xxq250
/
homework-jianmu
1
0
Fork 0
Code Issues Pull Requests 1 Actions Packages Projects Releases Wiki Activity

refactor extbuffer model.

This commit is contained in:
hjxilinx 2020-02-05 17:31:58 +08:00
parent d55dadd07f
commit 1de610a48d
21 changed files with 1620 additions and 1497 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
src/client/inc/tscSecondaryMerge.h
View File

@ -68,7 +68,7 @@ typedef struct SLocalReducer {
bool hasPrevRow; // cannot be released
bool hasUnprocessedRow;
tOrderDescriptor * pDesc;
tColModel * resColModel;
SColumnModel * resColModel;
tExtMemBuffer ** pExtMemBuffer; // disk-based buffer
SInterpolationInfo interpolationInfo; // interpolation support structure
char * pFinalRes; // result data after interpo
@ -92,7 +92,7 @@ typedef struct SSubqueryState {
typedef struct SRetrieveSupport {
tExtMemBuffer ** pExtMemBuffer; // for build loser tree
tOrderDescriptor *pOrderDescriptor;
tColModel * pFinalColModel; // colModel for final result
SColumnModel * pFinalColModel; // colModel for final result
SSubqueryState * pState;
int32_t subqueryIndex; // index of current vnode in vnode list
SSqlObj * pParentSqlObj;
@ -102,9 +102,9 @@ typedef struct SRetrieveSupport {
} SRetrieveSupport;
int32_t tscLocalReducerEnvCreate(SSqlObj *pSql, tExtMemBuffer ***pMemBuffer, tOrderDescriptor **pDesc,
tColModel **pFinalModel, uint32_t nBufferSize);
SColumnModel **pFinalModel, uint32_t nBufferSize);
void tscLocalReducerEnvDestroy(tExtMemBuffer **pMemBuffer, tOrderDescriptor *pDesc, tColModel *pFinalModel,
void tscLocalReducerEnvDestroy(tExtMemBuffer **pMemBuffer, tOrderDescriptor *pDesc, SColumnModel *pFinalModel,
int32_t numOfVnodes);
int32_t saveToBuffer(tExtMemBuffer *pMemoryBuf, tOrderDescriptor *pDesc, tFilePage *pPage, void *data,
@ -116,7 +116,7 @@ int32_t tscFlushTmpBuffer(tExtMemBuffer *pMemoryBuf, tOrderDescriptor *pDesc, tF
* create local reducer to launch the second-stage reduce process at client site
*/
void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrderDescriptor *pDesc,
tColModel *finalModel, SSqlCmd *pSqlCmd, SSqlRes *pRes);
SColumnModel *finalModel, SSqlCmd *pSqlCmd, SSqlRes *pRes);
void tscDestroyLocalReducer(SSqlObj *pSql);

3
src/client/src/tscFunctionImpl.c
View File

@ -27,6 +27,7 @@
#include "ttime.h"
#include "ttypes.h"
#include "tutil.h"
#include "tpercentile.h"
#define GET_INPUT_CHAR(x) (((char *)((x)->aInputElemBuf)) + ((x)->startOffset) * ((x)->inputBytes))
#define GET_INPUT_CHAR_INDEX(x, y) (GET_INPUT_CHAR(x) + (y) * (x)->inputBytes)
@ -2416,7 +2417,7 @@ static bool percentile_function_setup(SQLFunctionCtx *pCtx) {
SResultInfo *pResInfo = GET_RES_INFO(pCtx);
SSchema field[1] = {{pCtx->inputType, "dummyCol", 0, pCtx->inputBytes}};
tColModel *pModel = tColModelCreate(field, 1, 1000);
SColumnModel *pModel = createColumnModel(field, 1, 1000);
int32_t orderIdx = 0;
// tOrderDesc object

2
src/client/src/tscParseInsert.c
View File

@ -321,7 +321,7 @@ int32_t tsParseOneColumnData(SSchema *pSchema, SSQLToken *pToken, char *payload,
if (pToken->type == TK_NULL) {
*(uint32_t *)payload = TSDB_DATA_NCHAR_NULL;
} else {
// if the converted output len is over than pSchema->bytes, return error: 'Argument list too long'
// if the converted output len is over than pColumnModel->bytes, return error: 'Argument list too long'
if (!taosMbsToUcs4(pToken->z, pToken->n, payload, pSchema->bytes)) {
char buf[512] = {0};
snprintf(buf, 512, "%s", strerror(errno));

26
src/client/src/tscSQLParser.c
View File

@ -5453,8 +5453,6 @@ int32_t doCheckForQuery(SSqlObj* pSql, SQuerySQL* pQuerySql, int32_t index) {
const char* msg0 = "invalid table name";
const char* msg1 = "table name too long";
const char* msg2 = "point interpolation query needs timestamp";
const char* msg3 = "sliding value too small";
const char* msg4 = "sliding value no larger than the interval value";
const char* msg5 = "fill only available for interval query";
const char* msg6 = "start(end) time of query range required or time range too large";
const char* msg7 = "illegal number of tables in from clause";
@ -5587,30 +5585,6 @@ int32_t doCheckForQuery(SSqlObj* pSql, SQuerySQL* pQuerySql, int32_t index) {
if (!hasTimestampForPointInterpQuery(pQueryInfo)) {
return invalidSqlErrMsg(pQueryInfo->msg, msg2);
}
// // set sliding value, the query time range needs to be decide in the first place
// SSQLToken* pSliding = &pQuerySql->sliding;
// if (pSliding->n != 0) {
// if (!tscEmbedded && pCmd->inStream == 0 && hasDefaultQueryTimeRange(pQueryInfo)) { // sliding only allowed in stream
// const char* msg = "time range expected for sliding window query";
// return invalidSqlErrMsg(tscGetErrorMsgPayload(pCmd), msg);
// }
//
// getTimestampInUsFromStr(pSliding->z, pSliding->n, &pQueryInfo->nSlidingTime);
// if (pMeterMetaInfo->pMeterMeta->precision == TSDB_TIME_PRECISION_MILLI) {
// pQueryInfo->nSlidingTime /= 1000;
// }
//
// if (pQueryInfo->nSlidingTime < tsMinSlidingTime) {
// return invalidSqlErrMsg(pQueryInfo->msg, msg3);
// }
//
// if (pQueryInfo->nSlidingTime > pQueryInfo->nAggTimeInterval) {
// return invalidSqlErrMsg(pQueryInfo->msg, msg4);
// }
// } else {
// pQueryInfo->nSlidingTime = -1;
// }
// in case of join query, time range is required.
if (QUERY_IS_JOIN_QUERY(pQueryInfo->type)) {

150
src/client/src/tscSecondaryMerge.c
View File

@ -62,16 +62,19 @@ static void tscInitSqlContext(SSqlCmd *pCmd, SSqlRes *pRes, SLocalReducer *pRedu
for (int32_t i = 0; i < pQueryInfo->fieldsInfo.numOfOutputCols; ++i) {
SQLFunctionCtx *pCtx = &pReducer->pCtx[i];
pCtx->aOutputBuf = pReducer->pResultBuf->data + tscFieldInfoGetOffset(pQueryInfo, i) * pReducer->resColModel->maxCapacity;
pCtx->aOutputBuf = pReducer->pResultBuf->data + tscFieldInfoGetOffset(pQueryInfo, i) * pReducer->resColModel->capacity;
pCtx->order = pQueryInfo->order.order;
pCtx->functionId = pQueryInfo->exprsInfo.pExprs[i].functionId;
// input buffer hold only one point data
pCtx->aInputElemBuf = pReducer->pTempBuffer->data + pDesc->pSchema->colOffset[i];
int16_t offset = getColumnModelOffset(pDesc->pColumnModel, i);
SSchema* pSchema = getColumnModelSchema(pDesc->pColumnModel, i);
pCtx->aInputElemBuf = pReducer->pTempBuffer->data + offset;
// input data format comes from pModel
pCtx->inputType = pDesc->pSchema->pFields[i].type;
pCtx->inputBytes = pDesc->pSchema->pFields[i].bytes;
pCtx->inputType = pSchema->type;
pCtx->inputBytes = pSchema->bytes;
TAOS_FIELD *pField = tscFieldInfoGetField(pQueryInfo, i);
// output data format yet comes from pCmd.
@ -132,11 +135,11 @@ static void tscInitSqlContext(SSqlCmd *pCmd, SSqlRes *pRes, SLocalReducer *pRedu
* todo release allocated memory process with async process
*/
void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrderDescriptor *pDesc,
tColModel *finalmodel, SSqlCmd *pCmd, SSqlRes *pRes) {
SColumnModel *finalmodel, SSqlCmd *pCmd, SSqlRes *pRes) {
// offset of cmd in SSqlObj structure
char *pSqlObjAddr = (char *)pCmd - offsetof(SSqlObj, cmd);
if (pMemBuffer == NULL || pDesc->pSchema == NULL) {
if (pMemBuffer == NULL || pDesc->pColumnModel == NULL) {
tscLocalReducerEnvDestroy(pMemBuffer, pDesc, finalmodel, numOfBuffer);
tscError("%p no local buffer or intermediate result format model", pSqlObjAddr);
@ -162,9 +165,9 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
return;
}
if (pDesc->pSchema->maxCapacity >= pMemBuffer[0]->nPageSize) {
tscError("%p Invalid value of buffer capacity %d and page size %d ", pSqlObjAddr, pDesc->pSchema->maxCapacity,
pMemBuffer[0]->nPageSize);
if (pDesc->pColumnModel->capacity >= pMemBuffer[0]->pageSize) {
tscError("%p Invalid value of buffer capacity %d and page size %d ", pSqlObjAddr, pDesc->pColumnModel->capacity,
pMemBuffer[0]->pageSize);
tscLocalReducerEnvDestroy(pMemBuffer, pDesc, finalmodel, numOfBuffer);
pRes->code = TSDB_CODE_APP_ERROR;
@ -196,7 +199,7 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
int32_t numOfFlushoutInFile = pMemBuffer[i]->fileMeta.flushoutData.nLength;
for (int32_t j = 0; j < numOfFlushoutInFile; ++j) {
SLocalDataSource *pDS = (SLocalDataSource *)malloc(sizeof(SLocalDataSource) + pMemBuffer[0]->nPageSize);
SLocalDataSource *pDS = (SLocalDataSource *)malloc(sizeof(SLocalDataSource) + pMemBuffer[0]->pageSize);
if (pDS == NULL) {
tscError("%p failed to create merge structure", pSqlObjAddr);
pRes->code = TSDB_CODE_CLI_OUT_OF_MEMORY;
@ -219,7 +222,7 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
tscGetSrcColumnInfo(colInfo, pQueryInfo);
tColModelDisplayEx(pDesc->pSchema, pDS->filePage.data, pDS->filePage.numOfElems, pMemBuffer[0]->numOfElemsPerPage,
tColModelDisplayEx(pDesc->pColumnModel, pDS->filePage.data, pDS->filePage.numOfElems, pMemBuffer[0]->numOfElemsPerPage,
colInfo);
#endif
if (pDS->filePage.numOfElems == 0) { // no data in this flush
@ -259,7 +262,7 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
tscRestoreSQLFunctionForMetricQuery(pQueryInfo);
tscFieldInfoCalOffset(pQueryInfo);
if (pReducer->rowSize > pMemBuffer[0]->nPageSize) {
if (pReducer->rowSize > pMemBuffer[0]->pageSize) {
assert(false); // todo fixed row size is larger than the minimum page size;
}
@ -274,15 +277,15 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
pReducer->discardData = (tFilePage *)calloc(1, pReducer->rowSize + sizeof(tFilePage));
pReducer->discard = false;
pReducer->nResultBufSize = pMemBuffer[0]->nPageSize * 16;
pReducer->nResultBufSize = pMemBuffer[0]->pageSize * 16;
pReducer->pResultBuf = (tFilePage *)calloc(1, pReducer->nResultBufSize + sizeof(tFilePage));
int32_t finalRowLength = tscGetResRowLength(pQueryInfo);
pReducer->resColModel = finalmodel;
pReducer->resColModel->maxCapacity = pReducer->nResultBufSize / finalRowLength;
pReducer->resColModel->capacity = pReducer->nResultBufSize / finalRowLength;
assert(finalRowLength <= pReducer->rowSize);
pReducer->pFinalRes = calloc(1, pReducer->rowSize * pReducer->resColModel->maxCapacity);
pReducer->pFinalRes = calloc(1, pReducer->rowSize * pReducer->resColModel->capacity);
pReducer->pBufForInterpo = calloc(1, pReducer->nResultBufSize);
if (pReducer->pTempBuffer == NULL|| pReducer->discardData == NULL || pReducer->pResultBuf == NULL ||
@ -304,8 +307,8 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
tscCreateResPointerInfo(pRes, pQueryInfo);
tscInitSqlContext(pCmd, pRes, pReducer, pDesc);
// we change the maxCapacity of schema to denote that there is only one row in temp buffer
pReducer->pDesc->pSchema->maxCapacity = 1;
// we change the capacity of schema to denote that there is only one row in temp buffer
pReducer->pDesc->pColumnModel->capacity = 1;
//restore the limitation value at the last stage
if (tscOrderedProjectionQueryOnSTable(pQueryInfo, 0)) {
@ -333,7 +336,8 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
if (pQueryInfo->groupbyExpr.numOfGroupCols > 0) {
pInterpoInfo->pTags[0] = (char *)pInterpoInfo->pTags + POINTER_BYTES * pQueryInfo->groupbyExpr.numOfGroupCols;
for (int32_t i = 1; i < pQueryInfo->groupbyExpr.numOfGroupCols; ++i) {
pInterpoInfo->pTags[i] = pReducer->resColModel->pFields[startIndex + i - 1].bytes + pInterpoInfo->pTags[i - 1];
SSchema* pSchema = getColumnModelSchema(pReducer->resColModel, startIndex + i - 1);
pInterpoInfo->pTags[i] = pSchema->bytes + pInterpoInfo->pTags[i - 1];
}
} else {
assert(pInterpoInfo->pTags == NULL);
@ -346,16 +350,16 @@ static int32_t tscFlushTmpBufferImpl(tExtMemBuffer *pMemoryBuf, tOrderDescriptor
return 0;
}
assert(pPage->numOfElems <= pDesc->pSchema->maxCapacity);
assert(pPage->numOfElems <= pDesc->pColumnModel->capacity);
// sort before flush to disk, the data must be consecutively put on tFilePage.
if (pDesc->orderIdx.numOfOrderedCols > 0) {
if (pDesc->orderIdx.numOfCols > 0) {
tColDataQSort(pDesc, pPage->numOfElems, 0, pPage->numOfElems - 1, pPage->data, orderType);
}
#ifdef _DEBUG_VIEW
printf("%" PRIu64 " rows data flushed to disk after been sorted:\n", pPage->numOfElems);
tColModelDisplay(pDesc->pSchema, pPage->data, pPage->numOfElems, pPage->numOfElems);
tColModelDisplay(pDesc->pColumnModel, pPage->data, pPage->numOfElems, pPage->numOfElems);
#endif
// write to cache after being sorted
@ -383,18 +387,18 @@ int32_t tscFlushTmpBuffer(tExtMemBuffer *pMemoryBuf, tOrderDescriptor *pDesc, tF
int32_t saveToBuffer(tExtMemBuffer *pMemoryBuf, tOrderDescriptor *pDesc, tFilePage *pPage, void *data,
int32_t numOfRows, int32_t orderType) {
if (pPage->numOfElems + numOfRows <= pDesc->pSchema->maxCapacity) {
tColModelAppend(pDesc->pSchema, pPage, data, 0, numOfRows, numOfRows);
if (pPage->numOfElems + numOfRows <= pDesc->pColumnModel->capacity) {
tColModelAppend(pDesc->pColumnModel, pPage, data, 0, numOfRows, numOfRows);
return 0;
}
tColModel *pModel = pDesc->pSchema;
SColumnModel *pModel = pDesc->pColumnModel;
int32_t numOfRemainEntries = pDesc->pSchema->maxCapacity - pPage->numOfElems;
int32_t numOfRemainEntries = pDesc->pColumnModel->capacity - pPage->numOfElems;
tColModelAppend(pModel, pPage, data, 0, numOfRemainEntries, numOfRows);
/* current buffer is full, need to flushed to disk */
assert(pPage->numOfElems == pDesc->pSchema->maxCapacity);
assert(pPage->numOfElems == pDesc->pColumnModel->capacity);
int32_t ret = tscFlushTmpBuffer(pMemoryBuf, pDesc, pPage, orderType);
if (ret != 0) {
return -1;
@ -404,15 +408,15 @@ int32_t saveToBuffer(tExtMemBuffer *pMemoryBuf, tOrderDescriptor *pDesc, tFilePa
while (remain > 0) {
int32_t numOfWriteElems = 0;
if (remain > pModel->maxCapacity) {
numOfWriteElems = pModel->maxCapacity;
if (remain > pModel->capacity) {
numOfWriteElems = pModel->capacity;
} else {
numOfWriteElems = remain;
}
tColModelAppend(pModel, pPage, data, numOfRows - remain, numOfWriteElems, numOfRows);
if (pPage->numOfElems == pModel->maxCapacity) {
if (pPage->numOfElems == pModel->capacity) {
int32_t ret = tscFlushTmpBuffer(pMemoryBuf, pDesc, pPage, orderType);
if (ret != 0) {
return -1;
@ -508,7 +512,7 @@ void tscDestroyLocalReducer(SSqlObj *pSql) {
tscTrace("%p free local reducer finished", pSql);
}
static int32_t createOrderDescriptor(tOrderDescriptor **pOrderDesc, SSqlCmd *pCmd, tColModel *pModel) {
static int32_t createOrderDescriptor(tOrderDescriptor **pOrderDesc, SSqlCmd *pCmd, SColumnModel *pModel) {
int32_t numOfGroupByCols = 0;
SQueryInfo* pQueryInfo = tscGetQueryInfoDetail(pCmd, pCmd->clauseIndex);
@ -567,7 +571,7 @@ bool isSameGroup(SSqlCmd *pCmd, SLocalReducer *pReducer, char *pPrev, tFilePage
}
tOrderDescriptor *pOrderDesc = pReducer->pDesc;
int32_t numOfCols = pOrderDesc->orderIdx.numOfOrderedCols;
int32_t numOfCols = pOrderDesc->orderIdx.numOfCols;
// no group by columns, all data belongs to one group
if (numOfCols <= 0) {
@ -577,25 +581,25 @@ bool isSameGroup(SSqlCmd *pCmd, SLocalReducer *pReducer, char *pPrev, tFilePage
if (pOrderDesc->orderIdx.pData[numOfCols - 1] == PRIMARYKEY_TIMESTAMP_COL_INDEX) { //<= 0
// super table interval query
assert(pQueryInfo->nAggTimeInterval > 0);
pOrderDesc->orderIdx.numOfOrderedCols -= 1;
pOrderDesc->orderIdx.numOfCols -= 1;
} else { // simple group by query
assert(pQueryInfo->nAggTimeInterval == 0);
}
// only one row exists
int32_t ret = compare_a(pOrderDesc, 1, 0, pPrev, 1, 0, tmpBuffer->data);
pOrderDesc->orderIdx.numOfOrderedCols = numOfCols;
pOrderDesc->orderIdx.numOfCols = numOfCols;
return (ret == 0);
}
int32_t tscLocalReducerEnvCreate(SSqlObj *pSql, tExtMemBuffer ***pMemBuffer, tOrderDescriptor **pOrderDesc,
tColModel **pFinalModel, uint32_t nBufferSizes) {
SColumnModel **pFinalModel, uint32_t nBufferSizes) {
SSqlCmd *pCmd = &pSql->cmd;
SSqlRes *pRes = &pSql->res;
SSchema * pSchema = NULL;
tColModel *pModel = NULL;
SColumnModel *pModel = NULL;
*pFinalModel = NULL;
SQueryInfo* pQueryInfo = tscGetQueryInfoDetail(pCmd, pCmd->clauseIndex);
@ -630,14 +634,10 @@ int32_t tscLocalReducerEnvCreate(SSqlObj *pSql, tExtMemBuffer ***pMemBuffer, tOr
capacity = nBufferSizes / rlen;
}
pModel = tColModelCreate(pSchema, pQueryInfo->fieldsInfo.numOfOutputCols, capacity);
pModel = createColumnModel(pSchema, pQueryInfo->fieldsInfo.numOfOutputCols, capacity);
for (int32_t i = 0; i < pMeterMetaInfo->pMetricMeta->numOfVnodes; ++i) {
char tmpPath[512] = {0};
getTmpfilePath("tv_bf_db", tmpPath);
tscTrace("%p create [%d](%d) tmp file for subquery:%s", pSql, pMeterMetaInfo->pMetricMeta->numOfVnodes, i, tmpPath);
tExtMemBufferCreate(&(*pMemBuffer)[i], nBufferSizes, rlen, tmpPath, pModel);
(*pMemBuffer)[i] = createExtMemBuffer(nBufferSizes, rlen, pModel);
(*pMemBuffer)[i]->flushModel = MULTIPLE_APPEND_MODEL;
}
@ -655,7 +655,7 @@ int32_t tscLocalReducerEnvCreate(SSqlObj *pSql, tExtMemBuffer ***pMemBuffer, tOr
strcpy(pSchema[i].name, pField->name);
}
*pFinalModel = tColModelCreate(pSchema, pQueryInfo->fieldsInfo.numOfOutputCols, capacity);
*pFinalModel = createColumnModel(pSchema, pQueryInfo->fieldsInfo.numOfOutputCols, capacity);
tfree(pSchema);
return TSDB_CODE_SUCCESS;
@ -667,12 +667,12 @@ int32_t tscLocalReducerEnvCreate(SSqlObj *pSql, tExtMemBuffer ***pMemBuffer, tOr
* @param pFinalModel
* @param numOfVnodes
*/
void tscLocalReducerEnvDestroy(tExtMemBuffer **pMemBuffer, tOrderDescriptor *pDesc, tColModel *pFinalModel,
void tscLocalReducerEnvDestroy(tExtMemBuffer **pMemBuffer, tOrderDescriptor *pDesc, SColumnModel *pFinalModel,
int32_t numOfVnodes) {
tColModelDestroy(pFinalModel);
destroyColumnModel(pFinalModel);
tOrderDescDestroy(pDesc);
for (int32_t i = 0; i < numOfVnodes; ++i) {
tExtMemBufferDestroy(&pMemBuffer[i]);
pMemBuffer[i] = destoryExtMemBuffer(pMemBuffer[i]);
}
tfree(pMemBuffer);
@ -697,8 +697,8 @@ int32_t loadNewDataFromDiskFor(SLocalReducer *pLocalReducer, SLocalDataSource *p
#if defined(_DEBUG_VIEW)
printf("new page load to buffer\n");
tColModelDisplay(pOneInterDataSrc->pMemBuffer->pColModel, pOneInterDataSrc->filePage.data,
pOneInterDataSrc->filePage.numOfElems, pOneInterDataSrc->pMemBuffer->pColModel->maxCapacity);
tColModelDisplay(pOneInterDataSrc->pMemBuffer->pColumnModel, pOneInterDataSrc->filePage.data,
pOneInterDataSrc->filePage.numOfElems, pOneInterDataSrc->pMemBuffer->pColumnModel->capacity);
#endif
*needAdjustLoserTree = true;
} else {
@ -759,7 +759,7 @@ void savePrevRecordAndSetupInterpoInfo(SLocalReducer *pLocalReducer, SQueryInfo*
pLocalReducer->discard = true;
pLocalReducer->discardData->numOfElems = 0;
tColModel *pModel = pLocalReducer->pDesc->pSchema;
SColumnModel *pModel = pLocalReducer->pDesc->pColumnModel;
tColModelAppend(pModel, pLocalReducer->discardData, pLocalReducer->prevRowOfInput, 0, 1, 1);
}
@ -782,11 +782,12 @@ static void reversedCopyResultToDstBuf(SQueryInfo* pQueryInfo, SSqlRes *pRes, tF
}
static void reversedCopyFromInterpolationToDstBuf(SQueryInfo* pQueryInfo, SSqlRes *pRes, tFilePage **pResPages, SLocalReducer *pLocalReducer) {
assert(0);
for (int32_t i = 0; i < pQueryInfo->exprsInfo.numOfExprs; ++i) {
TAOS_FIELD *pField = tscFieldInfoGetField(pQueryInfo, i);
int32_t offset = tscFieldInfoGetOffset(pQueryInfo, i);
assert(offset == pLocalReducer->resColModel->colOffset[i]);
assert(offset == getColumnModelOffset(pLocalReducer->resColModel, i));
char *src = pResPages[i]->data + (pRes->numOfRows - 1) * pField->bytes;
char *dst = pRes->data + pRes->numOfRows * offset;
@ -880,7 +881,7 @@ static void doInterpolateResult(SSqlObj *pSql, SLocalReducer *pLocalReducer, boo
tFilePage **pResPages = malloc(POINTER_BYTES * pQueryInfo->fieldsInfo.numOfOutputCols);
for (int32_t i = 0; i < pQueryInfo->fieldsInfo.numOfOutputCols; ++i) {
TAOS_FIELD *pField = tscFieldInfoGetField(pQueryInfo, i);
pResPages[i] = calloc(1, sizeof(tFilePage) + pField->bytes * pLocalReducer->resColModel->maxCapacity);
pResPages[i] = calloc(1, sizeof(tFilePage) + pField->bytes * pLocalReducer->resColModel->capacity);
}
char ** srcData = (char **)malloc((POINTER_BYTES + sizeof(int32_t)) * pQueryInfo->fieldsInfo.numOfOutputCols);
@ -899,11 +900,11 @@ static void doInterpolateResult(SSqlObj *pSql, SLocalReducer *pLocalReducer, boo
TSKEY etime = taosGetRevisedEndKey(actualETime, pQueryInfo->order.order, pQueryInfo->nAggTimeInterval, pQueryInfo->intervalTimeUnit,
precision);
int32_t nrows = taosGetNumOfResultWithInterpo(pInterpoInfo, pPrimaryKeys, remains, pQueryInfo->nAggTimeInterval, etime,
pLocalReducer->resColModel->maxCapacity);
pLocalReducer->resColModel->capacity);
int32_t newRows = taosDoInterpoResult(pInterpoInfo, pQueryInfo->interpoType, pResPages, remains, nrows,
pQueryInfo->nAggTimeInterval, pPrimaryKeys, pLocalReducer->resColModel, srcData,
pQueryInfo->defaultVal, functions, pLocalReducer->resColModel->maxCapacity);
pQueryInfo->defaultVal, functions, pLocalReducer->resColModel->capacity);
assert(newRows <= nrows);
if (pQueryInfo->limit.offset < newRows) {
@ -960,11 +961,10 @@ static void doInterpolateResult(SSqlObj *pSql, SLocalReducer *pLocalReducer, boo
if (pQueryInfo->order.order == TSQL_SO_ASC) {
for (int32_t i = 0; i < pQueryInfo->fieldsInfo.numOfOutputCols; ++i) {
TAOS_FIELD *pField = tscFieldInfoGetField(pQueryInfo, i);
memcpy(pRes->data + pLocalReducer->resColModel->colOffset[i] * pRes->numOfRows, pResPages[i]->data,
pField->bytes * pRes->numOfRows);
int16_t offset = getColumnModelOffset(pLocalReducer->resColModel, i);
memcpy(pRes->data + offset * pRes->numOfRows, pResPages[i]->data, pField->bytes * pRes->numOfRows);
}
} else {
} else {//todo bug??
reversedCopyFromInterpolationToDstBuf(pQueryInfo, pRes, pResPages, pLocalReducer);
}
}
@ -979,13 +979,15 @@ static void doInterpolateResult(SSqlObj *pSql, SLocalReducer *pLocalReducer, boo
}
static void savePreviousRow(SLocalReducer *pLocalReducer, tFilePage *tmpBuffer) {
tColModel *pColModel = pLocalReducer->pDesc->pSchema;
assert(pColModel->maxCapacity == 1 && tmpBuffer->numOfElems == 1);
SColumnModel *pColumnModel = pLocalReducer->pDesc->pColumnModel;
assert(pColumnModel->capacity == 1 && tmpBuffer->numOfElems == 1);
// copy to previous temp buffer
for (int32_t i = 0; i < pLocalReducer->pDesc->pSchema->numOfCols; ++i) {
memcpy(pLocalReducer->prevRowOfInput + pColModel->colOffset[i], tmpBuffer->data + pColModel->colOffset[i],
pColModel->pFields[i].bytes);
for (int32_t i = 0; i < pColumnModel->numOfCols; ++i) {
SSchema* pSchema = getColumnModelSchema(pColumnModel, i);
int16_t offset = getColumnModelOffset(pColumnModel, i);
memcpy(pLocalReducer->prevRowOfInput + offset, tmpBuffer->data + offset, pSchema->bytes);
}
tmpBuffer->numOfElems = 0;
@ -1127,7 +1129,7 @@ bool needToMerge(SQueryInfo* pQueryInfo, SLocalReducer *pLocalReducer, tFilePage
ret = 1; // disable merge procedure
} else {
tOrderDescriptor *pDesc = pLocalReducer->pDesc;
if (pDesc->orderIdx.numOfOrderedCols > 0) {
if (pDesc->orderIdx.numOfCols > 0) {
if (pDesc->tsOrder == TSQL_SO_ASC) { // asc
// todo refactor comparator
ret = compare_a(pLocalReducer->pDesc, 1, 0, pLocalReducer->prevRowOfInput, 1, 0, tmpBuffer->data);
@ -1177,7 +1179,7 @@ bool doGenerateFinalResults(SSqlObj *pSql, SLocalReducer *pLocalReducer, bool no
SQueryInfo* pQueryInfo = tscGetQueryInfoDetail(pCmd, pCmd->clauseIndex);
tFilePage *pResBuf = pLocalReducer->pResultBuf;
tColModel *pModel = pLocalReducer->resColModel;
SColumnModel *pModel = pLocalReducer->resColModel;
pRes->code = TSDB_CODE_SUCCESS;
@ -1192,7 +1194,7 @@ bool doGenerateFinalResults(SSqlObj *pSql, SLocalReducer *pLocalReducer, bool no
return false;
}
tColModelCompact(pModel, pResBuf, pModel->maxCapacity);
tColModelCompact(pModel, pResBuf, pModel->capacity);
memcpy(pLocalReducer->pBufForInterpo, pResBuf->data, pLocalReducer->nResultBufSize);
#ifdef _DEBUG_VIEW
@ -1204,9 +1206,11 @@ bool doGenerateFinalResults(SSqlObj *pSql, SLocalReducer *pLocalReducer, bool no
int32_t startIndex = pQueryInfo->fieldsInfo.numOfOutputCols - pQueryInfo->groupbyExpr.numOfGroupCols;
for (int32_t i = 0; i < pQueryInfo->groupbyExpr.numOfGroupCols; ++i) {
int16_t offset = getColumnModelOffset(pModel, startIndex + i);
SSchema* pSchema = getColumnModelSchema(pModel, startIndex + i);
memcpy(pInterpoInfo->pTags[i],
pLocalReducer->pBufForInterpo + pModel->colOffset[startIndex + i] * pResBuf->numOfElems,
pModel->pFields[startIndex + i].bytes);
pLocalReducer->pBufForInterpo + offset * pResBuf->numOfElems, pSchema->bytes);
}
taosInterpoSetStartInfo(&pLocalReducer->interpolationInfo, pResBuf->numOfElems, pQueryInfo->interpoType);
@ -1218,7 +1222,7 @@ bool doGenerateFinalResults(SSqlObj *pSql, SLocalReducer *pLocalReducer, bool no
void resetOutputBuf(SQueryInfo* pQueryInfo, SLocalReducer *pLocalReducer) { // reset output buffer to the beginning
for (int32_t i = 0; i < pQueryInfo->fieldsInfo.numOfOutputCols; ++i) {
pLocalReducer->pCtx[i].aOutputBuf =
pLocalReducer->pResultBuf->data + tscFieldInfoGetOffset(pQueryInfo, i) * pLocalReducer->resColModel->maxCapacity;
pLocalReducer->pResultBuf->data + tscFieldInfoGetOffset(pQueryInfo, i) * pLocalReducer->resColModel->capacity;
}
memset(pLocalReducer->pResultBuf, 0, pLocalReducer->nResultBufSize + sizeof(tFilePage));
@ -1270,7 +1274,7 @@ static bool doInterpolationForCurrentGroup(SSqlObj *pSql) {
int32_t remain = taosNumOfRemainPoints(pInterpoInfo);
TSKEY ekey = taosGetRevisedEndKey(etime, pQueryInfo->order.order, pQueryInfo->nAggTimeInterval, pQueryInfo->intervalTimeUnit, p);
int32_t rows = taosGetNumOfResultWithInterpo(pInterpoInfo, (TSKEY *)pLocalReducer->pBufForInterpo, remain,
pQueryInfo->nAggTimeInterval, ekey, pLocalReducer->resColModel->maxCapacity);
pQueryInfo->nAggTimeInterval, ekey, pLocalReducer->resColModel->capacity);
if (rows > 0) { // do interpo
doInterpolateResult(pSql, pLocalReducer, false);
}
@ -1302,7 +1306,7 @@ static bool doHandleLastRemainData(SSqlObj *pSql) {
etime = taosGetRevisedEndKey(etime, pQueryInfo->order.order, pQueryInfo->nAggTimeInterval, pQueryInfo->intervalTimeUnit, precision);
int32_t rows = taosGetNumOfResultWithInterpo(pInterpoInfo, NULL, 0, pQueryInfo->nAggTimeInterval, etime,
pLocalReducer->resColModel->maxCapacity);
pLocalReducer->resColModel->capacity);
if (rows > 0) { // do interpo
doInterpolateResult(pSql, pLocalReducer, true);
}
@ -1391,7 +1395,7 @@ int32_t tscDoLocalreduce(SSqlObj *pSql) {
// clear buffer
handleUnprocessedRow(pCmd, pLocalReducer, tmpBuffer);
tColModel *pModel = pLocalReducer->pDesc->pSchema;
SColumnModel *pModel = pLocalReducer->pDesc->pColumnModel;
while (1) {
if (isAllSourcesCompleted(pLocalReducer)) {
@ -1408,14 +1412,14 @@ int32_t tscDoLocalreduce(SSqlObj *pSql) {
SLocalDataSource *pOneDataSrc = pLocalReducer->pLocalDataSrc[pTree->pNode[0].index];
tColModelAppend(pModel, tmpBuffer, pOneDataSrc->filePage.data, pOneDataSrc->rowIdx, 1,
pOneDataSrc->pMemBuffer->pColModel->maxCapacity);
pOneDataSrc->pMemBuffer->pColumnModel->capacity);
#if defined(_DEBUG_VIEW)
printf("chosen row:\t");
SSrcColumnInfo colInfo[256] = {0};
tscGetSrcColumnInfo(colInfo, pQueryInfo);
tColModelDisplayEx(pModel, tmpBuffer->data, tmpBuffer->numOfElems, pModel->maxCapacity, colInfo);
tColModelDisplayEx(pModel, tmpBuffer->data, tmpBuffer->numOfElems, pModel->capacity, colInfo);
#endif
if (pLocalReducer->discard) {
@ -1470,7 +1474,7 @@ int32_t tscDoLocalreduce(SSqlObj *pSql) {
* continue to process results instead of return results.
*/
if ((!sameGroup && pResBuf->numOfElems > 0) ||
(pResBuf->numOfElems == pLocalReducer->resColModel->maxCapacity)) {
(pResBuf->numOfElems == pLocalReducer->resColModel->capacity)) {
// does not belong to the same group
bool notSkipped = doGenerateFinalResults(pSql, pLocalReducer, !sameGroup);

14
src/client/src/tscServer.c
View File

@ -901,7 +901,7 @@ int tscLaunchSTableSubqueries(SSqlObj *pSql) {
tExtMemBuffer ** pMemoryBuf = NULL;
tOrderDescriptor *pDesc = NULL;
tColModel * pModel = NULL;
SColumnModel * pModel = NULL;
pRes->qhandle = 1; // hack the qhandle check
@ -1194,7 +1194,7 @@ void tscRetrieveFromVnodeCallBack(void *param, TAOS_RES *tres, int numOfRows) {
SSrcColumnInfo colInfo[256] = {0};
tscGetSrcColumnInfo(colInfo, pQueryInfo);
tColModelDisplayEx(pDesc->pSchema, pRes->data, pRes->numOfRows, pRes->numOfRows, colInfo);
tColModelDisplayEx(pDesc->pColumnModel, pRes->data, pRes->numOfRows, pRes->numOfRows, colInfo);
#endif
if (tsTotalTmpDirGB != 0 && tsAvailTmpDirGB < tsMinimalTmpDirGB) {
tscError("%p sub:%p client disk space remain %.3f GB, need at least %.3f GB, stop query", pPObj, pSql,
@ -1214,17 +1214,17 @@ void tscRetrieveFromVnodeCallBack(void *param, TAOS_RES *tres, int numOfRows) {
} else { // all data has been retrieved to client
/* data in from current vnode is stored in cache and disk */
uint32_t numOfRowsFromVnode = trsupport->pExtMemBuffer[idx]->numOfAllElems + trsupport->localBuffer->numOfElems;
uint32_t numOfRowsFromVnode = trsupport->pExtMemBuffer[idx]->numOfTotalElems + trsupport->localBuffer->numOfElems;
tscTrace("%p sub:%p all data retrieved from ip:%u,vid:%d, numOfRows:%d, orderOfSub:%d", pPObj, pSql, pSvd->ip,
pSvd->vnode, numOfRowsFromVnode, idx);
tColModelCompact(pDesc->pSchema, trsupport->localBuffer, pDesc->pSchema->maxCapacity);
tColModelCompact(pDesc->pColumnModel, trsupport->localBuffer, pDesc->pColumnModel->capacity);
#ifdef _DEBUG_VIEW
printf("%" PRIu64 " rows data flushed to disk:\n", trsupport->localBuffer->numOfElems);
SSrcColumnInfo colInfo[256] = {0};
tscGetSrcColumnInfo(colInfo, pQueryInfo);
tColModelDisplayEx(pDesc->pSchema, trsupport->localBuffer->data, trsupport->localBuffer->numOfElems,
tColModelDisplayEx(pDesc->pColumnModel, trsupport->localBuffer->data, trsupport->localBuffer->numOfElems,
trsupport->localBuffer->numOfElems, colInfo);
#endif
@ -1256,7 +1256,7 @@ void tscRetrieveFromVnodeCallBack(void *param, TAOS_RES *tres, int numOfRows) {
}
// all sub-queries are returned, start to local merge process
pDesc->pSchema->maxCapacity = trsupport->pExtMemBuffer[idx]->numOfElemsPerPage;
pDesc->pColumnModel->capacity = trsupport->pExtMemBuffer[idx]->numOfElemsPerPage;
tscTrace("%p retrieve from %d vnodes completed.final NumOfRows:%d,start to build loser tree", pPObj,
pState->numOfTotal, pState->numOfRetrievedRows);
@ -1516,7 +1516,7 @@ void tscUpdateVnodeInQueryMsg(SSqlObj *pSql, char *buf) {
char * pStart = buf + tsRpcHeadSize;
SQueryMeterMsg *pQueryMsg = (SQueryMeterMsg *)pStart;
if (UTIL_METER_IS_NOMRAL_METER(pMeterMetaInfo)) { // pSchema == NULL, query on meter
if (UTIL_METER_IS_NOMRAL_METER(pMeterMetaInfo)) { // pColumnModel == NULL, query on meter
SMeterMeta *pMeterMeta = pMeterMetaInfo->pMeterMeta;
pQueryMsg->vnode = htons(pMeterMeta->vpeerDesc[pSql->index].vnode);
} else { // query on metric

220
src/inc/textbuffer.h
View File

@ -19,19 +19,16 @@
extern "C" {
#endif
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "tutil.h"
#include "os.h"
#include "taosmsg.h"
#include "tutil.h"
#define DEFAULT_PAGE_SIZE 16384 // 16k larger than the SHistoInfo
#define MIN_BUFFER_SIZE (1 << 19)
#define MAX_TMPFILE_PATH_LENGTH PATH_MAX
#define INITIAL_ALLOCATION_BUFFER_SIZE 64
#define DEFAULT_PAGE_SIZE 16384 // 16k larger than the SHistoInfo
#define MIN_BUFFER_SIZE (1 << 19)
#define MAX_TMPFILE_PATH_LENGTH PATH_MAX
#define INITIAL_ALLOCATION_BUFFER_SIZE 64
// forward declare
// forward declaration
struct tTagSchema;
typedef enum EXT_BUFFER_FLUSH_MODEL {
@ -61,12 +58,12 @@ typedef struct tFlushoutData {
tFlushoutInfo *pFlushoutInfo;
} tFlushoutData;
typedef struct tFileMeta {
typedef struct SFileInfo {
uint32_t nFileSize; // in pages
uint32_t nPageSize;
uint32_t pageSize;
uint32_t numOfElemsInFile;
tFlushoutData flushoutData;
} tFileMeta;
} SFileInfo;
typedef struct tFilePage {
uint64_t numOfElems;
@ -78,65 +75,73 @@ typedef struct tFilePagesItem {
tFilePage item;
} tFilePagesItem;
typedef struct tColModel {
int32_t maxCapacity;
int32_t numOfCols;
int16_t * colOffset;
struct SSchema *pFields;
} tColModel;
typedef struct SSchemaEx {
struct SSchema field;
int16_t offset;
} SSchemaEx;
typedef struct tOrderIdx {
int32_t numOfOrderedCols;
typedef struct SColumnModel {
int32_t capacity;
int32_t numOfCols;
int16_t rowSize;
SSchemaEx *pFields;
} SColumnModel;
typedef struct SColumnOrderInfo {
int32_t numOfCols;
int16_t pData[];
} tOrderIdx;
} SColumnOrderInfo;
typedef struct tOrderDescriptor {
union {
struct tTagSchema *pTagSchema;
tColModel * pSchema;
};
int32_t tsOrder; // timestamp order type if exists
tOrderIdx orderIdx;
SColumnModel * pColumnModel;
int32_t tsOrder; // timestamp order type if exists
SColumnOrderInfo orderIdx;
} tOrderDescriptor;
typedef struct tExtMemBuffer {
int32_t nMaxSizeInPages;
int32_t inMemCapacity;
int32_t nElemSize;
int32_t nPageSize;
int32_t numOfAllElems;
int32_t pageSize;
int32_t numOfTotalElems;
int32_t numOfElemsInBuffer;
int32_t numOfElemsPerPage;
int16_t numOfInMemPages;
int16_t numOfPagesInMem;
tFilePagesItem *pHead;
tFilePagesItem *pTail;
tFileMeta fileMeta;
char dataFilePath[MAX_TMPFILE_PATH_LENGTH];
FILE *dataFile;
tColModel *pColModel;
char * path;
FILE * file;
SFileInfo fileMeta;
SColumnModel * pColumnModel;
EXT_BUFFER_FLUSH_MODEL flushModel;
} tExtMemBuffer;
/**
*
* @param fileNamePattern
* @param dstPath
*/
void getTmpfilePath(const char *fileNamePattern, char *dstPath);
/*
* create ext-memory buffer
/**
*
* @param inMemSize
* @param elemSize
* @param pModel
* @return
*/
void tExtMemBufferCreate(tExtMemBuffer **pMemBuffer, int32_t numOfBufferSize, int32_t elemSize,
const char *tmpDataFilePath, tColModel *pModel);
tExtMemBuffer *createExtMemBuffer(int32_t inMemSize, int32_t elemSize, SColumnModel *pModel);
/*
* destroy ext-memory buffer
/**
*
* @param pMemBuffer
* @return
*/
void tExtMemBufferDestroy(tExtMemBuffer **pMemBuffer);
void *destoryExtMemBuffer(tExtMemBuffer *pMemBuffer);
/*
/**
* @param pMemBuffer
* @param data input data pointer
* @param numOfRows number of rows in data
@ -145,12 +150,15 @@ void tExtMemBufferDestroy(tExtMemBuffer **pMemBuffer);
*/
int16_t tExtMemBufferPut(tExtMemBuffer *pMemBuffer, void *data, int32_t numOfRows);
/*
* flush all data into disk and release all in-memory buffer
/**
*
* @param pMemBuffer
* @return
*/
bool tExtMemBufferFlush(tExtMemBuffer *pMemBuffer);
/*
/**
*
* remove all data that has been put into buffer, including in buffer or
* ext-buffer(disk)
*/
@ -163,11 +171,44 @@ void tExtMemBufferClear(tExtMemBuffer *pMemBuffer);
*/
bool tExtMemBufferLoadData(tExtMemBuffer *pMemBuffer, tFilePage *pFilePage, int32_t flushIdx, int32_t pageIdx);
/**
*
* @param pMemBuffer
* @return
*/
bool tExtMemBufferIsAllDataInMem(tExtMemBuffer *pMemBuffer);
tColModel *tColModelCreate(SSchema *field, int32_t numOfCols, int32_t maxCapacity);
/**
*
* @param fields
* @param numOfCols
* @param blockCapacity
* @return
*/
SColumnModel *createColumnModel(SSchema *fields, int32_t numOfCols, int32_t blockCapacity);
void tColModelDestroy(tColModel *pModel);
/**
*
* @param pSrc
* @return
*/
SColumnModel *cloneColumnModel(SColumnModel *pSrc);
/**
*
* @param pModel
*/
void destroyColumnModel(SColumnModel *pModel);
/*
* compress data into consecutive block without hole in data
*/
void tColModelCompact(SColumnModel *pModel, tFilePage *inputBuffer, int32_t maxElemsCapacity);
void tColModelErase(SColumnModel *pModel, tFilePage *inputBuffer, int32_t maxCapacity, int32_t s, int32_t e);
SSchema *getColumnModelSchema(SColumnModel *pColumnModel, int32_t index);
int16_t getColumnModelOffset(SColumnModel *pColumnModel, int32_t index);
typedef struct SSrcColumnInfo {
int32_t functionId;
@ -177,68 +218,18 @@ typedef struct SSrcColumnInfo {
/*
* display data in column format model for debug purpose only
*/
void tColModelDisplay(tColModel *pModel, void *pData, int32_t numOfRows, int32_t maxCount);
void tColModelDisplay(SColumnModel *pModel, void *pData, int32_t numOfRows, int32_t maxCount);
void tColModelDisplayEx(tColModel *pModel, void *pData, int32_t numOfRows, int32_t maxCount, SSrcColumnInfo *pInfo);
void tColModelDisplayEx(SColumnModel *pModel, void *pData, int32_t numOfRows, int32_t maxCount, SSrcColumnInfo *pInfo);
/*
* compress data into consecutive block without hole in data
*/
void tColModelCompact(tColModel *pModel, tFilePage *inputBuffer, int32_t maxElemsCapacity);
void tColModelErase(tColModel *pModel, tFilePage *inputBuffer, int32_t maxCapacity, int32_t s, int32_t e);
tOrderDescriptor *tOrderDesCreate(int32_t *orderColIdx, int32_t numOfOrderCols, tColModel *pModel, int32_t tsOrderType);
tOrderDescriptor *tOrderDesCreate(const int32_t *orderColIdx, int32_t numOfOrderCols, SColumnModel *pModel,
int32_t tsOrderType);
void tOrderDescDestroy(tOrderDescriptor *pDesc);
void tColModelAppend(tColModel *dstModel, tFilePage *dstPage, void *srcData, int32_t srcStartRows,
void tColModelAppend(SColumnModel *dstModel, tFilePage *dstPage, void *srcData, int32_t srcStartRows,
int32_t numOfRowsToWrite, int32_t srcCapacity);
///////////////////////////////////////////////////////////////////////////////////////////////////////
typedef struct MinMaxEntry {
union {
double dMinVal;
int32_t iMinVal;
int64_t i64MinVal;
};
union {
double dMaxVal;
int32_t iMaxVal;
int64_t i64MaxVal;
};
} MinMaxEntry;
typedef struct tMemBucketSegment {
int32_t numOfSlots;
MinMaxEntry * pBoundingEntries;
tExtMemBuffer **pBuffer;
} tMemBucketSegment;
typedef struct tMemBucket {
int16_t numOfSegs;
int16_t nTotalSlots;
int16_t nSlotsOfSeg;
int16_t dataType;
int16_t nElemSize;
int32_t numOfElems;
int32_t nTotalBufferSize;
int32_t maxElemsCapacity;
int16_t nPageSize;
int16_t numOfTotalPages;
int16_t numOfAvailPages; /* remain available buffer pages */
tMemBucketSegment *pSegs;
tOrderDescriptor * pOrderDesc;
MinMaxEntry nRange;
void (*HashFunc)(struct tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx);
} tMemBucket;
typedef int (*__col_compar_fn_t)(tOrderDescriptor *, int32_t numOfRows, int32_t idx1, int32_t idx2, char *data);
void tColDataQSort(tOrderDescriptor *, int32_t numOfRows, int32_t start, int32_t end, char *data, int32_t orderType);
@ -253,19 +244,6 @@ int32_t compare_a(tOrderDescriptor *, int32_t numOfRow1, int32_t s1, char *data1
int32_t compare_d(tOrderDescriptor *, int32_t numOfRow1, int32_t s1, char *data1, int32_t numOfRow2, int32_t s2,
char *data2);
tMemBucket* tMemBucketCreate(int32_t totalSlots, int32_t nBufferSize, int16_t nElemSize,
int16_t dataType, tOrderDescriptor *pDesc);
void tMemBucketDestroy(tMemBucket *pBucket);
void tMemBucketPut(tMemBucket *pBucket, void *data, int32_t numOfRows);
double getPercentile(tMemBucket *pMemBucket, double percent);
void tBucketIntHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx);
void tBucketDoubleHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx);
#ifdef __cplusplus
}
#endif

2
src/inc/thistogram.h
View File

@ -20,8 +20,6 @@
extern "C" {
#endif
#include "tskiplist.h"
#define USE_ARRAYLIST
#define MAX_HISTOGRAM_BIN 500

2
src/inc/tinterpolation.h
View File

@ -78,7 +78,7 @@ int32_t taosNumOfRemainPoints(SInterpolationInfo *pInterpoInfo);
*/
int32_t taosDoInterpoResult(SInterpolationInfo *pInterpoInfo, int16_t interpoType, tFilePage **data,
int32_t numOfRawDataInRows, int32_t outputRows, int64_t nInterval,
const int64_t *pPrimaryKeyArray, tColModel *pModel, char **srcData, int64_t *defaultVal,
const int64_t *pPrimaryKeyArray, SColumnModel *pModel, char **srcData, int64_t *defaultVal,
const int32_t *functionIDs, int32_t bufSize);
int taosDoLinearInterpolation(int32_t type, SPoint *point1, SPoint *point2, SPoint *point);

77
src/inc/tpercentile.h Normal file
View File

@ -0,0 +1,77 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef TDENGINE_TPERCENTILE_H
#define TDENGINE_TPERCENTILE_H
#include "textbuffer.h"
typedef struct MinMaxEntry {
union {
double dMinVal;
int32_t iMinVal;
int64_t i64MinVal;
};
union {
double dMaxVal;
int32_t iMaxVal;
int64_t i64MaxVal;
};
} MinMaxEntry;
typedef struct tMemBucketSegment {
int32_t numOfSlots;
MinMaxEntry * pBoundingEntries;
tExtMemBuffer **pBuffer;
} tMemBucketSegment;
typedef struct tMemBucket {
int16_t numOfSegs;
int16_t nTotalSlots;
int16_t nSlotsOfSeg;
int16_t dataType;
int16_t nElemSize;
int32_t numOfElems;
int32_t nTotalBufferSize;
int32_t maxElemsCapacity;
int16_t pageSize;
int16_t numOfTotalPages;
int16_t numOfAvailPages; /* remain available buffer pages */
tMemBucketSegment *pSegs;
tOrderDescriptor * pOrderDesc;
MinMaxEntry nRange;
void (*HashFunc)(struct tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx);
} tMemBucket;
tMemBucket *tMemBucketCreate(int32_t totalSlots, int32_t nBufferSize, int16_t nElemSize, int16_t dataType,
tOrderDescriptor *pDesc);
void tMemBucketDestroy(tMemBucket *pBucket);
void tMemBucketPut(tMemBucket *pBucket, void *data, int32_t numOfRows);
double getPercentile(tMemBucket *pMemBucket, double percent);
void tBucketIntHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx);
void tBucketDoubleHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx);
#endif // TDENGINE_TPERCENTILE_H

18
src/system/detail/inc/vnodeTagMgmt.h
View File

@ -33,11 +33,13 @@ extern "C" {
* 1. we implement a quick sort algorithm, may remove it later.
*/
typedef struct tTagSchema {
struct SSchema *pSchema;
int32_t numOfCols;
int32_t colOffset[];
} tTagSchema;
//typedef struct tTagSchema {
// struct SSchema *pSchema;
// int32_t numOfCols;
// int32_t colOffset[];
//} tTagSchema;
typedef SColumnModel tTagSchema;
typedef struct tSidSet {
int32_t numOfSids;
@ -45,8 +47,8 @@ typedef struct tSidSet {
SMeterSidExtInfo **pSids;
int32_t * starterPos; // position of each subgroup, generated according to
tTagSchema *pTagSchema;
tOrderIdx orderIdx;
SColumnModel *pColumnModel;
SColumnOrderInfo orderIdx;
} tSidSet;
typedef int32_t (*__ext_compar_fn_t)(const void *p1, const void *p2, void *param);
@ -54,7 +56,7 @@ typedef int32_t (*__ext_compar_fn_t)(const void *p1, const void *p2, void *param
tSidSet *tSidSetCreate(struct SMeterSidExtInfo **pMeterSidExtInfo, int32_t numOfMeters, SSchema *pSchema,
int32_t numOfTags, SColIndexEx *colList, int32_t numOfOrderCols);
tTagSchema *tCreateTagSchema(SSchema *pSchema, int32_t numOfTagCols);
//tTagSchema *tCreateTagSchema(SSchema *pSchema, int32_t numOfTagCols);
int32_t *calculateSubGroup(void **pSids, int32_t numOfMeters, int32_t *numOfSubset, tOrderDescriptor *pOrderDesc,
__ext_compar_fn_t compareFn);

2
src/system/detail/src/mgmtDnodeInt.c
View File

@ -210,7 +210,7 @@ char *mgmtBuildCreateMeterIe(STabObj *pMeter, char *pMsg, int vnode) {
for (int i = 0; i < pMeter->numOfColumns; ++i) {
pCreateMeter->schema[i].type = pSchema[i].type;
/* strcpy(pCreateMeter->schema[i].name, pSchema[i].name); */
/* strcpy(pCreateMeter->schema[i].name, pColumnModel[i].name); */
pCreateMeter->schema[i].bytes = htons(pSchema[i].bytes);
pCreateMeter->schema[i].colId = htons(pSchema[i].colId);
}

24
src/system/detail/src/mgmtSupertableQuery.c
View File

@ -70,7 +70,7 @@ static int32_t tabObjResultComparator(const void* p1, const void* p2, void* para
STabObj* pNode1 = (STabObj*)p1;
STabObj* pNode2 = (STabObj*)p2;
for (int32_t i = 0; i < pOrderDesc->orderIdx.numOfOrderedCols; ++i) {
for (int32_t i = 0; i < pOrderDesc->orderIdx.numOfCols; ++i) {
int32_t colIdx = pOrderDesc->orderIdx.pData[i];
char* f1 = NULL;
@ -86,7 +86,9 @@ static int32_t tabObjResultComparator(const void* p1, const void* p2, void* para
} else {
f1 = mgmtMeterGetTag(pNode1, colIdx, NULL);
f2 = mgmtMeterGetTag(pNode2, colIdx, &schema);
assert(schema.type == pOrderDesc->pTagSchema->pSchema[colIdx].type);
SSchema* pSchema = getColumnModelSchema(pOrderDesc->pColumnModel, colIdx);
assert(schema.type == pSchema->type);
}
int32_t ret = doCompare(f1, f2, schema.type, schema.bytes);
@ -109,7 +111,7 @@ static int32_t tabObjResultComparator(const void* p1, const void* p2, void* para
* @param pOrderIndexInfo
* @param numOfTags
*/
static void mgmtUpdateOrderTagColIndex(SMetricMetaMsg* pMetricMetaMsg, int32_t tableIndex, tOrderIdx* pOrderIndexInfo,
static void mgmtUpdateOrderTagColIndex(SMetricMetaMsg* pMetricMetaMsg, int32_t tableIndex, SColumnOrderInfo* pOrderIndexInfo,
int32_t numOfTags) {
SMetricMetaElemMsg* pElem = (SMetricMetaElemMsg*)((char*)pMetricMetaMsg + pMetricMetaMsg->metaElem[tableIndex]);
SColIndexEx* groupColumnList = (SColIndexEx*)((char*)pMetricMetaMsg + pElem->groupbyTagColumnList);
@ -123,7 +125,7 @@ static void mgmtUpdateOrderTagColIndex(SMetricMetaMsg* pMetricMetaMsg, int32_t t
}
}
pOrderIndexInfo->numOfOrderedCols = numOfGroupbyTags;
pOrderIndexInfo->numOfCols = numOfGroupbyTags;
}
// todo merge sort function with losertree used
@ -143,14 +145,14 @@ void mgmtReorganizeMetersInMetricMeta(SMetricMetaMsg* pMetricMetaMsg, int32_t ta
*/
tOrderDescriptor* descriptor =
(tOrderDescriptor*)calloc(1, sizeof(tOrderDescriptor) + sizeof(int32_t) * pElem->numOfGroupCols);
descriptor->pTagSchema = tCreateTagSchema(pTagSchema, pMetric->numOfTags);
descriptor->orderIdx.numOfOrderedCols = pElem->numOfGroupCols;
descriptor->pColumnModel = createColumnModel(pTagSchema, pMetric->numOfTags, 1);
descriptor->orderIdx.numOfCols = pElem->numOfGroupCols;
int32_t* startPos = NULL;
int32_t numOfSubset = 1;
mgmtUpdateOrderTagColIndex(pMetricMetaMsg, tableIndex, &descriptor->orderIdx, pMetric->numOfTags);
if (descriptor->orderIdx.numOfOrderedCols > 0) {
if (descriptor->orderIdx.numOfCols > 0) {
tQSortEx(pRes->pRes, POINTER_BYTES, 0, pRes->num - 1, descriptor, tabObjResultComparator);
startPos = calculateSubGroup(pRes->pRes, pRes->num, &numOfSubset, descriptor, tabObjResultComparator);
} else {
@ -166,7 +168,7 @@ void mgmtReorganizeMetersInMetricMeta(SMetricMetaMsg* pMetricMetaMsg, int32_t ta
*/
qsort(pRes->pRes, (size_t)pRes->num, POINTER_BYTES, tabObjVGIDComparator);
free(descriptor->pTagSchema);
free(descriptor->pColumnModel);
free(descriptor);
free(startPos);
}
@ -291,15 +293,15 @@ static void orderResult(SMetricMetaMsg* pMetricMetaMsg, tQueryResultset* pRes, i
STabObj* pMetric = mgmtGetMeter(pElem->meterId);
SSchema* pTagSchema = (SSchema*)(pMetric->schema + pMetric->numOfColumns * sizeof(SSchema));
descriptor->pTagSchema = tCreateTagSchema(pTagSchema, pMetric->numOfTags);
descriptor->pColumnModel = createColumnModel(pTagSchema, pMetric->numOfTags, 1);
descriptor->orderIdx.pData[0] = colIndex;
descriptor->orderIdx.numOfOrderedCols = 1;
descriptor->orderIdx.numOfCols = 1;
// sort results list
tQSortEx(pRes->pRes, POINTER_BYTES, 0, pRes->num - 1, descriptor, tabObjResultComparator);
free(descriptor->pTagSchema);
free(descriptor->pColumnModel);
free(descriptor);
}

2
src/system/detail/src/vnodeFile.c
View File

@ -127,7 +127,7 @@ int vnodeCreateHeadDataFile(int vnode, int fileId, char *headName, char *dataNam
if (symlink(dDataName, dataName) != 0) return -1;
if (symlink(dLastName, lastName) != 0) return -1;
dPrint("vid:%d, fileId:%d, empty header file:%s dataFile:%s lastFile:%s on disk:%s is created ",
dPrint("vid:%d, fileId:%d, empty header file:%s file:%s lastFile:%s on disk:%s is created ",
vnode, fileId, headName, dataName, lastName, path);
return 0;

54
src/system/detail/src/vnodeQueryImpl.c
View File

@ -589,7 +589,7 @@ static void setExecParams(SQuery *pQuery, SQLFunctionCtx *pCtx, int64_t StartQue
char *primaryColumnData, int32_t size, int32_t functionId, SField *pField, bool hasNull,
int32_t blockStatus, void *param, int32_t scanFlag);
void createGroupResultBuf(SQuery *pQuery, SOutputRes *pOneResult, bool isMetricQuery);
void createGroupResultBuf(SQuery *pQuery, SOutputRes *pOneResult, bool isSTableQuery);
static void destroyGroupResultBuf(SOutputRes *pOneOutputRes, int32_t nOutputCols);
static int32_t binarySearchForBlockImpl(SCompBlock *pBlock, int32_t numOfBlocks, TSKEY skey, int32_t order) {
@ -2499,7 +2499,7 @@ static void setCtxTagColumnInfo(SQuery *pQuery, SQueryRuntimeEnv *pRuntimeEnv) {
}
static int32_t setupQueryRuntimeEnv(SMeterObj *pMeterObj, SQuery *pQuery, SQueryRuntimeEnv *pRuntimeEnv,
SSchema *pTagsSchema, int16_t order, bool isMetricQuery) {
tTagSchema *pTagsSchema, int16_t order, bool isSTableQuery) {
dTrace("QInfo:%p setup runtime env", GET_QINFO_ADDR(pQuery));
pRuntimeEnv->pMeterObj = pMeterObj;
@ -2520,8 +2520,10 @@ static int32_t setupQueryRuntimeEnv(SMeterObj *pMeterObj, SQuery *pQuery, SQuery
SQLFunctionCtx *pCtx = &pRuntimeEnv->pCtx[i];
if (TSDB_COL_IS_TAG(pSqlFuncMsg->colInfo.flag)) { // process tag column info
pCtx->inputType = pTagsSchema[pColIndexEx->colIdx].type;
pCtx->inputBytes = pTagsSchema[pColIndexEx->colIdx].bytes;
SSchema* pSchema = getColumnModelSchema(pTagsSchema, pColIndexEx->colIdx);
pCtx->inputType = pSchema->type;
pCtx->inputBytes = pSchema->bytes;
} else {
pCtx->inputType = GET_COLUMN_TYPE(pQuery, i);
pCtx->inputBytes = GET_COLUMN_BYTES(pQuery, i);
@ -2567,11 +2569,11 @@ static int32_t setupQueryRuntimeEnv(SMeterObj *pMeterObj, SQuery *pQuery, SQuery
// set the intermediate result output buffer
SResultInfo *pResInfo = &pRuntimeEnv->resultInfo[i];
setResultInfoBuf(pResInfo, pQuery->pSelectExpr[i].interResBytes, isMetricQuery);
setResultInfoBuf(pResInfo, pQuery->pSelectExpr[i].interResBytes, isSTableQuery);
}
// if it is group by normal column, do not set output buffer, the output buffer is pResult
if (!isGroupbyNormalCol(pQuery->pGroupbyExpr) && !isMetricQuery) {
if (!isGroupbyNormalCol(pQuery->pGroupbyExpr) && !isSTableQuery) {
resetCtxOutputBuf(pRuntimeEnv);
}
@ -4120,7 +4122,7 @@ static void allocMemForInterpo(SMeterQuerySupportObj *pSupporter, SQuery *pQuery
}
}
static int32_t allocateOutputBufForGroup(SMeterQuerySupportObj *pSupporter, SQuery *pQuery, bool isMetricQuery) {
static int32_t allocateOutputBufForGroup(SMeterQuerySupportObj *pSupporter, SQuery *pQuery, bool isSTableQuery) {
int32_t slot = 0;
if (isGroupbyNormalCol(pQuery->pGroupbyExpr) || (pQuery->nAggTimeInterval > 0 && pQuery->slidingTime > 0)) {
@ -4142,14 +4144,14 @@ static int32_t allocateOutputBufForGroup(SMeterQuerySupportObj *pSupporter, SQue
* for single table top/bottom query, the output for group by normal column, the output rows is
* equals to the maximum rows, instead of 1.
*/
if (!isMetricQuery && isTopBottomQuery(pQuery)) {
if (!isSTableQuery && isTopBottomQuery(pQuery)) {
assert(pQuery->numOfOutputCols > 1);
SSqlFunctionExpr *pExpr = &pQuery->pSelectExpr[1];
pOneRes->nAlloc = pExpr->pBase.arg[0].argValue.i64;
}
createGroupResultBuf(pQuery, pOneRes, isMetricQuery);
createGroupResultBuf(pQuery, pOneRes, isSTableQuery);
}
return TSDB_CODE_SUCCESS;
@ -4498,12 +4500,12 @@ int32_t vnodeMultiMeterQueryPrepare(SQInfo *pQInfo, SQuery *pQuery, void *param)
pQuery->lastKey = pQuery->skey;
// create runtime environment
SSchema *pTagSchema = NULL;
// SSchema *pColumnModel = NULL;
tTagSchema *pTagSchemaInfo = pSupporter->pSidSet->pTagSchema;
if (pTagSchemaInfo != NULL) {
pTagSchema = pTagSchemaInfo->pSchema;
}
tTagSchema *pTagSchemaInfo = pSupporter->pSidSet->pColumnModel;
// if (pTagSchemaInfo != NULL) {
// pColumnModel = pTagSchemaInfo->pSchema;
// }
// get one queried meter
SMeterObj *pMeter = getMeterObj(pSupporter->pMetersHashTable, pSupporter->pSidSet->pSids[0]->sid);
@ -4517,7 +4519,7 @@ int32_t vnodeMultiMeterQueryPrepare(SQInfo *pQInfo, SQuery *pQuery, void *param)
tsBufSetTraverseOrder(pRuntimeEnv->pTSBuf, order);
}
int32_t ret = setupQueryRuntimeEnv(pMeter, pQuery, &pSupporter->runtimeEnv, pTagSchema, TSQL_SO_ASC, true);
int32_t ret = setupQueryRuntimeEnv(pMeter, pQuery, &pSupporter->runtimeEnv, pTagSchemaInfo, TSQL_SO_ASC, true);
if (ret != TSDB_CODE_SUCCESS) {
return ret;
}
@ -5065,10 +5067,10 @@ static void doSetTagValueInParam(tTagSchema *pTagSchema, int32_t tagColIdx, SMet
tVariant *param) {
assert(tagColIdx >= 0);
int32_t *fieldValueOffset = pTagSchema->colOffset;
void * pStr = (char *)pMeterSidInfo->tags + fieldValueOffset[tagColIdx];
SSchema *pCol = &pTagSchema->pSchema[tagColIdx];
int16_t offset = getColumnModelOffset(pTagSchema, tagColIdx);
void * pStr = (char *)pMeterSidInfo->tags + offset;
SSchema *pCol = getColumnModelSchema(pTagSchema, tagColIdx);
tVariantDestroy(param);
@ -5081,7 +5083,7 @@ static void doSetTagValueInParam(tTagSchema *pTagSchema, int32_t tagColIdx, SMet
void vnodeSetTagValueInParam(tSidSet *pSidSet, SQueryRuntimeEnv *pRuntimeEnv, SMeterSidExtInfo *pMeterSidInfo) {
SQuery * pQuery = pRuntimeEnv->pQuery;
tTagSchema *pTagSchema = pSidSet->pTagSchema;
tTagSchema *pTagSchema = pSidSet->pColumnModel;
SSqlFuncExprMsg *pFuncMsg = &pQuery->pSelectExpr[0].pBase;
if (pQuery->numOfOutputCols == 1 && pFuncMsg->functionId == TSDB_FUNC_TS_COMP) {
@ -5691,7 +5693,7 @@ void enableFunctForMasterScan(SQueryRuntimeEnv *pRuntimeEnv, int32_t order) {
pQuery->order.order = (pQuery->order.order ^ 1);
}
void createGroupResultBuf(SQuery *pQuery, SOutputRes *pOneResult, bool isMetricQuery) {
void createGroupResultBuf(SQuery *pQuery, SOutputRes *pOneResult, bool isSTableQuery) {
int32_t numOfOutput = pQuery->numOfOutputCols;
pOneResult->resultInfo = calloc((size_t)numOfOutput, sizeof(SResultInfo));
@ -5704,7 +5706,7 @@ void createGroupResultBuf(SQuery *pQuery, SOutputRes *pOneResult, bool isMetricQ
pOneResult->result[i] = malloc(sizeof(tFilePage) + size * pOneResult->nAlloc);
pOneResult->result[i]->numOfElems = 0;
setResultInfoBuf(pResInfo, (int32_t)size, isMetricQuery);
setResultInfoBuf(pResInfo, (int32_t)size, isSTableQuery);
}
}
@ -7580,7 +7582,7 @@ int32_t saveResult(SMeterQuerySupportObj *pSupporter, SMeterQueryInfo *pMeterQue
sc[1].bytes = 8;
UNUSED(sc);
tColModel *cm = tColModelCreate(sc, pQuery->numOfOutputCols, pRuntimeEnv->numOfRowsPerPage);
SColumnModel *cm = createColumnModel(sc, pQuery->numOfOutputCols, pRuntimeEnv->numOfRowsPerPage);
// if (outputPage->numOfElems + numOfResult >= pRuntimeEnv->numOfRowsPerPage)
tColModelDisplay(cm, outputPage->data, outputPage->numOfElems, pRuntimeEnv->numOfRowsPerPage);
@ -7717,7 +7719,7 @@ static void applyIntervalQueryOnBlock(SMeterQuerySupportObj *pSupporter, SMeterD
saveIntervalQueryRange(pRuntimeEnv, pMeterQueryInfo);
} else {
doApplyIntervalQueryOnBlock(pSupporter, pMeterQueryInfo, pBlockInfo, pPrimaryKey, pFields, searchFn);
doApplyIntervalQueryOnBlock_rv(pSupporter, pMeterQueryInfo, pBlockInfo, pPrimaryKey, pFields, searchFn);
}
}
@ -7802,7 +7804,7 @@ static int32_t resultInterpolate(SQInfo *pQInfo, tFilePage **data, tFilePage **p
pSchema[i].type = pQuery->pSelectExpr[i].resType;
}
tColModel *pModel = tColModelCreate(pSchema, pQuery->numOfOutputCols, pQuery->pointsToRead);
SColumnModel *pModel = createColumnModel(pSchema, pQuery->numOfOutputCols, pQuery->pointsToRead);
char * srcData[TSDB_MAX_COLUMNS] = {0};
int32_t functions[TSDB_MAX_COLUMNS] = {0};
@ -7816,7 +7818,7 @@ static int32_t resultInterpolate(SQInfo *pQInfo, tFilePage **data, tFilePage **p
pQuery->nAggTimeInterval, (int64_t *)pDataSrc[0]->data, pModel, srcData,
pQuery->defaultVal, functions, pRuntimeEnv->pMeterObj->pointsPerFileBlock);
tColModelDestroy(pModel);
destroyColumnModel(pModel);
free(pSchema);
return numOfRes;

66
src/system/detail/src/vnodeShell.c
View File

@ -442,22 +442,22 @@ void vnodeExecuteRetrieveReq(SSchedMsg *pSched) {
if (code == TSDB_CODE_SUCCESS) {
size = vnodeGetResultSize((void *)(pRetrieve->qhandle), &numOfRows);
}
// buffer size for progress information, including meter count,
// and for each meter, including 'uid' and 'TSKEY'.
int progressSize = 0;
if (pQInfo->pMeterQuerySupporter != NULL)
progressSize = pQInfo->pMeterQuerySupporter->numOfMeters * (sizeof(int64_t) + sizeof(TSKEY)) + sizeof(int32_t);
else if (pQInfo->pObj != NULL)
progressSize = sizeof(int64_t) + sizeof(TSKEY) + sizeof(int32_t);
pStart = taosBuildRspMsgWithSize(pObj->thandle, TSDB_MSG_TYPE_RETRIEVE_RSP, progressSize + size + 100);
if (pStart == NULL) {
taosSendSimpleRsp(pObj->thandle, TSDB_MSG_TYPE_RETRIEVE_RSP, TSDB_CODE_SERV_OUT_OF_MEMORY);
goto _exit;
}
// buffer size for progress information, including meter count,
// and for each meter, including 'uid' and 'TSKEY'.
int progressSize = 0;
if (pQInfo->pMeterQuerySupporter != NULL)
progressSize = pQInfo->pMeterQuerySupporter->numOfMeters * (sizeof(int64_t) + sizeof(TSKEY)) + sizeof(int32_t);
else if (pQInfo->pObj != NULL)
progressSize = sizeof(int64_t) + sizeof(TSKEY) + sizeof(int32_t);
pStart = taosBuildRspMsgWithSize(pObj->thandle, TSDB_MSG_TYPE_RETRIEVE_RSP, progressSize + size + 100);
if (pStart == NULL) {
taosSendSimpleRsp(pObj->thandle, TSDB_MSG_TYPE_RETRIEVE_RSP, TSDB_CODE_SERV_OUT_OF_MEMORY);
goto _exit;
}
}
pMsg = pStart;
*pMsg = code;
@ -485,26 +485,28 @@ void vnodeExecuteRetrieveReq(SSchedMsg *pSched) {
// write the progress information of each meter to response
// this is required by subscriptions
if (pQInfo->pMeterQuerySupporter != NULL && pQInfo->pMeterQuerySupporter->pMeterSidExtInfo != NULL) {
*((int32_t*)pMsg) = htonl(pQInfo->pMeterQuerySupporter->numOfMeters);
pMsg += sizeof(int32_t);
for (int32_t i = 0; i < pQInfo->pMeterQuerySupporter->numOfMeters; i++) {
*((int64_t*)pMsg) = htobe64(pQInfo->pMeterQuerySupporter->pMeterSidExtInfo[i]->uid);
if (numOfRows > 0 && code == TSDB_CODE_SUCCESS) {
if (pQInfo->pMeterQuerySupporter != NULL && pQInfo->pMeterQuerySupporter->pMeterSidExtInfo != NULL) {
*((int32_t *)pMsg) = htonl(pQInfo->pMeterQuerySupporter->numOfMeters);
pMsg += sizeof(int32_t);
for (int32_t i = 0; i < pQInfo->pMeterQuerySupporter->numOfMeters; i++) {
*((int64_t *)pMsg) = htobe64(pQInfo->pMeterQuerySupporter->pMeterSidExtInfo[i]->uid);
pMsg += sizeof(int64_t);
*((TSKEY *)pMsg) = htobe64(pQInfo->pMeterQuerySupporter->pMeterSidExtInfo[i]->key);
pMsg += sizeof(TSKEY);
}
} else if (pQInfo->pObj != NULL) {
*((int32_t *)pMsg) = htonl(1);
pMsg += sizeof(int32_t);
*((int64_t *)pMsg) = htobe64(pQInfo->pObj->uid);
pMsg += sizeof(int64_t);
*((TSKEY*)pMsg) = htobe64(pQInfo->pMeterQuerySupporter->pMeterSidExtInfo[i]->key);
if (pQInfo->pointsRead > 0) {
*((TSKEY *)pMsg) = htobe64(pQInfo->query.lastKey + 1);
} else {
*((TSKEY *)pMsg) = htobe64(pQInfo->query.lastKey);
}
pMsg += sizeof(TSKEY);
}
} else if (pQInfo->pObj != NULL) {
*((int32_t*)pMsg) = htonl(1);
pMsg += sizeof(int32_t);
*((int64_t*)pMsg) = htobe64(pQInfo->pObj->uid);
pMsg += sizeof(int64_t);
if (pQInfo->pointsRead > 0) {
*((TSKEY*)pMsg) = htobe64(pQInfo->query.lastKey + 1);
} else {
*((TSKEY*)pMsg) = htobe64(pQInfo->query.lastKey);
}
pMsg += sizeof(TSKEY);
}
msgLen = pMsg - pStart;

86
src/system/detail/src/vnodeTagMgmt.c
View File

@ -24,10 +24,10 @@
#include "tast.h"
#include "vnodeTagMgmt.h"
#define GET_TAG_VAL_POINTER(s, col, sc, t) ((t *)(&((s)->tags[(sc)->colOffset[(col)]])))
#define GET_TAG_VAL_POINTER(s, col, sc, t) ((t *)(&((s)->tags[getColumnModelOffset(sc, col)])))
#define GET_TAG_VAL(s, col, sc, t) (*GET_TAG_VAL_POINTER(s, col, sc, t))
static void tTagsPrints(SMeterSidExtInfo *pMeterInfo, tTagSchema *pSchema, tOrderIdx *pOrder);
static void tTagsPrints(SMeterSidExtInfo *pMeterInfo, tTagSchema *pSchema, SColumnOrderInfo *pOrder);
static void tSidSetDisplay(tSidSet *pSets);
@ -65,7 +65,7 @@ int32_t meterSidComparator(const void *p1, const void *p2, void *param) {
SMeterSidExtInfo *s1 = (SMeterSidExtInfo *)p1;
SMeterSidExtInfo *s2 = (SMeterSidExtInfo *)p2;
for (int32_t i = 0; i < pOrderDesc->orderIdx.numOfOrderedCols; ++i) {
for (int32_t i = 0; i < pOrderDesc->orderIdx.numOfCols; ++i) {
int32_t colIdx = pOrderDesc->orderIdx.pData[i];
char * f1 = NULL;
@ -79,9 +79,9 @@ int32_t meterSidComparator(const void *p1, const void *p2, void *param) {
type = TSDB_DATA_TYPE_BINARY;
bytes = TSDB_METER_NAME_LEN;
} else {
f1 = GET_TAG_VAL_POINTER(s1, colIdx, pOrderDesc->pTagSchema, char);
f2 = GET_TAG_VAL_POINTER(s2, colIdx, pOrderDesc->pTagSchema, char);
SSchema *pSchema = &pOrderDesc->pTagSchema->pSchema[colIdx];
f1 = GET_TAG_VAL_POINTER(s1, colIdx, pOrderDesc->pColumnModel, char);
f2 = GET_TAG_VAL_POINTER(s2, colIdx, pOrderDesc->pColumnModel, char);
SSchema *pSchema = getColumnModelSchema(pOrderDesc->pColumnModel, colIdx);
type = pSchema->type;
bytes = pSchema->bytes;
}
@ -116,9 +116,9 @@ static void median(void **pMeterSids, size_t size, int32_t s1, int32_t s2, tOrde
compareFn(pMeterSids[s1], pMeterSids[s2], pOrderDesc) <= 0);
#ifdef _DEBUG_VIEW
tTagsPrints(pMeterSids[s1], pOrderDesc->pTagSchema, &pOrderDesc->orderIdx);
tTagsPrints(pMeterSids[midIdx], pOrderDesc->pTagSchema, &pOrderDesc->orderIdx);
tTagsPrints(pMeterSids[s2], pOrderDesc->pTagSchema, &pOrderDesc->orderIdx);
tTagsPrints(pMeterSids[s1], pOrderDesc->pColumnModel, &pOrderDesc->orderIdx);
tTagsPrints(pMeterSids[midIdx], pOrderDesc->pColumnModel, &pOrderDesc->orderIdx);
tTagsPrints(pMeterSids[s2], pOrderDesc->pColumnModel, &pOrderDesc->orderIdx);
#endif
}
@ -241,24 +241,24 @@ int32_t *calculateSubGroup(void **pSids, int32_t numOfMeters, int32_t *numOfSubs
return starterPos;
}
tTagSchema *tCreateTagSchema(SSchema *pSchema, int32_t numOfTagCols) {
if (numOfTagCols == 0 || pSchema == NULL) {
return NULL;
}
tTagSchema *pTagSchema =
(tTagSchema *)calloc(1, sizeof(tTagSchema) + numOfTagCols * sizeof(int32_t) + sizeof(SSchema) * numOfTagCols);
pTagSchema->colOffset[0] = 0;
pTagSchema->numOfCols = numOfTagCols;
for (int32_t i = 1; i < numOfTagCols; ++i) {
pTagSchema->colOffset[i] = (pTagSchema->colOffset[i - 1] + pSchema[i - 1].bytes);
}
pTagSchema->pSchema = (SSchema *)&(pTagSchema->colOffset[numOfTagCols]);
memcpy(pTagSchema->pSchema, pSchema, sizeof(SSchema) * numOfTagCols);
return pTagSchema;
}
//tTagSchema *tCreateTagSchema(SSchema *pSchema, int32_t numOfTagCols) {
// if (numOfTagCols == 0 || pSchema == NULL) {
// return NULL;
// }
//
// tTagSchema *pColumnModel =
// (tTagSchema *)calloc(1, sizeof(tTagSchema) + numOfTagCols * sizeof(int32_t) + sizeof(SSchema) * numOfTagCols);
//
// pColumnModel->colOffset[0] = 0;
// pColumnModel->numOfCols = numOfTagCols;
// for (int32_t i = 1; i < numOfTagCols; ++i) {
// pColumnModel->colOffset[i] = (pColumnModel->colOffset[i - 1] + pSchema[i - 1].bytes);
// }
//
// pColumnModel->pSchema = (SSchema *)&(pColumnModel->colOffset[numOfTagCols]);
// memcpy(pColumnModel->pSchema, pSchema, sizeof(SSchema) * numOfTagCols);
// return pColumnModel;
//}
tSidSet *tSidSetCreate(struct SMeterSidExtInfo **pMeterSidExtInfo, int32_t numOfMeters, SSchema *pSchema,
int32_t numOfTags, SColIndexEx *colList, int32_t numOfCols) {
@ -269,8 +269,8 @@ tSidSet *tSidSetCreate(struct SMeterSidExtInfo **pMeterSidExtInfo, int32_t numOf
pSidSet->numOfSids = numOfMeters;
pSidSet->pSids = pMeterSidExtInfo;
pSidSet->pTagSchema = tCreateTagSchema(pSchema, numOfTags);
pSidSet->orderIdx.numOfOrderedCols = numOfCols;
pSidSet->pColumnModel = createColumnModel(pSchema, numOfTags, 1);
pSidSet->orderIdx.numOfCols = numOfCols;
/*
* in case of "group by tbname,normal_col", the normal_col is ignored
@ -282,7 +282,7 @@ tSidSet *tSidSetCreate(struct SMeterSidExtInfo **pMeterSidExtInfo, int32_t numOf
}
}
pSidSet->orderIdx.numOfOrderedCols = numOfTagCols;
pSidSet->orderIdx.numOfCols = numOfTagCols;
pSidSet->starterPos = NULL;
return pSidSet;
@ -291,19 +291,19 @@ tSidSet *tSidSetCreate(struct SMeterSidExtInfo **pMeterSidExtInfo, int32_t numOf
void tSidSetDestroy(tSidSet **pSets) {
if ((*pSets) != NULL) {
tfree((*pSets)->starterPos);
tfree((*pSets)->pTagSchema)(*pSets)->pSids = NULL;
tfree((*pSets)->pColumnModel)(*pSets)->pSids = NULL;
tfree(*pSets);
}
}
void tTagsPrints(SMeterSidExtInfo *pMeterInfo, tTagSchema *pSchema, tOrderIdx *pOrder) {
void tTagsPrints(SMeterSidExtInfo *pMeterInfo, tTagSchema *pSchema, SColumnOrderInfo *pOrder) {
if (pSchema == NULL) {
return;
}
printf("sid: %-5d tags(", pMeterInfo->sid);
for (int32_t i = 0; i < pOrder->numOfOrderedCols; ++i) {
for (int32_t i = 0; i < pOrder->numOfCols; ++i) {
int32_t colIndex = pOrder->pData[i];
// it is the tbname column
@ -312,7 +312,9 @@ void tTagsPrints(SMeterSidExtInfo *pMeterInfo, tTagSchema *pSchema, tOrderIdx *p
continue;
}
switch (pSchema->pSchema[colIndex].type) {
SSchema* s = getColumnModelSchema(pSchema, colIndex);
switch (s->type) {
case TSDB_DATA_TYPE_INT:
printf("%d, ", GET_TAG_VAL(pMeterInfo, colIndex, pSchema, int32_t));
break;
@ -336,9 +338,9 @@ void tTagsPrints(SMeterSidExtInfo *pMeterInfo, tTagSchema *pSchema, tOrderIdx *p
break;
case TSDB_DATA_TYPE_NCHAR: {
char *data = GET_TAG_VAL_POINTER(pMeterInfo, colIndex, pSchema, char);
char buffer[512] = {0};
taosUcs4ToMbs(data, pSchema->pSchema[colIndex].bytes, buffer);
taosUcs4ToMbs(data, s->bytes, buffer);
printf("%s, ", buffer);
break;
}
@ -370,16 +372,16 @@ static void UNUSED_FUNC tSidSetDisplay(tSidSet *pSets) {
printf("the %d-th subgroup: \n", i + 1);
for (int32_t j = s; j < e; ++j) {
tTagsPrints(pSets->pSids[j], pSets->pTagSchema, &pSets->orderIdx);
tTagsPrints(pSets->pSids[j], pSets->pColumnModel, &pSets->orderIdx);
}
}
}
void tSidSetSort(tSidSet *pSets) {
pTrace("number of meters in sort: %d", pSets->numOfSids);
tOrderIdx *pOrderIdx = &pSets->orderIdx;
SColumnOrderInfo *pOrderIdx = &pSets->orderIdx;
if (pOrderIdx->numOfOrderedCols == 0 || pSets->numOfSids <= 1 || pSets->pTagSchema == NULL) { // no group by tags clause
if (pOrderIdx->numOfCols == 0 || pSets->numOfSids <= 1 || pSets->pColumnModel == NULL) { // no group by tags clause
pSets->numOfSubSet = 1;
pSets->starterPos = (int32_t *)malloc(sizeof(int32_t) * (pSets->numOfSubSet + 1));
pSets->starterPos[0] = 0;
@ -390,11 +392,11 @@ void tSidSetSort(tSidSet *pSets) {
#endif
} else {
tOrderDescriptor *descriptor =
(tOrderDescriptor *)calloc(1, sizeof(tOrderDescriptor) + sizeof(int16_t) * pSets->orderIdx.numOfOrderedCols);
descriptor->pTagSchema = pSets->pTagSchema;
(tOrderDescriptor *)calloc(1, sizeof(tOrderDescriptor) + sizeof(int16_t) * pSets->orderIdx.numOfCols);
descriptor->pColumnModel = pSets->pColumnModel;
descriptor->orderIdx = pSets->orderIdx;
memcpy(descriptor->orderIdx.pData, pOrderIdx->pData, sizeof(int16_t) * pSets->orderIdx.numOfOrderedCols);
memcpy(descriptor->orderIdx.pData, pOrderIdx->pData, sizeof(int16_t) * pSets->orderIdx.numOfCols);
tQSortEx((void **)pSets->pSids, POINTER_BYTES, 0, pSets->numOfSids - 1, descriptor, meterSidComparator);
pSets->starterPos =

2
src/system/detail/src/vnodeUtil.c
View File

@ -247,7 +247,7 @@ SSqlFunctionExpr* vnodeCreateSqlFunctionExpr(SQueryMeterMsg* pQueryMsg, int32_t*
SColIndexEx* pColumnIndexExInfo = &pExprs[i].pBase.colInfo;
// tag column schema is kept in pQueryMsg->pTagSchema
// tag column schema is kept in pQueryMsg->pColumnModel
if (TSDB_COL_IS_TAG(pColumnIndexExInfo->flag)) {
if (pColumnIndexExInfo->colIdx >= pQueryMsg->numOfTagsCols) {
*code = TSDB_CODE_INVALID_QUERY_MSG;

1281
src/util/src/textbuffer.c
View File

File diff suppressed because it is too large Load Diff

100
src/util/src/tinterpolation.c
View File

@ -205,16 +205,18 @@ static char* getPos(char* data, int32_t bytes, int32_t order, int32_t capacity,
// }
}
static void setTagsValueInInterpolation(tFilePage** data, char** pTags, tColModel* pModel, int32_t order, int32_t start,
static void setTagsValueInInterpolation(tFilePage** data, char** pTags, SColumnModel* pModel, int32_t order, int32_t start,
int32_t capacity, int32_t num) {
for (int32_t j = 0, i = start; i < pModel->numOfCols; ++i, ++j) {
char* val1 = getPos(data[i]->data, pModel->pFields[i].bytes, order, capacity, num);
assignVal(val1, pTags[j], pModel->pFields[i].bytes, pModel->pFields[i].type);
SSchema* pSchema = getColumnModelSchema(pModel, i);
char* val1 = getPos(data[i]->data, pSchema->bytes, order, capacity, num);
assignVal(val1, pTags[j], pSchema->bytes, pSchema->type);
}
}
static void doInterpoResultImpl(SInterpolationInfo* pInterpoInfo, int16_t interpoType, tFilePage** data,
tColModel* pModel, int32_t* num, char** srcData, int64_t nInterval, int64_t* defaultVal,
SColumnModel* pModel, int32_t* num, char** srcData, int64_t nInterval, int64_t* defaultVal,
int64_t currentTimestamp, int32_t capacity, int32_t numOfTags, char** pTags,
bool outOfBound) {
char** prevValues = &pInterpoInfo->prevValues;
@ -234,18 +236,23 @@ static void doInterpoResultImpl(SInterpolationInfo* pInterpoInfo, int16_t interp
char* pInterpolationData = INTERPOL_IS_ASC_INTERPOL(pInterpoInfo) ? *prevValues : *nextValues;
if (pInterpolationData != NULL) {
for (int32_t i = 1; i < numOfValCols; ++i) {
char* val1 = getPos(data[i]->data, pModel->pFields[i].bytes, pInterpoInfo->order, capacity, *num);
SSchema* pSchema = getColumnModelSchema(pModel, i);
int16_t offset = getColumnModelOffset(pModel, i);
char* val1 = getPos(data[i]->data, pSchema->bytes, pInterpoInfo->order, capacity, *num);
if (isNull(pInterpolationData + pModel->colOffset[i], pModel->pFields[i].type)) {
setNull(val1, pModel->pFields[i].type, pModel->pFields[i].bytes);
if (isNull(pInterpolationData + offset, pSchema->type)) {
setNull(val1, pSchema->type, pSchema->bytes);
} else {
assignVal(val1, pInterpolationData + pModel->colOffset[i], pModel->pFields[i].bytes, pModel->pFields[i].type);
assignVal(val1, pInterpolationData + offset, pSchema->bytes, pSchema->type);
}
}
} else { /* no prev value yet, set the value for null */
for (int32_t i = 1; i < numOfValCols; ++i) {
char* val1 = getPos(data[i]->data, pModel->pFields[i].bytes, pInterpoInfo->order, capacity, *num);
setNull(val1, pModel->pFields[i].type, pModel->pFields[i].bytes);
SSchema* pSchema = getColumnModelSchema(pModel, i);
char* val1 = getPos(data[i]->data, pSchema->bytes, pInterpoInfo->order, capacity, *num);
setNull(val1, pSchema->type, pSchema->bytes);
}
}
@ -254,34 +261,41 @@ static void doInterpoResultImpl(SInterpolationInfo* pInterpoInfo, int16_t interp
// TODO : linear interpolation supports NULL value
if (*prevValues != NULL && !outOfBound) {
for (int32_t i = 1; i < numOfValCols; ++i) {
int32_t type = pModel->pFields[i].type;
char* val1 = getPos(data[i]->data, pModel->pFields[i].bytes, pInterpoInfo->order, capacity, *num);
SSchema* pSchema = getColumnModelSchema(pModel, i);
int16_t offset = getColumnModelOffset(pModel, i);
int16_t type = pSchema->type;
char* val1 = getPos(data[i]->data, pSchema->bytes, pInterpoInfo->order, capacity, *num);
if (type == TSDB_DATA_TYPE_BINARY || type == TSDB_DATA_TYPE_NCHAR || type == TSDB_DATA_TYPE_BOOL) {
setNull(val1, pModel->pFields[i].type, pModel->pFields[i].bytes);
setNull(val1, type, pSchema->bytes);
continue;
}
point1 = (SPoint){.key = *(TSKEY*)(*prevValues), .val = *prevValues + pModel->colOffset[i]};
point2 = (SPoint){.key = currentTimestamp, .val = srcData[i] + pInterpoInfo->rowIdx * pModel->pFields[i].bytes};
point1 = (SPoint){.key = *(TSKEY*)(*prevValues), .val = *prevValues + offset};
point2 = (SPoint){.key = currentTimestamp, .val = srcData[i] + pInterpoInfo->rowIdx * pSchema->bytes};
point = (SPoint){.key = pInterpoInfo->startTimestamp, .val = val1};
taosDoLinearInterpolation(pModel->pFields[i].type, &point1, &point2, &point);
taosDoLinearInterpolation(type, &point1, &point2, &point);
}
setTagsValueInInterpolation(data, pTags, pModel, pInterpoInfo->order, numOfValCols, capacity, *num);
} else {
for (int32_t i = 1; i < numOfValCols; ++i) {
char* val1 = getPos(data[i]->data, pModel->pFields[i].bytes, pInterpoInfo->order, capacity, *num);
setNull(val1, pModel->pFields[i].type, pModel->pFields[i].bytes);
SSchema* pSchema = getColumnModelSchema(pModel, i);
char* val1 = getPos(data[i]->data, pSchema->bytes, pInterpoInfo->order, capacity, *num);
setNull(val1, pSchema->type, pSchema->bytes);
}
setTagsValueInInterpolation(data, pTags, pModel, pInterpoInfo->order, numOfValCols, capacity, *num);
}
} else { /* default value interpolation */
for (int32_t i = 1; i < numOfValCols; ++i) {
char* val1 = getPos(data[i]->data, pModel->pFields[i].bytes, pInterpoInfo->order, capacity, *num);
assignVal(val1, (char*)&defaultVal[i], pModel->pFields[i].bytes, pModel->pFields[i].type);
SSchema* pSchema = getColumnModelSchema(pModel, i);
char* val1 = getPos(data[i]->data, pSchema->bytes, pInterpoInfo->order, capacity, *num);
assignVal(val1, (char*)&defaultVal[i], pSchema->bytes, pSchema->type);
}
setTagsValueInInterpolation(data, pTags, pModel, pInterpoInfo->order, numOfValCols, capacity, *num);
@ -295,7 +309,7 @@ static void doInterpoResultImpl(SInterpolationInfo* pInterpoInfo, int16_t interp
int32_t taosDoInterpoResult(SInterpolationInfo* pInterpoInfo, int16_t interpoType, tFilePage** data,
int32_t numOfRawDataInRows, int32_t outputRows, int64_t nInterval,
const int64_t* pPrimaryKeyArray, tColModel* pModel, char** srcData, int64_t* defaultVal,
const int64_t* pPrimaryKeyArray, SColumnModel* pModel, char** srcData, int64_t* defaultVal,
const int32_t* functionIDs, int32_t bufSize) {
int32_t num = 0;
pInterpoInfo->numOfCurrentInterpo = 0;
@ -328,17 +342,21 @@ int32_t taosDoInterpoResult(SInterpolationInfo* pInterpoInfo, int16_t interpoTyp
(pInterpoInfo->startTimestamp > currentTimestamp && !INTERPOL_IS_ASC_INTERPOL(pInterpoInfo))) {
/* set the next value for interpolation */
if (*nextValues == NULL) {
*nextValues =
calloc(1, pModel->colOffset[pModel->numOfCols - 1] + pModel->pFields[pModel->numOfCols - 1].bytes);
*nextValues = calloc(1, pModel->rowSize);
for (int i = 1; i < pModel->numOfCols; i++) {
setNull(*nextValues + pModel->colOffset[i], pModel->pFields[i].type, pModel->pFields[i].bytes);
int16_t offset = getColumnModelOffset(pModel, i);
SSchema* pSchema = getColumnModelSchema(pModel, i);
setNull(*nextValues + offset, pSchema->type, pSchema->bytes);
}
}
int32_t offset = pInterpoInfo->rowIdx;
for (int32_t tlen = 0, i = 0; i < pModel->numOfCols - numOfTags; ++i) {
memcpy(*nextValues + tlen, srcData[i] + offset * pModel->pFields[i].bytes, pModel->pFields[i].bytes);
tlen += pModel->pFields[i].bytes;
SSchema* pSchema = getColumnModelSchema(pModel, i);
memcpy(*nextValues + tlen, srcData[i] + offset * pSchema->bytes, pSchema->bytes);
tlen += pSchema->bytes;
}
}
@ -358,37 +376,41 @@ int32_t taosDoInterpoResult(SInterpolationInfo* pInterpoInfo, int16_t interpoTyp
if (pInterpoInfo->startTimestamp == currentTimestamp) {
if (*prevValues == NULL) {
*prevValues =
calloc(1, pModel->colOffset[pModel->numOfCols - 1] + pModel->pFields[pModel->numOfCols - 1].bytes);
*prevValues = calloc(1, pModel->rowSize);
for (int i = 1; i < pModel->numOfCols; i++) {
setNull(*prevValues + pModel->colOffset[i], pModel->pFields[i].type, pModel->pFields[i].bytes);
int16_t offset = getColumnModelOffset(pModel, i);
SSchema* pSchema = getColumnModelSchema(pModel, i);
setNull(*prevValues + offset, pSchema->type, pSchema->bytes);
}
}
// assign rows to dst buffer
int32_t i = 0;
for (int32_t tlen = 0; i < pModel->numOfCols - numOfTags; ++i) {
char* val1 = getPos(data[i]->data, pModel->pFields[i].bytes, pInterpoInfo->order, bufSize, num);
int16_t offset = getColumnModelOffset(pModel, i);
SSchema* pSchema = getColumnModelSchema(pModel, i);
char* val1 = getPos(data[i]->data, pSchema->bytes, pInterpoInfo->order, bufSize, num);
if (i == 0 ||
(functionIDs[i] != TSDB_FUNC_COUNT &&
!isNull(srcData[i] + pInterpoInfo->rowIdx * pModel->pFields[i].bytes, pModel->pFields[i].type)) ||
!isNull(srcData[i] + pInterpoInfo->rowIdx * pSchema->bytes, pSchema->type)) ||
(functionIDs[i] == TSDB_FUNC_COUNT &&
*(int64_t*)(srcData[i] + pInterpoInfo->rowIdx * pModel->pFields[i].bytes) != 0)) {
assignVal(val1, srcData[i] + pInterpoInfo->rowIdx * pModel->pFields[i].bytes, pModel->pFields[i].bytes,
pModel->pFields[i].type);
memcpy(*prevValues + tlen, srcData[i] + pInterpoInfo->rowIdx * pModel->pFields[i].bytes,
pModel->pFields[i].bytes);
*(int64_t*)(srcData[i] + pInterpoInfo->rowIdx * pSchema->bytes) != 0)) {
assignVal(val1, srcData[i] + pInterpoInfo->rowIdx * pSchema->bytes, pSchema->bytes, pSchema->type);
memcpy(*prevValues + tlen, srcData[i] + pInterpoInfo->rowIdx * pSchema->bytes, pSchema->bytes);
} else { // i > 0 and isNULL, do interpolation
if (interpoType == TSDB_INTERPO_PREV) {
assignVal(val1, *prevValues + pModel->colOffset[i], pModel->pFields[i].bytes, pModel->pFields[i].type);
assignVal(val1, *prevValues + offset, pSchema->bytes, pSchema->type);
} else if (interpoType == TSDB_INTERPO_LINEAR) {
// TODO:
} else {
assignVal(val1, (char*)&defaultVal[i], pModel->pFields[i].bytes, pModel->pFields[i].type);
assignVal(val1, (char*)&defaultVal[i], pSchema->bytes, pSchema->type);
}
}
tlen += pModel->pFields[i].bytes;
tlen += pSchema->bytes;
}
/* set the tag value for final result */

976
src/util/src/tpercentile.c Normal file
View File

@ -0,0 +1,976 @@
/*
* Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
*
* This program is free software: you can use, redistribute, and/or modify
* it under the terms of the GNU Affero General Public License, version 3
* or later ("AGPL"), as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "os.h"
#include "taosmsg.h"
#include "tsdb.h"
#include "tlog.h"
#include "ttypes.h"
#include "tpercentile.h"
tExtMemBuffer *releaseBucketsExceptFor(tMemBucket *pMemBucket, int16_t segIdx, int16_t slotIdx) {
tExtMemBuffer *pBuffer = NULL;
for (int32_t i = 0; i < pMemBucket->numOfSegs; ++i) {
tMemBucketSegment *pSeg = &pMemBucket->pSegs[i];
for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
if (i == segIdx && j == slotIdx) {
pBuffer = pSeg->pBuffer[j];
} else {
if (pSeg->pBuffer && pSeg->pBuffer[j]) {
pSeg->pBuffer[j] = destoryExtMemBuffer(pSeg->pBuffer[j]);
}
}
}
}
return pBuffer;
}
static tFilePage *loadIntoBucketFromDisk(tMemBucket *pMemBucket, int32_t segIdx, int32_t slotIdx,
tOrderDescriptor *pDesc) {
// release all data in other slots
tExtMemBuffer *pMemBuffer = pMemBucket->pSegs[segIdx].pBuffer[slotIdx];
tFilePage * buffer = (tFilePage *)calloc(1, pMemBuffer->nElemSize * pMemBuffer->numOfTotalElems + sizeof(tFilePage));
int32_t oldCapacity = pDesc->pColumnModel->capacity;
pDesc->pColumnModel->capacity = pMemBuffer->numOfTotalElems;
if (!tExtMemBufferIsAllDataInMem(pMemBuffer)) {
pMemBuffer = releaseBucketsExceptFor(pMemBucket, segIdx, slotIdx);
assert(pMemBuffer->numOfTotalElems > 0);
// load data in disk to memory
tFilePage *pPage = (tFilePage *)calloc(1, pMemBuffer->pageSize);
for (int32_t i = 0; i < pMemBuffer->fileMeta.flushoutData.nLength; ++i) {
tFlushoutInfo *pFlushInfo = &pMemBuffer->fileMeta.flushoutData.pFlushoutInfo[i];
int32_t ret = fseek(pMemBuffer->file, pFlushInfo->startPageId * pMemBuffer->pageSize, SEEK_SET);
UNUSED(ret);
for (uint32_t j = 0; j < pFlushInfo->numOfPages; ++j) {
ret = fread(pPage, pMemBuffer->pageSize, 1, pMemBuffer->file);
UNUSED(ret);
assert(pPage->numOfElems > 0);
tColModelAppend(pDesc->pColumnModel, buffer, pPage->data, 0, pPage->numOfElems, pPage->numOfElems);
printf("id: %d count: %" PRIu64 "\n", j, buffer->numOfElems);
}
}
tfree(pPage);
assert(buffer->numOfElems == pMemBuffer->fileMeta.numOfElemsInFile);
}
// load data in pMemBuffer to buffer
tFilePagesItem *pListItem = pMemBuffer->pHead;
while (pListItem != NULL) {
tColModelAppend(pDesc->pColumnModel, buffer, pListItem->item.data, 0, pListItem->item.numOfElems,
pListItem->item.numOfElems);
pListItem = pListItem->pNext;
}
tColDataQSort(pDesc, buffer->numOfElems, 0, buffer->numOfElems - 1, buffer->data, TSQL_SO_ASC);
pDesc->pColumnModel->capacity = oldCapacity; // restore value
return buffer;
}
double findOnlyResult(tMemBucket *pMemBucket) {
assert(pMemBucket->numOfElems == 1);
for (int32_t i = 0; i < pMemBucket->numOfSegs; ++i) {
tMemBucketSegment *pSeg = &pMemBucket->pSegs[i];
if (pSeg->pBuffer) {
for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
tExtMemBuffer *pBuffer = pSeg->pBuffer[j];
if (pBuffer) {
assert(pBuffer->numOfTotalElems == 1);
tFilePage *pPage = &pBuffer->pHead->item;
if (pBuffer->numOfElemsInBuffer == 1) {
switch (pMemBucket->dataType) {
case TSDB_DATA_TYPE_INT:
return *(int32_t *)pPage->data;
case TSDB_DATA_TYPE_SMALLINT:
return *(int16_t *)pPage->data;
case TSDB_DATA_TYPE_TINYINT:
return *(int8_t *)pPage->data;
case TSDB_DATA_TYPE_BIGINT:
return (double)(*(int64_t *)pPage->data);
case TSDB_DATA_TYPE_DOUBLE: {
double dv = GET_DOUBLE_VAL(pPage->data);
//return *(double *)pPage->data;
return dv;
}
case TSDB_DATA_TYPE_FLOAT: {
float fv = GET_FLOAT_VAL(pPage->data);
//return *(float *)pPage->data;
return fv;
}
default:
return 0;
}
}
}
}
}
}
return 0;
}
void tBucketBigIntHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx) {
int64_t v = *(int64_t *)value;
if (pBucket->nRange.i64MaxVal == INT64_MIN) {
if (v >= 0) {
*segIdx = ((v >> (64 - 9)) >> 6) + 8;
*slotIdx = (v >> (64 - 9)) & 0x3F;
} else { // v<0
*segIdx = ((-v) >> (64 - 9)) >> 6;
*slotIdx = ((-v) >> (64 - 9)) & 0x3F;
*segIdx = 7 - (*segIdx);
}
} else {
// todo hash for bigint and float and double
int64_t span = pBucket->nRange.i64MaxVal - pBucket->nRange.i64MinVal;
if (span < pBucket->nTotalSlots) {
int32_t delta = (int32_t)(v - pBucket->nRange.i64MinVal);
*segIdx = delta / pBucket->nSlotsOfSeg;
*slotIdx = delta % pBucket->nSlotsOfSeg;
} else {
double x = (double)span / pBucket->nTotalSlots;
double posx = (v - pBucket->nRange.i64MinVal) / x;
if (v == pBucket->nRange.i64MaxVal) {
posx -= 1;
}
*segIdx = ((int32_t)posx) / pBucket->nSlotsOfSeg;
*slotIdx = ((int32_t)posx) % pBucket->nSlotsOfSeg;
}
}
}
// todo refactor to more generic
void tBucketIntHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx) {
int32_t v = *(int32_t *)value;
if (pBucket->nRange.iMaxVal == INT32_MIN) {
/*
* taking negative integer into consideration,
* there is only half of pBucket->segs available for non-negative integer
*/
// int32_t numOfSlots = pBucket->nTotalSlots>>1;
// int32_t bits = bitsOfNumber(numOfSlots)-1;
if (v >= 0) {
*segIdx = ((v >> (32 - 9)) >> 6) + 8;
*slotIdx = (v >> (32 - 9)) & 0x3F;
} else { // v<0
*segIdx = ((-v) >> (32 - 9)) >> 6;
*slotIdx = ((-v) >> (32 - 9)) & 0x3F;
*segIdx = 7 - (*segIdx);
}
} else {
// divide a range of [iMinVal, iMaxVal] into 1024 buckets
int32_t span = pBucket->nRange.iMaxVal - pBucket->nRange.iMinVal;
if (span < pBucket->nTotalSlots) {
int32_t delta = v - pBucket->nRange.iMinVal;
*segIdx = delta / pBucket->nSlotsOfSeg;
*slotIdx = delta % pBucket->nSlotsOfSeg;
} else {
double x = (double)span / pBucket->nTotalSlots;
double posx = (v - pBucket->nRange.iMinVal) / x;
if (v == pBucket->nRange.iMaxVal) {
posx -= 1;
}
*segIdx = ((int32_t)posx) / pBucket->nSlotsOfSeg;
*slotIdx = ((int32_t)posx) % pBucket->nSlotsOfSeg;
}
}
}
void tBucketDoubleHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx) {
// double v = *(double *)value;
double v = GET_DOUBLE_VAL(value);
if (pBucket->nRange.dMinVal == DBL_MAX) {
/*
* taking negative integer into consideration,
* there is only half of pBucket->segs available for non-negative integer
*/
double x = DBL_MAX / (pBucket->nTotalSlots >> 1);
double posx = (v + DBL_MAX) / x;
*segIdx = ((int32_t)posx) / pBucket->nSlotsOfSeg;
*slotIdx = ((int32_t)posx) % pBucket->nSlotsOfSeg;
} else {
// divide a range of [dMinVal, dMaxVal] into 1024 buckets
double span = pBucket->nRange.dMaxVal - pBucket->nRange.dMinVal;
if (span < pBucket->nTotalSlots) {
int32_t delta = (int32_t)(v - pBucket->nRange.dMinVal);
*segIdx = delta / pBucket->nSlotsOfSeg;
*slotIdx = delta % pBucket->nSlotsOfSeg;
} else {
double x = span / pBucket->nTotalSlots;
double posx = (v - pBucket->nRange.dMinVal) / x;
if (v == pBucket->nRange.dMaxVal) {
posx -= 1;
}
*segIdx = ((int32_t)posx) / pBucket->nSlotsOfSeg;
*slotIdx = ((int32_t)posx) % pBucket->nSlotsOfSeg;
}
if (*segIdx < 0 || *segIdx > 16 || *slotIdx < 0 || *slotIdx > 64) {
pError("error in hash process. segment is: %d, slot id is: %d\n", *segIdx, *slotIdx);
}
}
}
tMemBucket *tMemBucketCreate(int32_t totalSlots, int32_t nBufferSize, int16_t nElemSize, int16_t dataType,
tOrderDescriptor *pDesc) {
tMemBucket *pBucket = (tMemBucket *)malloc(sizeof(tMemBucket));
pBucket->nTotalSlots = totalSlots;
pBucket->nSlotsOfSeg = 1 << 6; // 64 Segments, 16 slots each seg.
pBucket->dataType = dataType;
pBucket->nElemSize = nElemSize;
pBucket->pageSize = DEFAULT_PAGE_SIZE;
pBucket->numOfElems = 0;
pBucket->numOfSegs = pBucket->nTotalSlots / pBucket->nSlotsOfSeg;
pBucket->nTotalBufferSize = nBufferSize;
pBucket->maxElemsCapacity = pBucket->nTotalBufferSize / pBucket->nElemSize;
pBucket->numOfTotalPages = pBucket->nTotalBufferSize / pBucket->pageSize;
pBucket->numOfAvailPages = pBucket->numOfTotalPages;
pBucket->pOrderDesc = pDesc;
switch (pBucket->dataType) {
case TSDB_DATA_TYPE_INT:
case TSDB_DATA_TYPE_SMALLINT:
case TSDB_DATA_TYPE_TINYINT: {
pBucket->nRange.iMinVal = INT32_MAX;
pBucket->nRange.iMaxVal = INT32_MIN;
pBucket->HashFunc = tBucketIntHash;
break;
};
case TSDB_DATA_TYPE_DOUBLE:
case TSDB_DATA_TYPE_FLOAT: {
pBucket->nRange.dMinVal = DBL_MAX;
pBucket->nRange.dMaxVal = -DBL_MAX;
pBucket->HashFunc = tBucketDoubleHash;
break;
};
case TSDB_DATA_TYPE_BIGINT: {
pBucket->nRange.i64MinVal = INT64_MAX;
pBucket->nRange.i64MaxVal = INT64_MIN;
pBucket->HashFunc = tBucketBigIntHash;
break;
};
default: {
pError("MemBucket:%p,not support data type %d,failed", *pBucket, pBucket->dataType);
tfree(pBucket);
return NULL;
}
}
int32_t numOfCols = pDesc->pColumnModel->numOfCols;
if (numOfCols != 1) {
pError("MemBucket:%p,only consecutive data is allowed,invalid numOfCols:%d", pBucket, numOfCols);
tfree(pBucket);
return NULL;
}
SSchema* pSchema = getColumnModelSchema(pDesc->pColumnModel, 0);
if (pSchema->type != dataType) {
pError("MemBucket:%p,data type is not consistent,%d in schema, %d in param", pBucket, pSchema->type, dataType);
tfree(pBucket);
return NULL;
}
if (pBucket->numOfTotalPages < pBucket->nTotalSlots) {
pWarn("MemBucket:%p,total buffer pages %d are not enough for all slots", pBucket, pBucket->numOfTotalPages);
}
pBucket->pSegs = (tMemBucketSegment *)malloc(pBucket->numOfSegs * sizeof(tMemBucketSegment));
for (int32_t i = 0; i < pBucket->numOfSegs; ++i) {
pBucket->pSegs[i].numOfSlots = pBucket->nSlotsOfSeg;
pBucket->pSegs[i].pBuffer = NULL;
pBucket->pSegs[i].pBoundingEntries = NULL;
}
pTrace("MemBucket:%p,created,buffer size:%d,elem size:%d", pBucket, pBucket->numOfTotalPages * DEFAULT_PAGE_SIZE,
pBucket->nElemSize);
return pBucket;
}
void tMemBucketDestroy(tMemBucket *pBucket) {
if (pBucket == NULL) {
return;
}
if (pBucket->pSegs) {
for (int32_t i = 0; i < pBucket->numOfSegs; ++i) {
tMemBucketSegment *pSeg = &(pBucket->pSegs[i]);
tfree(pSeg->pBoundingEntries);
if (pSeg->pBuffer == NULL || pSeg->numOfSlots == 0) {
continue;
}
for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
if (pSeg->pBuffer[j] != NULL) {
pSeg->pBuffer[j] = destoryExtMemBuffer(pSeg->pBuffer[j]);
}
}
tfree(pSeg->pBuffer);
}
}
tfree(pBucket->pSegs);
tfree(pBucket);
}
/*
* find the slots which accounts for largest proportion of total in-memory buffer
*/
static void tBucketGetMaxMemSlot(tMemBucket *pBucket, int16_t *segIdx, int16_t *slotIdx) {
*segIdx = -1;
*slotIdx = -1;
int32_t val = 0;
for (int32_t k = 0; k < pBucket->numOfSegs; ++k) {
tMemBucketSegment *pSeg = &pBucket->pSegs[k];
for (int32_t i = 0; i < pSeg->numOfSlots; ++i) {
if (pSeg->pBuffer == NULL || pSeg->pBuffer[i] == NULL) {
continue;
}
if (val < pSeg->pBuffer[i]->numOfInMemPages) {
val = pSeg->pBuffer[i]->numOfInMemPages;
*segIdx = k;
*slotIdx = i;
}
}
}
}
static void resetBoundingBox(tMemBucketSegment *pSeg, int32_t type) {
switch (type) {
case TSDB_DATA_TYPE_BIGINT: {
for (int32_t i = 0; i < pSeg->numOfSlots; ++i) {
pSeg->pBoundingEntries[i].i64MaxVal = INT64_MIN;
pSeg->pBoundingEntries[i].i64MinVal = INT64_MAX;
}
break;
};
case TSDB_DATA_TYPE_INT:
case TSDB_DATA_TYPE_SMALLINT:
case TSDB_DATA_TYPE_TINYINT: {
for (int32_t i = 0; i < pSeg->numOfSlots; ++i) {
pSeg->pBoundingEntries[i].iMaxVal = INT32_MIN;
pSeg->pBoundingEntries[i].iMinVal = INT32_MAX;
}
break;
};
case TSDB_DATA_TYPE_DOUBLE:
case TSDB_DATA_TYPE_FLOAT: {
for (int32_t i = 0; i < pSeg->numOfSlots; ++i) {
pSeg->pBoundingEntries[i].dMaxVal = -DBL_MAX;
pSeg->pBoundingEntries[i].dMinVal = DBL_MAX;
}
break;
}
}
}
void tMemBucketUpdateBoundingBox(MinMaxEntry *r, char *data, int32_t dataType) {
switch (dataType) {
case TSDB_DATA_TYPE_INT: {
int32_t val = *(int32_t *)data;
if (r->iMinVal > val) {
r->iMinVal = val;
}
if (r->iMaxVal < val) {
r->iMaxVal = val;
}
break;
};
case TSDB_DATA_TYPE_BIGINT: {
int64_t val = *(int64_t *)data;
if (r->i64MinVal > val) {
r->i64MinVal = val;
}
if (r->i64MaxVal < val) {
r->i64MaxVal = val;
}
break;
};
case TSDB_DATA_TYPE_SMALLINT: {
int32_t val = *(int16_t *)data;
if (r->iMinVal > val) {
r->iMinVal = val;
}
if (r->iMaxVal < val) {
r->iMaxVal = val;
}
break;
};
case TSDB_DATA_TYPE_TINYINT: {
int32_t val = *(int8_t *)data;
if (r->iMinVal > val) {
r->iMinVal = val;
}
if (r->iMaxVal < val) {
r->iMaxVal = val;
}
break;
};
case TSDB_DATA_TYPE_DOUBLE: {
// double val = *(double *)data;
double val = GET_DOUBLE_VAL(data);
if (r->dMinVal > val) {
r->dMinVal = val;
}
if (r->dMaxVal < val) {
r->dMaxVal = val;
}
break;
};
case TSDB_DATA_TYPE_FLOAT: {
// double val = *(float *)data;
double val = GET_FLOAT_VAL(data);
if (r->dMinVal > val) {
r->dMinVal = val;
}
if (r->dMaxVal < val) {
r->dMaxVal = val;
}
break;
};
default: { assert(false); }
}
}
/*
* in memory bucket, we only accept the simple data consecutive put in a row/column
* no column-model in this case.
*/
void tMemBucketPut(tMemBucket *pBucket, void *data, int32_t numOfRows) {
pBucket->numOfElems += numOfRows;
int16_t segIdx = 0, slotIdx = 0;
for (int32_t i = 0; i < numOfRows; ++i) {
char *d = (char *)data + i * tDataTypeDesc[pBucket->dataType].nSize;
switch (pBucket->dataType) {
case TSDB_DATA_TYPE_SMALLINT: {
int32_t val = *(int16_t *)d;
(pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
break;
}
case TSDB_DATA_TYPE_TINYINT: {
int32_t val = *(int8_t *)d;
(pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
break;
}
case TSDB_DATA_TYPE_INT: {
int32_t val = *(int32_t *)d;
(pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
break;
}
case TSDB_DATA_TYPE_BIGINT: {
int64_t val = *(int64_t *)d;
(pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
break;
}
case TSDB_DATA_TYPE_DOUBLE: {
// double val = *(double *)d;
double val = GET_DOUBLE_VAL(d);
(pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
break;
}
case TSDB_DATA_TYPE_FLOAT: {
// double val = *(float *)d;
double val = GET_FLOAT_VAL(d);
(pBucket->HashFunc)(pBucket, &val, &segIdx, &slotIdx);
break;
}
}
tMemBucketSegment *pSeg = &pBucket->pSegs[segIdx];
if (pSeg->pBoundingEntries == NULL) {
pSeg->pBoundingEntries = (MinMaxEntry *)malloc(sizeof(MinMaxEntry) * pBucket->nSlotsOfSeg);
resetBoundingBox(pSeg, pBucket->dataType);
}
if (pSeg->pBuffer == NULL) {
pSeg->pBuffer = (tExtMemBuffer **)calloc(pBucket->nSlotsOfSeg, sizeof(void *));
}
if (pSeg->pBuffer[slotIdx] == NULL) {
pSeg->pBuffer[slotIdx] = createExtMemBuffer(pBucket->numOfTotalPages * pBucket->pageSize, pBucket->nElemSize,
pBucket->pOrderDesc->pColumnModel);
pSeg->pBuffer[slotIdx]->flushModel = SINGLE_APPEND_MODEL;
pBucket->pOrderDesc->pColumnModel->capacity = pSeg->pBuffer[slotIdx]->numOfElemsPerPage;
}
tMemBucketUpdateBoundingBox(&pSeg->pBoundingEntries[slotIdx], d, pBucket->dataType);
// ensure available memory pages to allocate
int16_t cseg = 0, cslot = 0;
if (pBucket->numOfAvailPages == 0) {
pTrace("MemBucket:%p,max avail size:%d, no avail memory pages,", pBucket, pBucket->numOfTotalPages);
tBucketGetMaxMemSlot(pBucket, &cseg, &cslot);
if (cseg == -1 || cslot == -1) {
pError("MemBucket:%p,failed to find appropriated avail buffer", pBucket);
return;
}
if (cseg != segIdx || cslot != slotIdx) {
pBucket->numOfAvailPages += pBucket->pSegs[cseg].pBuffer[cslot]->numOfInMemPages;
int32_t avail = pBucket->pSegs[cseg].pBuffer[cslot]->numOfInMemPages;
UNUSED(avail);
tExtMemBufferFlush(pBucket->pSegs[cseg].pBuffer[cslot]);
pTrace("MemBucket:%p,seg:%d,slot:%d flushed to disk,new avail pages:%d", pBucket, cseg, cslot,
pBucket->numOfAvailPages);
} else {
pTrace("MemBucket:%p,failed to choose slot to flush to disk seg:%d,slot:%d", pBucket, cseg, cslot);
}
}
int16_t consumedPgs = pSeg->pBuffer[slotIdx]->numOfInMemPages;
int16_t newPgs = tExtMemBufferPut(pSeg->pBuffer[slotIdx], d, 1);
/*
* trigger 1. page re-allocation, to reduce the available pages
* 2. page flushout, to increase the available pages
*/
pBucket->numOfAvailPages += (consumedPgs - newPgs);
}
}
void releaseBucket(tMemBucket *pMemBucket, int32_t segIdx, int32_t slotIdx) {
if (segIdx < 0 || segIdx > pMemBucket->numOfSegs || slotIdx < 0) {
return;
}
tMemBucketSegment *pSeg = &pMemBucket->pSegs[segIdx];
if (slotIdx < 0 || slotIdx >= pSeg->numOfSlots || pSeg->pBuffer[slotIdx] == NULL) {
return;
}
pSeg->pBuffer[slotIdx] = destoryExtMemBuffer(pSeg->pBuffer[slotIdx]);
}
////////////////////////////////////////////////////////////////////////////////////////////
static void findMaxMinValue(tMemBucket *pMemBucket, double *maxVal, double *minVal) {
*minVal = DBL_MAX;
*maxVal = -DBL_MAX;
for (int32_t i = 0; i < pMemBucket->numOfSegs; ++i) {
tMemBucketSegment *pSeg = &pMemBucket->pSegs[i];
if (pSeg->pBuffer == NULL) {
continue;
}
switch (pMemBucket->dataType) {
case TSDB_DATA_TYPE_INT:
case TSDB_DATA_TYPE_SMALLINT:
case TSDB_DATA_TYPE_TINYINT: {
for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
double minv = pSeg->pBoundingEntries[j].iMinVal;
double maxv = pSeg->pBoundingEntries[j].iMaxVal;
if (*minVal > minv) {
*minVal = minv;
}
if (*maxVal < maxv) {
*maxVal = maxv;
}
}
break;
}
case TSDB_DATA_TYPE_DOUBLE:
case TSDB_DATA_TYPE_FLOAT: {
for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
double minv = pSeg->pBoundingEntries[j].dMinVal;
double maxv = pSeg->pBoundingEntries[j].dMaxVal;
if (*minVal > minv) {
*minVal = minv;
}
if (*maxVal < maxv) {
*maxVal = maxv;
}
}
break;
}
case TSDB_DATA_TYPE_BIGINT: {
for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
double minv = (double)pSeg->pBoundingEntries[j].i64MinVal;
double maxv = (double)pSeg->pBoundingEntries[j].i64MaxVal;
if (*minVal > minv) {
*minVal = minv;
}
if (*maxVal < maxv) {
*maxVal = maxv;
}
}
break;
}
}
}
}
static MinMaxEntry getMinMaxEntryOfNearestSlotInNextSegment(tMemBucket *pMemBucket, int32_t segIdx) {
int32_t i = segIdx + 1;
while (i < pMemBucket->numOfSegs && pMemBucket->pSegs[i].numOfSlots == 0) ++i;
tMemBucketSegment *pSeg = &pMemBucket->pSegs[i];
assert(pMemBucket->numOfSegs > i && pMemBucket->pSegs[i].pBuffer != NULL);
i = 0;
while (i < pMemBucket->nSlotsOfSeg && pSeg->pBuffer[i] == NULL) ++i;
assert(i < pMemBucket->nSlotsOfSeg);
return pSeg->pBoundingEntries[i];
}
/*
*
* now, we need to find the minimum value of the next slot for
* interpolating the percentile value
* j is the last slot of current segment, we need to get the first
* slot of the next segment.
*/
static MinMaxEntry getMinMaxEntryOfNextSlotWithData(tMemBucket *pMemBucket, int32_t segIdx, int32_t slotIdx) {
tMemBucketSegment *pSeg = &pMemBucket->pSegs[segIdx];
MinMaxEntry next;
if (slotIdx == pSeg->numOfSlots - 1) { // find next segment with data
return getMinMaxEntryOfNearestSlotInNextSegment(pMemBucket, segIdx);
} else {
int32_t j = slotIdx + 1;
for (; j < pMemBucket->nSlotsOfSeg && pMemBucket->pSegs[segIdx].pBuffer[j] == 0; ++j) {
};
if (j == pMemBucket->nSlotsOfSeg) { // current slot has no available
// slot,try next segment
return getMinMaxEntryOfNearestSlotInNextSegment(pMemBucket, segIdx);
} else {
next = pSeg->pBoundingEntries[slotIdx + 1];
assert(pSeg->pBuffer[slotIdx + 1] != NULL);
}
}
return next;
}
bool isIdenticalData(tMemBucket *pMemBucket, int32_t segIdx, int32_t slotIdx);
char *getFirstElemOfMemBuffer(tMemBucketSegment *pSeg, int32_t slotIdx, tFilePage *pPage);
double getPercentileImpl(tMemBucket *pMemBucket, int32_t count, double fraction) {
int32_t num = 0;
for (int32_t i = 0; i < pMemBucket->numOfSegs; ++i) {
tMemBucketSegment *pSeg = &pMemBucket->pSegs[i];
for (int32_t j = 0; j < pSeg->numOfSlots; ++j) {
if (pSeg->pBuffer == NULL || pSeg->pBuffer[j] == NULL) {
continue;
}
// required value in current slot
if (num < (count + 1) && num + pSeg->pBuffer[j]->numOfTotalElems >= (count + 1)) {
if (pSeg->pBuffer[j]->numOfTotalElems + num == (count + 1)) {
/*
* now, we need to find the minimum value of the next slot for interpolating the percentile value
* j is the last slot of current segment, we need to get the first slot of the next segment.
*
*/
MinMaxEntry next = getMinMaxEntryOfNextSlotWithData(pMemBucket, i, j);
double maxOfThisSlot = 0;
double minOfNextSlot = 0;
switch (pMemBucket->dataType) {
case TSDB_DATA_TYPE_INT:
case TSDB_DATA_TYPE_SMALLINT:
case TSDB_DATA_TYPE_TINYINT: {
maxOfThisSlot = pSeg->pBoundingEntries[j].iMaxVal;
minOfNextSlot = next.iMinVal;
break;
};
case TSDB_DATA_TYPE_FLOAT:
case TSDB_DATA_TYPE_DOUBLE: {
maxOfThisSlot = pSeg->pBoundingEntries[j].dMaxVal;
minOfNextSlot = next.dMinVal;
break;
};
case TSDB_DATA_TYPE_BIGINT: {
maxOfThisSlot = (double)pSeg->pBoundingEntries[j].i64MaxVal;
minOfNextSlot = (double)next.i64MinVal;
break;
}
};
assert(minOfNextSlot > maxOfThisSlot);
double val = (1 - fraction) * maxOfThisSlot + fraction * minOfNextSlot;
return val;
}
if (pSeg->pBuffer[j]->numOfTotalElems <= pMemBucket->maxElemsCapacity) {
// data in buffer and file are merged together to be processed.
tFilePage *buffer = loadIntoBucketFromDisk(pMemBucket, i, j, pMemBucket->pOrderDesc);
int32_t currentIdx = count - num;
char * thisVal = buffer->data + pMemBucket->nElemSize * currentIdx;
char * nextVal = thisVal + pMemBucket->nElemSize;
double td, nd;
switch (pMemBucket->dataType) {
case TSDB_DATA_TYPE_SMALLINT: {
td = *(int16_t *)thisVal;
nd = *(int16_t *)nextVal;
break;
}
case TSDB_DATA_TYPE_TINYINT: {
td = *(int8_t *)thisVal;
nd = *(int8_t *)nextVal;
break;
}
case TSDB_DATA_TYPE_INT: {
td = *(int32_t *)thisVal;
nd = *(int32_t *)nextVal;
break;
};
case TSDB_DATA_TYPE_FLOAT: {
// td = *(float *)thisVal;
// nd = *(float *)nextVal;
td = GET_FLOAT_VAL(thisVal);
nd = GET_FLOAT_VAL(nextVal);
break;
}
case TSDB_DATA_TYPE_DOUBLE: {
// td = *(double *)thisVal;
td = GET_DOUBLE_VAL(thisVal);
// nd = *(double *)nextVal;
nd = GET_DOUBLE_VAL(nextVal);
break;
}
case TSDB_DATA_TYPE_BIGINT: {
td = (double)*(int64_t *)thisVal;
nd = (double)*(int64_t *)nextVal;
break;
}
}
double val = (1 - fraction) * td + fraction * nd;
tfree(buffer);
return val;
} else { // incur a second round bucket split
if (isIdenticalData(pMemBucket, i, j)) {
tExtMemBuffer *pMemBuffer = pSeg->pBuffer[j];
tFilePage *pPage = (tFilePage *)malloc(pMemBuffer->pageSize);
char *thisVal = getFirstElemOfMemBuffer(pSeg, j, pPage);
double finalResult = 0.0;
switch (pMemBucket->dataType) {
case TSDB_DATA_TYPE_SMALLINT: {
finalResult = *(int16_t *)thisVal;
break;
}
case TSDB_DATA_TYPE_TINYINT: {
finalResult = *(int8_t *)thisVal;
break;
}
case TSDB_DATA_TYPE_INT: {
finalResult = *(int32_t *)thisVal;
break;
};
case TSDB_DATA_TYPE_FLOAT: {
// finalResult = *(float *)thisVal;
finalResult = GET_FLOAT_VAL(thisVal);
break;
}
case TSDB_DATA_TYPE_DOUBLE: {
// finalResult = *(double *)thisVal;
finalResult = GET_DOUBLE_VAL(thisVal);
break;
}
case TSDB_DATA_TYPE_BIGINT: {
finalResult = (double)(*(int64_t *)thisVal);
break;
}
}
free(pPage);
return finalResult;
}
pTrace("MemBucket:%p,start second round bucketing", pMemBucket);
if (pSeg->pBuffer[j]->numOfElemsInBuffer != 0) {
pTrace("MemBucket:%p,flush %d pages to disk, clear status", pMemBucket, pSeg->pBuffer[j]->numOfInMemPages);
pMemBucket->numOfAvailPages += pSeg->pBuffer[j]->numOfInMemPages;
tExtMemBufferFlush(pSeg->pBuffer[j]);
}
tExtMemBuffer *pMemBuffer = pSeg->pBuffer[j];
pSeg->pBuffer[j] = NULL;
// release all
for (int32_t tt = 0; tt < pMemBucket->numOfSegs; ++tt) {
tMemBucketSegment *pSeg = &pMemBucket->pSegs[tt];
for (int32_t ttx = 0; ttx < pSeg->numOfSlots; ++ttx) {
if (pSeg->pBuffer && pSeg->pBuffer[ttx]) {
pSeg->pBuffer[ttx] = destoryExtMemBuffer(pSeg->pBuffer[ttx]);
}
}
}
pMemBucket->nRange.i64MaxVal = pSeg->pBoundingEntries->i64MaxVal;
pMemBucket->nRange.i64MinVal = pSeg->pBoundingEntries->i64MinVal;
pMemBucket->numOfElems = 0;
for (int32_t tt = 0; tt < pMemBucket->numOfSegs; ++tt) {
tMemBucketSegment *pSeg = &pMemBucket->pSegs[tt];
for (int32_t ttx = 0; ttx < pSeg->numOfSlots; ++ttx) {
if (pSeg->pBoundingEntries) {
resetBoundingBox(pSeg, pMemBucket->dataType);
}
}
}
tFilePage *pPage = (tFilePage *)malloc(pMemBuffer->pageSize);
tFlushoutInfo *pFlushInfo = &pMemBuffer->fileMeta.flushoutData.pFlushoutInfo[0];
assert(pFlushInfo->numOfPages == pMemBuffer->fileMeta.nFileSize);
int32_t ret = fseek(pMemBuffer->file, pFlushInfo->startPageId * pMemBuffer->pageSize, SEEK_SET);
UNUSED(ret);
for (uint32_t jx = 0; jx < pFlushInfo->numOfPages; ++jx) {
ret = fread(pPage, pMemBuffer->pageSize, 1, pMemBuffer->file);
UNUSED(ret);
tMemBucketPut(pMemBucket, pPage->data, pPage->numOfElems);
}
fclose(pMemBuffer->file);
if (unlink(pMemBuffer->path) != 0) {
pError("MemBucket:%p, remove tmp file %s failed", pMemBucket, pMemBuffer->path);
}
tfree(pMemBuffer);
tfree(pPage);
return getPercentileImpl(pMemBucket, count - num, fraction);
}
} else {
num += pSeg->pBuffer[j]->numOfTotalElems;
}
}
}
return 0;
}
double getPercentile(tMemBucket *pMemBucket, double percent) {
if (pMemBucket->numOfElems == 0) {
return 0.0;
}
if (pMemBucket->numOfElems == 1) { // return the only element
return findOnlyResult(pMemBucket);
}
percent = fabs(percent);
// validate the parameters
if (fabs(percent - 100.0) < DBL_EPSILON || (percent < DBL_EPSILON)) {
double minx = 0, maxx = 0;
/*
* find the min/max value, no need to scan all data in bucket
*/
findMaxMinValue(pMemBucket, &maxx, &minx);
return fabs(percent - 100) < DBL_EPSILON ? maxx : minx;
}
double percentVal = (percent * (pMemBucket->numOfElems - 1)) / ((double)100.0);
int32_t orderIdx = (int32_t)percentVal;
// do put data by using buckets
return getPercentileImpl(pMemBucket, orderIdx, percentVal - orderIdx);
}
/*
* check if data in one slot are all identical
* only need to compare with the bounding box
*/
bool isIdenticalData(tMemBucket *pMemBucket, int32_t segIdx, int32_t slotIdx) {
tMemBucketSegment *pSeg = &pMemBucket->pSegs[segIdx];
if (pMemBucket->dataType == TSDB_DATA_TYPE_INT || pMemBucket->dataType == TSDB_DATA_TYPE_BIGINT ||
pMemBucket->dataType == TSDB_DATA_TYPE_SMALLINT || pMemBucket->dataType == TSDB_DATA_TYPE_TINYINT) {
return pSeg->pBoundingEntries[slotIdx].i64MinVal == pSeg->pBoundingEntries[slotIdx].i64MaxVal;
}
if (pMemBucket->dataType == TSDB_DATA_TYPE_FLOAT || pMemBucket->dataType == TSDB_DATA_TYPE_DOUBLE) {
return fabs(pSeg->pBoundingEntries[slotIdx].dMaxVal - pSeg->pBoundingEntries[slotIdx].dMinVal) < DBL_EPSILON;
}
return false;
}
/*
* get the first element of one slot into memory.
* if no data of current slot in memory, load it from disk
*/
char *getFirstElemOfMemBuffer(tMemBucketSegment *pSeg, int32_t slotIdx, tFilePage *pPage) {
tExtMemBuffer *pMemBuffer = pSeg->pBuffer[slotIdx];
char * thisVal = NULL;
if (pSeg->pBuffer[slotIdx]->numOfElemsInBuffer != 0) {
thisVal = pSeg->pBuffer[slotIdx]->pHead->item.data;
} else {
/*
* no data in memory, load one page into memory
*/
tFlushoutInfo *pFlushInfo = &pMemBuffer->fileMeta.flushoutData.pFlushoutInfo[0];
assert(pFlushInfo->numOfPages == pMemBuffer->fileMeta.nFileSize);
fseek(pMemBuffer->file, pFlushInfo->startPageId * pMemBuffer->pageSize, SEEK_SET);
size_t ret = fread(pPage, pMemBuffer->pageSize, 1, pMemBuffer->file);
UNUSED(ret);
thisVal = pPage->data;
}
return thisVal;
}