1);
goto _err; goto _err;
} }
code = tBlockDataMerge(&bData1, &bData2, pBlockData); if (aBufN[1]) {
if (code) { n = taosWriteFile(pFD, pWriter->aBuf[1], aBufN[1]);
tBlockDataDestroy(&bData1, 1); if (n < 0) {
tBlockDataDestroy(&bData2, 1); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
} }
tBlockDataDestroy(&bData1, 1); if (aBufN[0]) {
tBlockDataDestroy(&bData2, 1); n = taosWriteFile(pFD, pWriter->aBuf[0], aBufN[0]);
if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
}
} }
return code; // update info
if (toLast) {
_err: pWriter->fSst[pWriter->wSet.nSstF - 1].size += pBlkInfo->szBlock;
tsdbError("vgId:%d tsdb read data block failed since %s", TD_VID(pReader->pTsdb->pVnode), tstrerror(code)); } else {
return code; pWriter->fData.size += pBlkInfo->szBlock;
} }
int32_t tsdbReadSstBlock(SDataFReader *pReader, int32_t iSst, SSstBlk *pSstBlk, SBlockData *pBlockData) {
int32_t code = 0;
code = tsdbReadBlockDataImpl(pReader, &pSstBlk->bInfo, 1, pBlockData); // ================= SMA ====================
if (pSmaInfo) {
code = tsdbWriteBlockSma(pWriter, pBlockData, pSmaInfo);
if (code) goto _err; if (code) goto _err;
}
_exit:
tsdbTrace("vgId:%d tsdb write block data, suid:%" PRId64 " uid:%" PRId64 " nRow:%d, offset:%" PRId64 " size:%d",
TD_VID(pWriter->pTsdb->pVnode), pBlockData->suid, pBlockData->uid, pBlockData->nRow, pBlkInfo->offset,
pBlkInfo->szBlock);
return code; return code;
_err: _err:
tsdbError("vgId:%d tsdb read last block failed since %s", TD_VID(pReader->pTsdb->pVnode), tstrerror(code)); tsdbError("vgId:%d tsdb write block data failed since %s", TD_VID(pWriter->pTsdb->pVnode), tstrerror(code));
return code;
}
int32_t tsdbReadSstBlockEx(SDataFReader *pReader, int32_t iSst, SSstBlk *pSstBlk, SBlockData *pBlockData) {
int32_t code = 0;
// read
code = tsdbReadAndCheck(pReader->aLastFD[iSst], pSstBlk->bInfo.offset, &pReader->aBuf[0], pSstBlk->bInfo.szBlock, 0);
if (code) goto _exit;
// decmpr
code = tDecmprBlockData(pReader->aBuf[0], pSstBlk->bInfo.szBlock, pBlockData, &pReader->aBuf[1]);
if (code) goto _exit;
_exit:
return code; return code;
} }
// SDataFWriter ==================================================== int32_t tsdbDFileSetCopy(STsdb *pTsdb, SDFileSet *pSetFrom, SDFileSet *pSetTo) {
int32_t tsdbDataFWriterOpen(SDataFWriter **ppWriter, STsdb *pTsdb, SDFileSet *pSet) {
int32_t code = 0; int32_t code = 0;
int32_t flag;
int64_t n; int64_t n;
SDataFWriter *pWriter = NULL; int64_t size;
char fname[TSDB_FILENAME_LEN]; TdFilePtr pOutFD = NULL; // TODO
char hdr[TSDB_FHDR_SIZE] = {0}; TdFilePtr PInFD = NULL; // TODO
char fNameFrom[TSDB_FILENAME_LEN];
char fNameTo[TSDB_FILENAME_LEN];
// alloc // head
pWriter = taosMemoryCalloc(1, sizeof(*pWriter)); tsdbHeadFileName(pTsdb, pSetFrom->diskId, pSetFrom->fid, pSetFrom->pHeadF, fNameFrom);
if (pWriter == NULL) { tsdbHeadFileName(pTsdb, pSetTo->diskId, pSetTo->fid, pSetTo->pHeadF, fNameTo);
code = TSDB_CODE_OUT_OF_MEMORY;
goto _err; pOutFD = taosOpenFile(fNameTo, TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC);
} if (pOutFD == NULL) {
pWriter->pTsdb = pTsdb; code = TAOS_SYSTEM_ERROR(errno);
pWriter->wSet = (SDFileSet){ goto _err;
.diskId = pSet->diskId,
.fid = pSet->fid,
.pHeadF = &pWriter->fHead,
.pDataF = &pWriter->fData,
.pSmaF = &pWriter->fSma,
.nSstF = pSet->nSstF //
};
pWriter->fHead = *pSet->pHeadF;
pWriter->fData = *pSet->pDataF;
pWriter->fSma = *pSet->pSmaF;
for (int8_t iSst = 0; iSst < pSet->nSstF; iSst++) {
pWriter->wSet.aSstF[iSst] = &pWriter->fSst[iSst];
pWriter->fSst[iSst] = *pSet->aSstF[iSst];
} }
// head PInFD = taosOpenFile(fNameFrom, TD_FILE_READ);
flag = TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC; if (PInFD == NULL) {
tsdbHeadFileName(pTsdb, pWriter->wSet.diskId, pWriter->wSet.fid, &pWriter->fHead, fname);
pWriter->pHeadFD = taosOpenFile(fname, flag);
if (pWriter->pHeadFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
n = taosWriteFile(pWriter->pHeadFD, hdr, TSDB_FHDR_SIZE); n = taosFSendFile(pOutFD, PInFD, 0, pSetFrom->pHeadF->size);
if (n < 0) { if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
taosCloseFile(&pOutFD);
ASSERT(n == TSDB_FHDR_SIZE); taosCloseFile(&PInFD);
pWriter->fHead.size += TSDB_FHDR_SIZE;
// data // data
if (pWriter->fData.size == 0) { tsdbDataFileName(pTsdb, pSetFrom->diskId, pSetFrom->fid, pSetFrom->pDataF, fNameFrom);
flag = TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC; tsdbDataFileName(pTsdb, pSetTo->diskId, pSetTo->fid, pSetTo->pDataF, fNameTo);
} else {
flag = TD_FILE_WRITE; pOutFD = taosOpenFile(fNameTo, TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC);
} if (pOutFD == NULL) {
tsdbDataFileName(pTsdb, pWriter->wSet.diskId, pWriter->wSet.fid, &pWriter->fData, fname);
pWriter->pDataFD = taosOpenFile(fname, flag);
if (pWriter->pDataFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
if (pWriter->fData.size == 0) {
n = taosWriteFile(pWriter->pDataFD, hdr, TSDB_FHDR_SIZE); PInFD = taosOpenFile(fNameFrom, TD_FILE_READ);
if (n < 0) { if (PInFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
pWriter->fData.size += TSDB_FHDR_SIZE; n = taosFSendFile(pOutFD, PInFD, 0, pSetFrom->pDataF->size);
} else {
n = taosLSeekFile(pWriter->pDataFD, 0, SEEK_END);
if (n < 0) { if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
taosCloseFile(&pOutFD);
taosCloseFile(&PInFD);
ASSERT(n == pWriter->fData.size); // sst
} tsdbSstFileName(pTsdb, pSetFrom->diskId, pSetFrom->fid, pSetFrom->aSstF[0], fNameFrom);
tsdbSstFileName(pTsdb, pSetTo->diskId, pSetTo->fid, pSetTo->aSstF[0], fNameTo);
// sma pOutFD = taosOpenFile(fNameTo, TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC);
if (pWriter->fSma.size == 0) { if (pOutFD == NULL) {
flag = TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC;
} else {
flag = TD_FILE_WRITE;
}
tsdbSmaFileName(pTsdb, pWriter->wSet.diskId, pWriter->wSet.fid, &pWriter->fSma, fname);
pWriter->pSmaFD = taosOpenFile(fname, flag);
if (pWriter->pSmaFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
if (pWriter->fSma.size == 0) {
n = taosWriteFile(pWriter->pSmaFD, hdr, TSDB_FHDR_SIZE); PInFD = taosOpenFile(fNameFrom, TD_FILE_READ);
if (n < 0) { if (PInFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
pWriter->fSma.size += TSDB_FHDR_SIZE; n = taosFSendFile(pOutFD, PInFD, 0, pSetFrom->aSstF[0]->size);
} else {
n = taosLSeekFile(pWriter->pSmaFD, 0, SEEK_END);
if (n < 0) { if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
taosCloseFile(&pOutFD);
taosCloseFile(&PInFD);
ASSERT(n == pWriter->fSma.size); // sma
tsdbSmaFileName(pTsdb, pSetFrom->diskId, pSetFrom->fid, pSetFrom->pSmaF, fNameFrom);
tsdbSmaFileName(pTsdb, pSetTo->diskId, pSetTo->fid, pSetTo->pSmaF, fNameTo);
pOutFD = taosOpenFile(fNameTo, TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC);
if (pOutFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
} }
// sst PInFD = taosOpenFile(fNameFrom, TD_FILE_READ);
ASSERT(pWriter->fSst[pSet->nSstF - 1].size == 0); if (PInFD == NULL) {
flag = TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC;
tsdbSstFileName(pTsdb, pWriter->wSet.diskId, pWriter->wSet.fid, &pWriter->fSst[pSet->nSstF - 1], fname);
pWriter->pLastFD = taosOpenFile(fname, flag);
if (pWriter->pLastFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
n = taosWriteFile(pWriter->pLastFD, hdr, TSDB_FHDR_SIZE);
n = taosFSendFile(pOutFD, PInFD, 0, pSetFrom->pSmaF->size);
if (n < 0) { if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
pWriter->fSst[pWriter->wSet.nSstF - 1].size += TSDB_FHDR_SIZE; taosCloseFile(&pOutFD);
taosCloseFile(&PInFD);
*ppWriter = pWriter;
return code; return code;
_err: _err:
tsdbError("vgId:%d, tsdb data file writer open failed since %s", TD_VID(pTsdb->pVnode), tstrerror(code)); tsdbError("vgId:%d, tsdb DFileSet copy failed since %s", TD_VID(pTsdb->pVnode), tstrerror(code));
*ppWriter = NULL;
return code; return code;
} }
int32_t tsdbDataFWriterClose(SDataFWriter **ppWriter, int8_t sync) { // SDataFReader ====================================================
int32_t tsdbDataFReaderOpen(SDataFReader **ppReader, STsdb *pTsdb, SDFileSet *pSet) {
int32_t code = 0; int32_t code = 0;
STsdb *pTsdb = NULL; SDataFReader *pReader;
char fname[TSDB_FILENAME_LEN];
if (*ppWriter == NULL) goto _exit; // alloc
pReader = (SDataFReader *)taosMemoryCalloc(1, sizeof(*pReader));
if (pReader == NULL) {
code = TSDB_CODE_OUT_OF_MEMORY;
goto _err;
}
pReader->pTsdb = pTsdb;
pReader->pSet = pSet;
pTsdb = (*ppWriter)->pTsdb; // open impl
if (sync) { // head
if (taosFsyncFile((*ppWriter)->pHeadFD) < 0) { tsdbHeadFileName(pTsdb, pSet->diskId, pSet->fid, pSet->pHeadF, fname);
pReader->pHeadFD = taosOpenFile(fname, TD_FILE_READ);
if (pReader->pHeadFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
if (taosFsyncFile((*ppWriter)->pDataFD) < 0) { // data
tsdbDataFileName(pTsdb, pSet->diskId, pSet->fid, pSet->pDataF, fname);
pReader->pDataFD = taosOpenFile(fname, TD_FILE_READ);
if (pReader->pDataFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
if (taosFsyncFile((*ppWriter)->pSmaFD) < 0) { // sma
tsdbSmaFileName(pTsdb, pSet->diskId, pSet->fid, pSet->pSmaF, fname);
pReader->pSmaFD = taosOpenFile(fname, TD_FILE_READ);
if (pReader->pSmaFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
if (taosFsyncFile((*ppWriter)->pLastFD) < 0) { // sst
for (int32_t iSst = 0; iSst < pSet->nSstF; iSst++) {
tsdbSstFileName(pTsdb, pSet->diskId, pSet->fid, pSet->aSstF[iSst], fname);
pReader->aLastFD[iSst] = taosOpenFile(fname, TD_FILE_READ);
if (pReader->aLastFD[iSst] == NULL) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
} }
if (taosCloseFile(&(*ppWriter)->pHeadFD) < 0) { *ppReader = pReader;
return code;
_err:
tsdbError("vgId:%d, tsdb data file reader open failed since %s", TD_VID(pTsdb->pVnode), tstrerror(code));
*ppReader = NULL;
return code;
}
int32_t tsdbDataFReaderClose(SDataFReader **ppReader) {
int32_t code = 0;
if (*ppReader == NULL) goto _exit;
// head
if (taosCloseFile(&(*ppReader)->pHeadFD) < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
if (taosCloseFile(&(*ppWriter)->pDataFD) < 0) { // data
if (taosCloseFile(&(*ppReader)->pDataFD) < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
if (taosCloseFile(&(*ppWriter)->pSmaFD) < 0) { // sma
if (taosCloseFile(&(*ppReader)->pSmaFD) < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
if (taosCloseFile(&(*ppWriter)->pLastFD) < 0) { // sst
for (int32_t iSst = 0; iSst < (*ppReader)->pSet->nSstF; iSst++) {
if (taosCloseFile(&(*ppReader)->aLastFD[iSst]) < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
}
for (int32_t iBuf = 0; iBuf < sizeof((*ppWriter)->aBuf) / sizeof(uint8_t *); iBuf++) { for (int32_t iBuf = 0; iBuf < sizeof((*ppReader)->aBuf) / sizeof(uint8_t *); iBuf++) {
tFree((*ppWriter)->aBuf[iBuf]); tFree((*ppReader)->aBuf[iBuf]);
} }
taosMemoryFree(*ppWriter); taosMemoryFree(*ppReader);
_exit: _exit:
*ppWriter = NULL; *ppReader = NULL;
return code; return code;
_err: _err:
tsdbError("vgId:%d, data file writer close failed since %s", TD_VID(pTsdb->pVnode), tstrerror(code)); tsdbError("vgId:%d, data file reader close failed since %s", TD_VID((*ppReader)->pTsdb->pVnode), tstrerror(code));
return code; return code;
} }
int32_t tsdbUpdateDFileSetHeader(SDataFWriter *pWriter) { int32_t tsdbReadBlockIdx(SDataFReader *pReader, SArray *aBlockIdx) {
int32_t code = 0; int32_t code = 0;
int64_t offset = pReader->pSet->pHeadF->offset;
int64_t size = pReader->pSet->pHeadF->size - offset;
int64_t n; int64_t n;
char hdr[TSDB_FHDR_SIZE]; uint32_t delimiter;
// head ==============
memset(hdr, 0, TSDB_FHDR_SIZE);
tPutHeadFile(hdr, &pWriter->fHead);
taosCalcChecksumAppend(0, hdr, TSDB_FHDR_SIZE);
n = taosLSeekFile(pWriter->pHeadFD, 0, SEEK_SET);
if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
}
n = taosWriteFile(pWriter->pHeadFD, hdr, TSDB_FHDR_SIZE); taosArrayClear(aBlockIdx);
if (n < 0) { if (size == 0) {
code = TAOS_SYSTEM_ERROR(errno); goto _exit;
goto _err;
} }
// data ============== // alloc
memset(hdr, 0, TSDB_FHDR_SIZE); code = tRealloc(&pReader->aBuf[0], size);
tPutDataFile(hdr, &pWriter->fData); if (code) goto _err;
taosCalcChecksumAppend(0, hdr, TSDB_FHDR_SIZE);
n = taosLSeekFile(pWriter->pDataFD, 0, SEEK_SET); // seek
if (n < 0) { if (taosLSeekFile(pReader->pHeadFD, offset, SEEK_SET) < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
n = taosWriteFile(pWriter->pDataFD, hdr, TSDB_FHDR_SIZE); // read
n = taosReadFile(pReader->pHeadFD, pReader->aBuf[0], size);
if (n < 0) { if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} } else if (n < size) {
code = TSDB_CODE_FILE_CORRUPTED;
// sma ==============
memset(hdr, 0, TSDB_FHDR_SIZE);
tPutSmaFile(hdr, &pWriter->fSma);
taosCalcChecksumAppend(0, hdr, TSDB_FHDR_SIZE);
n = taosLSeekFile(pWriter->pSmaFD, 0, SEEK_SET);
if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
n = taosWriteFile(pWriter->pSmaFD, hdr, TSDB_FHDR_SIZE); // check
if (n < 0) { if (!taosCheckChecksumWhole(pReader->aBuf[0], size)) {
code = TAOS_SYSTEM_ERROR(errno); code = TSDB_CODE_FILE_CORRUPTED;
goto _err; goto _err;
} }
// sst ============== // decode
memset(hdr, 0, TSDB_FHDR_SIZE); n = 0;
tPutSstFile(hdr, &pWriter->fSst[pWriter->wSet.nSstF - 1]); n = tGetU32(pReader->aBuf[0] + n, &delimiter);
taosCalcChecksumAppend(0, hdr, TSDB_FHDR_SIZE); ASSERT(delimiter == TSDB_FILE_DLMT);
n = taosLSeekFile(pWriter->pLastFD, 0, SEEK_SET); while (n < size - sizeof(TSCKSUM)) {
if (n < 0) { SBlockIdx blockIdx;
code = TAOS_SYSTEM_ERROR(errno); n += tGetBlockIdx(pReader->aBuf[0] + n, &blockIdx);
goto _err;
}
n = taosWriteFile(pWriter->pLastFD, hdr, TSDB_FHDR_SIZE); if (taosArrayPush(aBlockIdx, &blockIdx) == NULL) {
if (n < 0) { code = TSDB_CODE_OUT_OF_MEMORY;
code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} }
}
ASSERT(n + sizeof(TSCKSUM) == size);
_exit:
return code; return code;
_err: _err:
tsdbError("vgId:%d, update DFileSet header failed since %s", TD_VID(pWriter->pTsdb->pVnode), tstrerror(code)); tsdbError("vgId:%d, read block idx failed since %s", TD_VID(pReader->pTsdb->pVnode), tstrerror(code));
return code; return code;
} }
int32_t tsdbWriteBlockIdx(SDataFWriter *pWriter, SArray *aBlockIdx) { int32_t tsdbReadSstBlk(SDataFReader *pReader, int32_t iSst, SArray *aSstBlk) {
int32_t code = 0; int32_t code = 0;
SHeadFile *pHeadFile = &pWriter->fHead; int64_t offset = pReader->pSet->aSstF[iSst]->offset;
int64_t size = 0; int64_t size = pReader->pSet->aSstF[iSst]->size - offset;
int64_t n; int64_t n;
uint32_t delimiter;
// check taosArrayClear(aSstBlk);
if (taosArrayGetSize(aBlockIdx) == 0) { if (size == 0) {
pHeadFile->offset = pHeadFile->size;
goto _exit; goto _exit;
} }
// prepare
size = sizeof(uint32_t);
for (int32_t iBlockIdx = 0; iBlockIdx < taosArrayGetSize(aBlockIdx); iBlockIdx++) {
size += tPutBlockIdx(NULL, taosArrayGet(aBlockIdx, iBlockIdx));
}
size += sizeof(TSCKSUM);
// alloc // alloc
code = tRealloc(&pWriter->aBuf[0], size); code = tRealloc(&pReader->aBuf[0], size);
if (code) goto _err; if (code) goto _err;
// build // seek
n = 0; if (taosLSeekFile(pReader->aLastFD[iSst], offset, SEEK_SET) < 0) {
n = tPutU32(pWriter->aBuf[0] + n, TSDB_FILE_DLMT); code = TAOS_SYSTEM_ERROR(errno);
for (int32_t iBlockIdx = 0; iBlockIdx < taosArrayGetSize(aBlockIdx); iBlockIdx++) { goto _err;
n += tPutBlockIdx(pWriter->aBuf[0] + n, taosArrayGet(aBlockIdx, iBlockIdx));
} }
taosCalcChecksumAppend(0, pWriter->aBuf[0], size);
ASSERT(n + sizeof(TSCKSUM) == size);
// write // read
n = taosWriteFile(pWriter->pHeadFD, pWriter->aBuf[0], size); n = taosReadFile(pReader->aLastFD[iSst], pReader->aBuf[0], size);
if (n < 0) { if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} else if (n < size) {
code = TSDB_CODE_FILE_CORRUPTED;
goto _err;
} }
// update // check
pHeadFile->offset = pHeadFile->size; if (!taosCheckChecksumWhole(pReader->aBuf[0], size)) {
pHeadFile->size += size; code = TSDB_CODE_FILE_CORRUPTED;
goto _err;
}
// decode
n = 0;
n = tGetU32(pReader->aBuf[0] + n, &delimiter);
ASSERT(delimiter == TSDB_FILE_DLMT);
while (n < size - sizeof(TSCKSUM)) {
SSstBlk blockl;
n += tGetSstBlk(pReader->aBuf[0] + n, &blockl);
if (taosArrayPush(aSstBlk, &blockl) == NULL) {
code = TSDB_CODE_OUT_OF_MEMORY;
goto _err;
}
}
ASSERT(n + sizeof(TSCKSUM) == size);
_exit: _exit:
tsdbTrace("vgId:%d write block idx, offset:%" PRId64 " size:%" PRId64 " nBlockIdx:%d", TD_VID(pWriter->pTsdb->pVnode),
pHeadFile->offset, size, taosArrayGetSize(aBlockIdx));
return code; return code;
_err: _err:
tsdbError("vgId:%d, write block idx failed since %s", TD_VID(pWriter->pTsdb->pVnode), tstrerror(code)); tsdbError("vgId:%d read blockl failed since %s", TD_VID(pReader->pTsdb->pVnode), tstrerror(code));
return code; return code;
} }
int32_t tsdbWriteBlock(SDataFWriter *pWriter, SMapData *mBlock, SBlockIdx *pBlockIdx) { int32_t tsdbReadBlock(SDataFReader *pReader, SBlockIdx *pBlockIdx, SMapData *mBlock) {
int32_t code = 0; int32_t code = 0;
SHeadFile *pHeadFile = &pWriter->fHead; int64_t offset = pBlockIdx->offset;
int64_t size; int64_t size = pBlockIdx->size;
int64_t n; int64_t n;
int64_t tn;
ASSERT(mBlock->nItem > 0);
// alloc // alloc
size = sizeof(uint32_t) + tPutMapData(NULL, mBlock) + sizeof(TSCKSUM); code = tRealloc(&pReader->aBuf[0], size);
code = tRealloc(&pWriter->aBuf[0], size);
if (code) goto _err; if (code) goto _err;
// build // seek
n = 0; if (taosLSeekFile(pReader->pHeadFD, offset, SEEK_SET) < 0) {
n += tPutU32(pWriter->aBuf[0] + n, TSDB_FILE_DLMT); code = TAOS_SYSTEM_ERROR(errno);
n += tPutMapData(pWriter->aBuf[0] + n, mBlock); goto _err;
taosCalcChecksumAppend(0, pWriter->aBuf[0], size); }
ASSERT(n + sizeof(TSCKSUM) == size);
// write // read
n = taosWriteFile(pWriter->pHeadFD, pWriter->aBuf[0], size); n = taosReadFile(pReader->pHeadFD, pReader->aBuf[0], size);
if (n < 0) { if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} else if (n < size) {
code = TSDB_CODE_FILE_CORRUPTED;
goto _err;
} }
// update // check
pBlockIdx->offset = pHeadFile->size; if (!taosCheckChecksumWhole(pReader->aBuf[0], size)) {
pBlockIdx->size = size; code = TSDB_CODE_FILE_CORRUPTED;
pHeadFile->size += size; goto _err;
}
// decode
n = 0;
uint32_t delimiter;
n += tGetU32(pReader->aBuf[0] + n, &delimiter);
ASSERT(delimiter == TSDB_FILE_DLMT);
tn = tGetMapData(pReader->aBuf[0] + n, mBlock);
if (tn < 0) {
code = TSDB_CODE_OUT_OF_MEMORY;
goto _err;
}
n += tn;
ASSERT(n + sizeof(TSCKSUM) == size);
tsdbTrace("vgId:%d, write block, file ID:%d commit ID:%d suid:%" PRId64 " uid:%" PRId64 " offset:%" PRId64
" size:%" PRId64 " nItem:%d",
TD_VID(pWriter->pTsdb->pVnode), pWriter->wSet.fid, pHeadFile->commitID, pBlockIdx->suid, pBlockIdx->uid,
pBlockIdx->offset, pBlockIdx->size, mBlock->nItem);
return code; return code;
_err: _err:
tsdbError("vgId:%d, write block failed since %s", TD_VID(pWriter->pTsdb->pVnode), tstrerror(code)); tsdbError("vgId:%d, read block failed since %s", TD_VID(pReader->pTsdb->pVnode), tstrerror(code));
return code; return code;
} }
int32_t tsdbWriteSstBlk(SDataFWriter *pWriter, SArray *aSstBlk) { int32_t tsdbReadBlockSma(SDataFReader *pReader, SDataBlk *pDataBlk, SArray *aColumnDataAgg) {
int32_t code = 0; int32_t code = 0;
SSstFile *pSstFile = &pWriter->fSst[pWriter->wSet.nSstF - 1]; SSmaInfo *pSmaInfo = &pDataBlk->smaInfo;
int64_t size;
int64_t n;
// check ASSERT(pSmaInfo->size > 0);
if (taosArrayGetSize(aSstBlk) == 0) {
pSstFile->offset = pSstFile->size;
goto _exit;
}
// size taosArrayClear(aColumnDataAgg);
size = sizeof(uint32_t); // TSDB_FILE_DLMT
for (int32_t iBlockL = 0; iBlockL < taosArrayGetSize(aSstBlk); iBlockL++) {
size += tPutSstBlk(NULL, taosArrayGet(aSstBlk, iBlockL));
}
size += sizeof(TSCKSUM);
// alloc // alloc
code = tRealloc(&pWriter->aBuf[0], size); int32_t size = pSmaInfo->size + sizeof(TSCKSUM);
code = tRealloc(&pReader->aBuf[0], size);
if (code) goto _err; if (code) goto _err;
// encode // seek
n = 0; int64_t n = taosLSeekFile(pReader->pSmaFD, pSmaInfo->offset, SEEK_SET);
n += tPutU32(pWriter->aBuf[0] + n, TSDB_FILE_DLMT); if (n < 0) {
for (int32_t iBlockL = 0; iBlockL < taosArrayGetSize(aSstBlk); iBlockL++) { code = TAOS_SYSTEM_ERROR(errno);
n += tPutSstBlk(pWriter->aBuf[0] + n, taosArrayGet(aSstBlk, iBlockL)); goto _err;
} else if (n < pSmaInfo->offset) {
code = TSDB_CODE_FILE_CORRUPTED;
goto _err;
} }
taosCalcChecksumAppend(0, pWriter->aBuf[0], size);
ASSERT(n + sizeof(TSCKSUM) == size); // read
n = taosReadFile(pReader->pSmaFD, pReader->aBuf[0], size);
// write
n = taosWriteFile(pWriter->pLastFD, pWriter->aBuf[0], size);
if (n < 0) { if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno); code = TAOS_SYSTEM_ERROR(errno);
goto _err; goto _err;
} else if (n < size) {
code = TSDB_CODE_FILE_CORRUPTED;
goto _err;
} }
// update // check
pSstFile->offset = pSstFile->size; if (!taosCheckChecksumWhole(pReader->aBuf[0], size)) {
pSstFile->size += size; code = TSDB_CODE_FILE_CORRUPTED;
goto _err;
}
// decode
n = 0;
while (n < pSmaInfo->size) {
SColumnDataAgg sma;
n += tGetColumnDataAgg(pReader->aBuf[0] + n, &sma);
if (taosArrayPush(aColumnDataAgg, &sma) == NULL) {
code = TSDB_CODE_OUT_OF_MEMORY;
goto _err;
}
}
_exit:
tsdbTrace("vgId:%d tsdb write blockl, loffset:%" PRId64 " size:%" PRId64, TD_VID(pWriter->pTsdb->pVnode),
pSstFile->offset, size);
return code; return code;
_err: _err:
tsdbError("vgId:%d tsdb write blockl failed since %s", TD_VID(pWriter->pTsdb->pVnode), tstrerror(code)); tsdbError("vgId:%d tsdb read block sma failed since %s", TD_VID(pReader->pTsdb->pVnode), tstrerror(code));
return code; return code;
} }
static void tsdbUpdateBlockInfo(SBlockData *pBlockData, SDataBlk *pDataBlk) { static int32_t tsdbReadBlockDataImpl(SDataFReader *pReader, SBlockInfo *pBlkInfo, int8_t fromLast,
for (int32_t iRow = 0; iRow < pBlockData->nRow; iRow++) { SBlockData *pBlockData) {
TSDBKEY key = {.ts = pBlockData->aTSKEY[iRow], .version = pBlockData->aVersion[iRow]}; int32_t code = 0;
tBlockDataClear(pBlockData);
if (iRow == 0) { TdFilePtr pFD = fromLast ? pReader->aLastFD[0] : pReader->pDataFD; // (todo)
if (tsdbKeyCmprFn(&pDataBlk->minKey, &key) > 0) {
pDataBlk->minKey = key; // uid + version + tskey
} code = tsdbReadAndCheck(pFD, pBlkInfo->offset, &pReader->aBuf[0], pBlkInfo->szKey, 1);
if (code) goto _err;
SDiskDataHdr hdr;
uint8_t *p = pReader->aBuf[0] + tGetDiskDataHdr(pReader->aBuf[0], &hdr);
ASSERT(hdr.delimiter == TSDB_FILE_DLMT);
ASSERT(pBlockData->suid == hdr.suid);
ASSERT(pBlockData->uid == hdr.uid);
pBlockData->nRow = hdr.nRow;
// uid
if (hdr.uid == 0) {
ASSERT(hdr.szUid);
code = tsdbDecmprData(p, hdr.szUid, TSDB_DATA_TYPE_BIGINT, hdr.cmprAlg, (uint8_t **)&pBlockData->aUid,
sizeof(int64_t) * hdr.nRow, &pReader->aBuf[1]);
if (code) goto _err;
} else { } else {
if (pBlockData->aTSKEY[iRow] == pBlockData->aTSKEY[iRow - 1]) { ASSERT(!hdr.szUid);
pDataBlk->hasDup = 1;
}
} }
p += hdr.szUid;
if (iRow == pBlockData->nRow - 1 && tsdbKeyCmprFn(&pDataBlk->maxKey, &key) < 0) { // version
pDataBlk->maxKey = key; code = tsdbDecmprData(p, hdr.szVer, TSDB_DATA_TYPE_BIGINT, hdr.cmprAlg, (uint8_t **)&pBlockData->aVersion,
} sizeof(int64_t) * hdr.nRow, &pReader->aBuf[1]);
if (code) goto _err;
p += hdr.szVer;
// TSKEY
code = tsdbDecmprData(p, hdr.szKey, TSDB_DATA_TYPE_TIMESTAMP, hdr.cmprAlg, (uint8_t **)&pBlockData->aTSKEY,
sizeof(TSKEY) * hdr.nRow, &pReader->aBuf[1]);
if (code) goto _err;
p += hdr.szKey;
ASSERT(p - pReader->aBuf[0] == pBlkInfo->szKey - sizeof(TSCKSUM));
// read and decode columns
if (taosArrayGetSize(pBlockData->aIdx) == 0) goto _exit;
pDataBlk->minVer = TMIN(pDataBlk->minVer, key.version); if (hdr.szBlkCol > 0) {
pDataBlk->maxVer = TMAX(pDataBlk->maxVer, key.version); int64_t offset = pBlkInfo->offset + pBlkInfo->szKey;
code = tsdbReadAndCheck(pFD, offset, &pReader->aBuf[0], hdr.szBlkCol + sizeof(TSCKSUM), 1);
if (code) goto _err;
} }
pDataBlk->nRow += pBlockData->nRow;
}
static int32_t tsdbWriteBlockSma(SDataFWriter *pWriter, SBlockData *pBlockData, SSmaInfo *pSmaInfo) {
int32_t code = 0;
pSmaInfo->offset = 0; SBlockCol blockCol = {.cid = 0};
pSmaInfo->size = 0; SBlockCol *pBlockCol = &blockCol;
int32_t n = 0;
// encode
for (int32_t iColData = 0; iColData < taosArrayGetSize(pBlockData->aIdx); iColData++) { for (int32_t iColData = 0; iColData < taosArrayGetSize(pBlockData->aIdx); iColData++) {
SColData *pColData = tBlockDataGetColDataByIdx(pBlockData, iColData); SColData *pColData = tBlockDataGetColDataByIdx(pBlockData, iColData);
if ((!pColData->smaOn) || IS_VAR_DATA_TYPE(pColData->type)) continue; while (pBlockCol && pBlockCol->cid < pColData->cid) {
if (n < hdr.szBlkCol) {
SColumnDataAgg sma; n += tGetBlockCol(pReader->aBuf[0] + n, pBlockCol);
tsdbCalcColDataSMA(pColData, &sma); } else {
ASSERT(n == hdr.szBlkCol);
pBlockCol = NULL;
}
}
code = tRealloc(&pWriter->aBuf[0], pSmaInfo->size + tPutColumnDataAgg(NULL, &sma)); if (pBlockCol == NULL || pBlockCol->cid > pColData->cid) {
// add a lot of NONE
for (int32_t iRow = 0; iRow < hdr.nRow; iRow++) {
code = tColDataAppendValue(pColData, &COL_VAL_NONE(pColData->cid, pColData->type));
if (code) goto _err; if (code) goto _err;
pSmaInfo->size += tPutColumnDataAgg(pWriter->aBuf[0] + pSmaInfo->size, &sma);
} }
} else {
ASSERT(pBlockCol->type == pColData->type);
ASSERT(pBlockCol->flag && pBlockCol->flag != HAS_NONE);
// write if (pBlockCol->flag == HAS_NULL) {
if (pSmaInfo->size) { // add a lot of NULL
int32_t size = pSmaInfo->size + sizeof(TSCKSUM); for (int32_t iRow = 0; iRow < hdr.nRow; iRow++) {
code = tColDataAppendValue(pColData, &COL_VAL_NULL(pBlockCol->cid, pBlockCol->type));
code = tRealloc(&pWriter->aBuf[0], size);
if (code) goto _err; if (code) goto _err;
}
} else {
// decode from binary
int64_t offset = pBlkInfo->offset + pBlkInfo->szKey + hdr.szBlkCol + sizeof(TSCKSUM) + pBlockCol->offset;
int32_t size = pBlockCol->szBitmap + pBlockCol->szOffset + pBlockCol->szValue + sizeof(TSCKSUM);
taosCalcChecksumAppend(0, pWriter->aBuf[0], size); code = tsdbReadAndCheck(pFD, offset, &pReader->aBuf[1], size, 0);
if (code) goto _err;
int64_t n = taosWriteFile(pWriter->pSmaFD, pWriter->aBuf[0], size); code = tsdbDecmprColData(pReader->aBuf[1], pBlockCol, hdr.cmprAlg, hdr.nRow, pColData, &pReader->aBuf[2]);
if (n < 0) { if (code) goto _err;
code = TAOS_SYSTEM_ERROR(errno); }
goto _err;
} }
pSmaInfo->offset = pWriter->fSma.size;
pWriter->fSma.size += size;
} }
_exit:
return code; return code;
_err: _err:
tsdbError("vgId:%d tsdb write block sma failed since %s", TD_VID(pWriter->pTsdb->pVnode), tstrerror(code)); tsdbError("vgId:%d tsdb read block data impl failed since %s", TD_VID(pReader->pTsdb->pVnode), tstrerror(code));
return code; return code;
} }
int32_t tsdbWriteBlockData(SDataFWriter *pWriter, SBlockData *pBlockData, SBlockInfo *pBlkInfo, SSmaInfo *pSmaInfo, int32_t tsdbReadDataBlock(SDataFReader *pReader, SDataBlk *pDataBlk, SBlockData *pBlockData) {
int8_t cmprAlg, int8_t toLast) {
int32_t code = 0; int32_t code = 0;
ASSERT(pBlockData->nRow > 0); code = tsdbReadBlockDataImpl(pReader, &pDataBlk->aSubBlock[0], 0, pBlockData);
pBlkInfo->offset = toLast ? pWriter->fSst[pWriter->wSet.nSstF - 1].size : pWriter->fData.size;
pBlkInfo->szBlock = 0;
pBlkInfo->szKey = 0;
int32_t aBufN[4] = {0};
code = tCmprBlockData(pBlockData, cmprAlg, NULL, NULL, pWriter->aBuf, aBufN);
if (code) goto _err; if (code) goto _err;
// write ================= if (pDataBlk->nSubBlock > 1) {
TdFilePtr pFD = toLast ? pWriter->pLastFD : pWriter->pDataFD; SBlockData bData1;
SBlockData bData2;
pBlkInfo->szKey = aBufN[3] + aBufN[2]; // create
pBlkInfo->szBlock = aBufN[0] + aBufN[1] + aBufN[2] + aBufN[3]; code = tBlockDataCreate(&bData1);
if (code) goto _err;
code = tBlockDataCreate(&bData2);
if (code) goto _err;
int64_t n = taosWriteFile(pFD, pWriter->aBuf[3], aBufN[3]); // init
if (n < 0) { tBlockDataInitEx(&bData1, pBlockData);
code = TAOS_SYSTEM_ERROR(errno); tBlockDataInitEx(&bData2, pBlockData);
goto _err;
}
n = taosWriteFile(pFD, pWriter->aBuf[2], aBufN[2]); for (int32_t iSubBlock = 1; iSubBlock < pDataBlk->nSubBlock; iSubBlock++) {
if (n < 0) { code = tsdbReadBlockDataImpl(pReader, &pDataBlk->aSubBlock[iSubBlock], 0, &bData1);
code = TAOS_SYSTEM_ERROR(errno); if (code) {
tBlockDataDestroy(&bData1, 1);
tBlockDataDestroy(&bData2, 1);
goto _err; goto _err;
} }
if (aBufN[1]) { code = tBlockDataCopy(pBlockData, &bData2);
n = taosWriteFile(pFD, pWriter->aBuf[1], aBufN[1]); if (code) {
if (n < 0) { tBlockDataDestroy(&bData1, 1);
code = TAOS_SYSTEM_ERROR(errno); tBlockDataDestroy(&bData2, 1);
goto _err; goto _err;
} }
}
if (aBufN[0]) { code = tBlockDataMerge(&bData1, &bData2, pBlockData);
n = taosWriteFile(pFD, pWriter->aBuf[0], aBufN[0]); if (code) {
if (n < 0) { tBlockDataDestroy(&bData1, 1);
code = TAOS_SYSTEM_ERROR(errno); tBlockDataDestroy(&bData2, 1);
goto _err; goto _err;
} }
} }
// update info tBlockDataDestroy(&bData1, 1);
if (toLast) { tBlockDataDestroy(&bData2, 1);
pWriter->fSst[pWriter->wSet.nSstF - 1].size += pBlkInfo->szBlock;
} else {
pWriter->fData.size += pBlkInfo->szBlock;
}
// ================= SMA ====================
if (pSmaInfo) {
code = tsdbWriteBlockSma(pWriter, pBlockData, pSmaInfo);
if (code) goto _err;
} }
_exit:
tsdbTrace("vgId:%d tsdb write block data, suid:%" PRId64 " uid:%" PRId64 " nRow:%d, offset:%" PRId64 " size:%d",
TD_VID(pWriter->pTsdb->pVnode), pBlockData->suid, pBlockData->uid, pBlockData->nRow, pBlkInfo->offset,
pBlkInfo->szBlock);
return code; return code;
_err: _err:
tsdbError("vgId:%d tsdb write block data failed since %s", TD_VID(pWriter->pTsdb->pVnode), tstrerror(code)); tsdbError("vgId:%d tsdb read data block failed since %s", TD_VID(pReader->pTsdb->pVnode), tstrerror(code));
return code; return code;
} }
int32_t tsdbDFileSetCopy(STsdb *pTsdb, SDFileSet *pSetFrom, SDFileSet *pSetTo) { int32_t tsdbReadSstBlock(SDataFReader *pReader, int32_t iSst, SSstBlk *pSstBlk, SBlockData *pBlockData) {
int32_t code = 0; int32_t code = 0;
int64_t n;
int64_t size;
TdFilePtr pOutFD = NULL; // TODO
TdFilePtr PInFD = NULL; // TODO
char fNameFrom[TSDB_FILENAME_LEN];
char fNameTo[TSDB_FILENAME_LEN];
// head
tsdbHeadFileName(pTsdb, pSetFrom->diskId, pSetFrom->fid, pSetFrom->pHeadF, fNameFrom);
tsdbHeadFileName(pTsdb, pSetTo->diskId, pSetTo->fid, pSetTo->pHeadF, fNameTo);
pOutFD = taosOpenFile(fNameTo, TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC);
if (pOutFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
}
PInFD = taosOpenFile(fNameFrom, TD_FILE_READ);
if (PInFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
}
n = taosFSendFile(pOutFD, PInFD, 0, pSetFrom->pHeadF->size);
if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
}
taosCloseFile(&pOutFD);
taosCloseFile(&PInFD);
// data
tsdbDataFileName(pTsdb, pSetFrom->diskId, pSetFrom->fid, pSetFrom->pDataF, fNameFrom);
tsdbDataFileName(pTsdb, pSetTo->diskId, pSetTo->fid, pSetTo->pDataF, fNameTo);
pOutFD = taosOpenFile(fNameTo, TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC);
if (pOutFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
}
PInFD = taosOpenFile(fNameFrom, TD_FILE_READ);
if (PInFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
}
n = taosFSendFile(pOutFD, PInFD, 0, pSetFrom->pDataF->size);
if (n < 0) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
}
taosCloseFile(&pOutFD);
taosCloseFile(&PInFD);
// sst
tsdbSstFileName(pTsdb, pSetFrom->diskId, pSetFrom->fid, pSetFrom->aSstF[0], fNameFrom);
tsdbSstFileName(pTsdb, pSetTo->diskId, pSetTo->fid, pSetTo->aSstF[0], fNameTo);
pOutFD = taosOpenFile(fNameTo, TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC);
if (pOutFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
}
PInFD = taosOpenFile(fNameFrom, TD_FILE_READ);
if (PInFD == NULL) {
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
}
n = taosFSendFile(pOutFD, PInFD, 0, pSetFrom->aSstF[0]->size); code = tsdbReadBlockDataImpl(pReader, &pSstBlk->bInfo, 1, pBlockData);
if (n < 0) { if (code) goto _err;
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
}
taosCloseFile(&pOutFD);
taosCloseFile(&PInFD);
// sma return code;
tsdbSmaFileName(pTsdb, pSetFrom->diskId, pSetFrom->fid, pSetFrom->pSmaF, fNameFrom);
tsdbSmaFileName(pTsdb, pSetTo->diskId, pSetTo->fid, pSetTo->pSmaF, fNameTo);
pOutFD = taosOpenFile(fNameTo, TD_FILE_WRITE | TD_FILE_CREATE | TD_FILE_TRUNC); _err:
if (pOutFD == NULL) { tsdbError("vgId:%d tsdb read last block failed since %s", TD_VID(pReader->pTsdb->pVnode), tstrerror(code));
code = TAOS_SYSTEM_ERROR(errno); return code;
goto _err; }
}
PInFD = taosOpenFile(fNameFrom, TD_FILE_READ); int32_t tsdbReadSstBlockEx(SDataFReader *pReader, int32_t iSst, SSstBlk *pSstBlk, SBlockData *pBlockData) {
if (PInFD == NULL) { int32_t code = 0;
code = TAOS_SYSTEM_ERROR(errno);
goto _err;
}
n = taosFSendFile(pOutFD, PInFD, 0, pSetFrom->pSmaF->size); // read
if (n < 0) { code = tsdbReadAndCheck(pReader->aLastFD[iSst], pSstBlk->bInfo.offset, &pReader->aBuf[0], pSstBlk->bInfo.szBlock, 0);
code = TAOS_SYSTEM_ERROR(errno); if (code) goto _exit;
goto _err;
}
taosCloseFile(&pOutFD);
taosCloseFile(&PInFD);
return code; // decmpr
code = tDecmprBlockData(pReader->aBuf[0], pSstBlk->bInfo.szBlock, pBlockData, &pReader->aBuf[1]);
if (code) goto _exit;
_err: _exit:
tsdbError("vgId:%d, tsdb DFileSet copy failed since %s", TD_VID(pTsdb->pVnode), tstrerror(code));
return code; return code;
} }
...@@ -1558,7 +1534,6 @@ _err: ...@@ -1558,7 +1534,6 @@ _err:
tsdbError("vgId:%d, update del file hdr failed since %s", TD_VID(pWriter->pTsdb->pVnode), tstrerror(code)); tsdbError("vgId:%d, update del file hdr failed since %s", TD_VID(pWriter->pTsdb->pVnode), tstrerror(code));
return code; return code;
} }
// SDelFReader ==================================================== // SDelFReader ====================================================
struct SDelFReader { struct SDelFReader {
STsdb *pTsdb; STsdb *pTsdb;
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册