diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md index 18b700974c..3b1a66839d 100644 --- a/CONTRIBUTING.md +++ b/CONTRIBUTING.md @@ -6,10 +6,10 @@ We appreciate contributions from all developers. Feel free to follow us, fork th Any users can report bugs to us through the [github issue tracker](https://github.com/taosdata/TDengine/issues). We appreciate a detailed description of the problem you met. It is better to provide the detailed steps on reproducing the bug. Otherwise, an appendix with log files generated by the bug is welcome. -## Sign the contributor license agreement +## Read the contributor license agreement -It is required to sign the Contributor Licence Agreement(CLA) before a user submitting your code patch. Follow the [TaosData CLA](https://www.taosdata.com/en/contributor/) link to access the agreement and instructions on how to sign it. +It is required to agree the Contributor Licence Agreement(CLA) before a user submitting his/her code patch. Follow the [TaosData CLA](https://www.taosdata.com/en/contributor/) link to read through the agreement. ## Submit your code -Before submitting your code, make sure to [sign the contributor license agreement](#sign-the-contributor-license-agreement) beforehand. Your submission should solve an issue or add a feature registered in the [github issue tracker](https://github.com/taosdata/TDengine/issues). If no corresponding issue or feature is found in the issue tracker, please create one. When submitting your code to our repository, please create a pull request with the issue number included. +Before submitting your code, make sure to [read the contributor license agreement](#read-the-contributor-license-agreement) beforehand. If you don't accept the aggreement, please stop submitting. Your submission means you have accepted the agreement. Your submission should solve an issue or add a feature registered in the [github issue tracker](https://github.com/taosdata/TDengine/issues). If no corresponding issue or feature is found in the issue tracker, please create one. When submitting your code to our repository, please create a pull request with the issue number included. diff --git a/README.md b/README.md index ada2d8df6b..35a192d148 100644 --- a/README.md +++ b/README.md @@ -23,19 +23,39 @@ For user manual, system design and architecture, engineering blogs, refer to [TD # Building At the moment, TDengine only supports building and running on Linux systems. You can choose to [install from packages](https://www.taosdata.com/en/getting-started/#Install-from-Package) or from the source code. This quick guide is for installation from the source only. -To build TDengine, use [CMake](https://cmake.org/) 2.8 or higher versions in the project directory. - -Install CMake for example on Ubuntu: +To build TDengine, use [CMake](https://cmake.org/) 2.8 or higher versions in the project directory. Install CMake for example on Ubuntu: ``` sudo apt-get install -y cmake build-essential ``` +To compile and package the JDBC driver source code, you should have a Java jdk-8 or higher and Apache Maven 2.7 or higher installed. +To install openjdk-8 on Ubuntu: +``` +sudo apt-get install openjdk-8-jdk +``` +To install Apache Maven on Ubuntu: +``` +sudo apt-get install maven +``` + Build TDengine: ```cmd mkdir build && cd build cmake .. && cmake --build . ``` + +# Quick Run +To quickly start a TDengine server after building, run the command below in terminal: +```cmd +./build/bin/taosd -c test/cfg +``` +In another terminal, use the TDengine shell to connect the server: +``` +./build/bin/taos -c test/cfg +``` +option "-c test/cfg" specifies the system configuration file directory. + # Installing After building successfully, TDengine can be installed by: ```cmd @@ -44,47 +64,20 @@ make install Users can find more information about directories installed on the system in the [directory and files](https://www.taosdata.com/en/documentation/administrator/#Directory-and-Files) section. It should be noted that installing from source code does not configure service management for TDengine. Users can also choose to [install from packages](https://www.taosdata.com/en/getting-started/#Install-from-Package) for it. - -# Running - -To start the TDengine server, run the command below in terminal: -```cmd -./build/bin/taosd -c test/cfg -``` -In another terminal, use the TDengine shell to connect the server: -``` -./build/bin/taos -c test/cfg -``` - - -Start the service in the terminal. +To start the service after installation, in a terminal, use: ```cmd taosd ``` -Then users can use the [TDengine shell](https://www.taosdata.com/en/getting-started/#TDengine-Shell) to connect the TDengine server. +Then users can use the [TDengine shell](https://www.taosdata.com/en/getting-started/#TDengine-Shell) to connect the TDengine server. In a terminal, use: ```cmd taos ``` -If the terminal connects the server successfully, welcome messages and version info are printed. Otherwise, an error message is shown. +If TDengine shell connects the server successfully, welcome messages and version info are printed. Otherwise, an error message is shown. # Try TDengine -It is easy to run SQL commands in the terminal which is the same as other SQL databases. +It is easy to run SQL commands from TDengine shell which is the same as other SQL databases. ```sql create database db; use db; diff --git a/documentation/webdocs/markdowndocs/advanced features.md b/documentation/webdocs/markdowndocs/advanced features.md index 44d99535c7..e841a5a6a5 100644 --- a/documentation/webdocs/markdowndocs/advanced features.md +++ b/documentation/webdocs/markdowndocs/advanced features.md @@ -62,7 +62,7 @@ Time series data is a sequence of data points over time. Inside a table, the dat To reduce the development complexity and improve data consistency, TDengine provides the pub/sub functionality. To publish a message, you simply insert a record into a table. Compared with popular messaging tool Kafka, you subscribe to a table or a SQL query statement, instead of a topic. Once new data points arrive, TDengine will notify the application. The process is just like Kafka. -The detailed API will be introduced in the [connectors](https://www.taosdata.com/en/documentation/advanced-features/) section. +The detailed API will be introduced in the [connectors](https://www.taosdata.com/en/documentation/connector/) section. ##Caching TDengine allocates a fixed-size buffer in memory, the newly arrived data will be written into the buffer first. Every device or table gets one or more memory blocks. For typical IoT scenarios, the hot data shall always be newly arrived data, they are more important for timely analysis. Based on this observation, TDengine manages the cache blocks in First-In-First-Out strategy. If no enough space in the buffer, the oldest data will be saved into hard disk first, then be overwritten by newly arrived data. TDengine also guarantees every device can keep at least one block of data in the buffer. @@ -77,4 +77,4 @@ Through this design, caching tools like Redis are no longer needed in the system TDengine creates one or more virtual nodes(vnode) in each data node. Each vnode contains data for multiple tables and has its own buffer. The buffer of a vnode is fully separated from the buffer of another vnode, not shared. But the tables in a vnode share the same buffer. -System configuration parameter cacheBlockSize configures the cache block size in bytes, and another parameter cacheNumOfBlocks configures the number of cache blocks. The total memory for the buffer of a vnode is `cacheBlockSize * cacheNumOfBlocks`​. Another system parameter `numOfBlocksPerMeter` configures the maximum number of cache blocks a table can use. When you create a database, you can specify these parameters. \ No newline at end of file +System configuration parameter cacheBlockSize configures the cache block size in bytes, and another parameter cacheNumOfBlocks configures the number of cache blocks. The total memory for the buffer of a vnode is `cacheBlockSize * cacheNumOfBlocks`​. Another system parameter `numOfBlocksPerMeter` configures the maximum number of cache blocks a table can use. When you create a database, you can specify these parameters. diff --git a/src/client/inc/tscSecondaryMerge.h b/src/client/inc/tscSecondaryMerge.h index 2f47fe36d5..ce3f4091d5 100644 --- a/src/client/inc/tscSecondaryMerge.h +++ b/src/client/inc/tscSecondaryMerge.h @@ -37,13 +37,13 @@ extern "C" { struct SQLFunctionCtx; -typedef struct SLocalDataSrc { +typedef struct SLocalDataSource { tExtMemBuffer *pMemBuffer; int32_t flushoutIdx; int32_t pageId; int32_t rowIdx; tFilePage filePage; -} SLocalDataSrc; +} SLocalDataSource; enum { TSC_LOCALREDUCE_READY = 0x0, @@ -52,7 +52,7 @@ enum { }; typedef struct SLocalReducer { - SLocalDataSrc **pLocalDataSrc; + SLocalDataSource **pLocalDataSrc; int32_t numOfBuffer; int32_t numOfCompleted; diff --git a/src/client/inc/tscUtil.h b/src/client/inc/tscUtil.h index 7d470c35a4..c5c634a9c5 100644 --- a/src/client/inc/tscUtil.h +++ b/src/client/inc/tscUtil.h @@ -41,21 +41,24 @@ typedef struct SParsedColElem { } SParsedColElem; typedef struct SParsedDataColInfo { - bool ordered; // denote if the timestamp in one data block ordered or not int16_t numOfCols; int16_t numOfAssignedCols; SParsedColElem elems[TSDB_MAX_COLUMNS]; bool hasVal[TSDB_MAX_COLUMNS]; - int64_t prevTimestamp; } SParsedDataColInfo; -SInsertedDataBlocks* tscCreateDataBlock(int32_t size); -void tscDestroyDataBlock(SInsertedDataBlocks** pDataBlock); +STableDataBlocks* tscCreateDataBlock(int32_t size); +void tscDestroyDataBlock(STableDataBlocks* pDataBlock); +void tscAppendDataBlock(SDataBlockList* pList, STableDataBlocks* pBlocks); SDataBlockList* tscCreateBlockArrayList(); -void tscDestroyBlockArrayList(SDataBlockList** pList); -int32_t tscCopyDataBlockToPayload(SSqlObj* pSql, SInsertedDataBlocks* pDataBlock); +void* tscDestroyBlockArrayList(SDataBlockList* pList); +int32_t tscCopyDataBlockToPayload(SSqlObj* pSql, STableDataBlocks* pDataBlock); void tscFreeUnusedDataBlocks(SDataBlockList* pList); +void tscMergeTableDataBlocks(SSqlObj* pSql, SDataBlockList* pDataList); +STableDataBlocks* tscGetDataBlockFromList(void* pHashList, SDataBlockList* pDataBlockList, int64_t id, int32_t size, + int32_t startOffset, int32_t rowSize, char* tableId); +STableDataBlocks* tscCreateDataBlockEx(size_t size, int32_t rowSize, int32_t startOffset, char* name); SVnodeSidList* tscGetVnodeSidList(SMetricMeta* pMetricmeta, int32_t vnodeIdx); SMeterSidExtInfo* tscGetMeterSidInfo(SVnodeSidList* pSidList, int32_t idx); @@ -66,8 +69,7 @@ bool tscIsTwoStageMergeMetricQuery(SSqlObj* pSql); /** * * for the projection query on metric or point interpolation query on metric, - * we iterate all the meters, instead of invoke query on all qualified meters - * simultaneously. + * we iterate all the meters, instead of invoke query on all qualified meters simultaneously. * * @param pSql sql object * @return @@ -124,8 +126,7 @@ void tscIncStreamExecutionCount(void* pStream); bool tscValidateColumnId(SSqlCmd* pCmd, int32_t colId); -// get starter position of metric query condition (query on tags) in -// SSqlCmd.payload +// get starter position of metric query condition (query on tags) in SSqlCmd.payload char* tsGetMetricQueryCondPos(STagCond* pCond); void tscTagCondAssign(STagCond* pDst, STagCond* pSrc); void tscTagCondRelease(STagCond* pCond); @@ -139,6 +140,7 @@ void tscCleanSqlCmd(SSqlCmd* pCmd); bool tscShouldFreeAsyncSqlObj(SSqlObj* pSql); void tscDoQuery(SSqlObj* pSql); +void sortRemoveDuplicates(STableDataBlocks* dataBuf); #ifdef __cplusplus } #endif diff --git a/src/client/inc/tsclient.h b/src/client/inc/tsclient.h index 0c244e0ce9..5ed0550cd8 100644 --- a/src/client/inc/tsclient.h +++ b/src/client/inc/tsclient.h @@ -169,16 +169,22 @@ typedef struct STagCond { char * pData; } STagCond; -typedef struct SInsertedDataBlocks { - char meterId[TSDB_METER_ID_LEN]; - int64_t size; - uint32_t nAllocSize; - uint32_t numOfMeters; +typedef struct STableDataBlocks { + char meterId[TSDB_METER_ID_LEN]; + int64_t vgid; + int64_t size; + + int64_t prevTS; + bool ordered; + + int32_t numOfMeters; + int32_t rowSize; + uint32_t nAllocSize; union { char *filename; char *pData; }; -} SInsertedDataBlocks; +} STableDataBlocks; typedef struct SDataBlockList { int32_t idx; @@ -186,7 +192,7 @@ typedef struct SDataBlockList { int32_t nAlloc; char * userParam; /* user assigned parameters for async query */ void * udfp; /* user defined function pointer, used in async model */ - SInsertedDataBlocks **pData; + STableDataBlocks **pData; } SDataBlockList; typedef struct { diff --git a/src/client/src/tscAsync.c b/src/client/src/tscAsync.c index 660db08030..925a3f3c84 100644 --- a/src/client/src/tscAsync.c +++ b/src/client/src/tscAsync.c @@ -410,7 +410,7 @@ void tscAsyncInsertMultiVnodesProxy(void *param, TAOS_RES *tres, int numOfRows) tscTrace("%p Async insertion completed, destroy data block list", pSql); // release data block data - tscDestroyBlockArrayList(&pCmd->pDataBlocks); + pCmd->pDataBlocks = tscDestroyBlockArrayList(pCmd->pDataBlocks); // all data has been sent to vnode, call user function (*pSql->fp)(pSql->param, tres, numOfRows); diff --git a/src/client/src/tscParseInsert.c b/src/client/src/tscParseInsert.c index a8cd14cd67..7dec7280f8 100644 --- a/src/client/src/tscParseInsert.c +++ b/src/client/src/tscParseInsert.c @@ -53,11 +53,11 @@ return TSDB_CODE_INVALID_SQL; \ } while (0) -static void setErrMsg(char* msg, char* sql); -static int32_t tscAllocateMemIfNeed(SInsertedDataBlocks* pDataBlock, int32_t rowSize); +static void setErrMsg(char *msg, char *sql); +static int32_t tscAllocateMemIfNeed(STableDataBlocks *pDataBlock, int32_t rowSize); // get formation -static int32_t getNumericType(const char* data) { +static int32_t getNumericType(const char *data) { if (*data == '-' || *data == '+') { data += 1; } @@ -73,7 +73,7 @@ static int32_t getNumericType(const char* data) { } } -static int64_t tscToInteger(char* data, char** endPtr) { +static int64_t tscToInteger(char *data, char **endPtr) { int32_t numType = getNumericType(data); int32_t radix = 10; @@ -86,14 +86,14 @@ static int64_t tscToInteger(char* data, char** endPtr) { return strtoll(data, endPtr, radix); } -int tsParseTime(char* value, int32_t valuelen, int64_t* time, char** next, char* error, int16_t timePrec) { - char* token; +int tsParseTime(char *value, int32_t valuelen, int64_t *time, char **next, char *error, int16_t timePrec) { + char * token; int tokenlen; int64_t interval; int64_t useconds = 0; - char* pTokenEnd = *next; + char *pTokenEnd = *next; tscGetToken(pTokenEnd, &token, &tokenlen); if (tokenlen == 0 && strlen(value) == 0) { INVALID_SQL_RET_MSG(error, "missing time stamp"); @@ -135,9 +135,9 @@ int tsParseTime(char* value, int32_t valuelen, int64_t* time, char** next, char* } /* - * time expression: - * e.g., now+12a, now-5h - */ + * time expression: + * e.g., now+12a, now-5h + */ pTokenEnd = tscGetToken(pTokenEnd, &token, &tokenlen); if (tokenlen && (*token == '+' || *token == '-')) { pTokenEnd = tscGetToken(pTokenEnd, &value, &valuelen); @@ -166,32 +166,32 @@ int tsParseTime(char* value, int32_t valuelen, int64_t* time, char** next, char* return TSDB_CODE_SUCCESS; } -int32_t tsParseOneColumnData(SSchema* pSchema, char* value, int valuelen, char* payload, char* msg, char** str, +int32_t tsParseOneColumnData(SSchema *pSchema, char *value, int valuelen, char *payload, char *msg, char **str, bool primaryKey, int16_t timePrec) { int64_t temp; - int32_t nullInt = *(int32_t*)TSDB_DATA_NULL_STR_L; - char* endptr = NULL; + int32_t nullInt = *(int32_t *)TSDB_DATA_NULL_STR_L; + char * endptr = NULL; errno = 0; // reset global error code switch (pSchema->type) { case TSDB_DATA_TYPE_BOOL: { // bool - if (valuelen == 4 && nullInt == *(int32_t*)value) { - *(uint8_t*)payload = TSDB_DATA_BOOL_NULL; + if (valuelen == 4 && nullInt == *(int32_t *)value) { + *(uint8_t *)payload = TSDB_DATA_BOOL_NULL; } else { if (strncmp(value, "true", valuelen) == 0) { - *(uint8_t*)payload = TSDB_TRUE; + *(uint8_t *)payload = TSDB_TRUE; } else if (strncmp(value, "false", valuelen) == 0) { - *(uint8_t*)payload = TSDB_FALSE; + *(uint8_t *)payload = TSDB_FALSE; } else { int64_t v = strtoll(value, NULL, 10); - *(uint8_t*)payload = (int8_t)((v == 0) ? TSDB_FALSE : TSDB_TRUE); + *(uint8_t *)payload = (int8_t)((v == 0) ? TSDB_FALSE : TSDB_TRUE); } } break; } case TSDB_DATA_TYPE_TINYINT: - if (valuelen == 4 && nullInt == *(int32_t*)value) { - *((int32_t*)payload) = TSDB_DATA_TINYINT_NULL; + if (valuelen == 4 && nullInt == *(int32_t *)value) { + *((int32_t *)payload) = TSDB_DATA_TINYINT_NULL; } else { int64_t v = tscToInteger(value, &endptr); if (errno == ERANGE || v > INT8_MAX || v < INT8_MIN) { @@ -199,18 +199,18 @@ int32_t tsParseOneColumnData(SSchema* pSchema, char* value, int valuelen, char* } int8_t v8 = (int8_t)v; - if (isNull((char*)&v8, pSchema->type)) { + if (isNull((char *)&v8, pSchema->type)) { INVALID_SQL_RET_MSG(msg, "data is overflow"); } - *((int8_t*)payload) = v8; + *((int8_t *)payload) = v8; } break; case TSDB_DATA_TYPE_SMALLINT: - if (valuelen == 4 && nullInt == *(int32_t*)value) { - *((int32_t*)payload) = TSDB_DATA_SMALLINT_NULL; + if (valuelen == 4 && nullInt == *(int32_t *)value) { + *((int32_t *)payload) = TSDB_DATA_SMALLINT_NULL; } else { int64_t v = tscToInteger(value, &endptr); @@ -219,17 +219,17 @@ int32_t tsParseOneColumnData(SSchema* pSchema, char* value, int valuelen, char* } int16_t v16 = (int16_t)v; - if (isNull((char*)&v16, pSchema->type)) { + if (isNull((char *)&v16, pSchema->type)) { INVALID_SQL_RET_MSG(msg, "data is overflow"); } - *((int16_t*)payload) = v16; + *((int16_t *)payload) = v16; } break; case TSDB_DATA_TYPE_INT: - if (valuelen == 4 && nullInt == *(int32_t*)value) { - *((int32_t*)payload) = TSDB_DATA_INT_NULL; + if (valuelen == 4 && nullInt == *(int32_t *)value) { + *((int32_t *)payload) = TSDB_DATA_INT_NULL; } else { int64_t v = tscToInteger(value, &endptr); @@ -238,36 +238,36 @@ int32_t tsParseOneColumnData(SSchema* pSchema, char* value, int valuelen, char* } int32_t v32 = (int32_t)v; - if (isNull((char*)&v32, pSchema->type)) { + if (isNull((char *)&v32, pSchema->type)) { INVALID_SQL_RET_MSG(msg, "data is overflow"); } - *((int32_t*)payload) = v32; + *((int32_t *)payload) = v32; } break; case TSDB_DATA_TYPE_BIGINT: - if (valuelen == 4 && nullInt == *(int32_t*)value) { - *((int64_t*)payload) = TSDB_DATA_BIGINT_NULL; + if (valuelen == 4 && nullInt == *(int32_t *)value) { + *((int64_t *)payload) = TSDB_DATA_BIGINT_NULL; } else { int64_t v = tscToInteger(value, &endptr); - if (isNull((char*)&v, pSchema->type) || errno == ERANGE) { + if (isNull((char *)&v, pSchema->type) || errno == ERANGE) { INVALID_SQL_RET_MSG(msg, "data is overflow"); } - *((int64_t*)payload) = v; + *((int64_t *)payload) = v; } break; case TSDB_DATA_TYPE_FLOAT: - if (valuelen == 4 && nullInt == *(int32_t*)value) { - *((int32_t*)payload) = TSDB_DATA_FLOAT_NULL; + if (valuelen == 4 && nullInt == *(int32_t *)value) { + *((int32_t *)payload) = TSDB_DATA_FLOAT_NULL; } else { float v = (float)strtod(value, &endptr); - if (isNull((char*)&v, pSchema->type) || isinf(v) || isnan(v)) { - *((int32_t*)payload) = TSDB_DATA_FLOAT_NULL; + if (isNull((char *)&v, pSchema->type) || isinf(v) || isnan(v)) { + *((int32_t *)payload) = TSDB_DATA_FLOAT_NULL; } else { - *((float*)payload) = v; + *((float *)payload) = v; } if (str != NULL) { @@ -280,14 +280,14 @@ int32_t tsParseOneColumnData(SSchema* pSchema, char* value, int valuelen, char* break; case TSDB_DATA_TYPE_DOUBLE: - if (valuelen == 4 && nullInt == *(int32_t*)value) { - *((int64_t*)payload) = TSDB_DATA_DOUBLE_NULL; + if (valuelen == 4 && nullInt == *(int32_t *)value) { + *((int64_t *)payload) = TSDB_DATA_DOUBLE_NULL; } else { double v = strtod(value, &endptr); - if (isNull((char*)&v, pSchema->type) || isinf(v) || isnan(v)) { - *((int32_t*)payload) = TSDB_DATA_FLOAT_NULL; + if (isNull((char *)&v, pSchema->type) || isinf(v) || isnan(v)) { + *((int32_t *)payload) = TSDB_DATA_FLOAT_NULL; } else { - *((double*)payload) = v; + *((double *)payload) = v; } if (str != NULL) { @@ -300,8 +300,10 @@ int32_t tsParseOneColumnData(SSchema* pSchema, char* value, int valuelen, char* break; case TSDB_DATA_TYPE_BINARY: - // binary data cannot be null-terminated char string, otherwise the last char of the string is lost - if (valuelen == 4 && nullInt == *(int32_t*)value) { + /* + * binary data cannot be null-terminated char string, otherwise the last char of the string is lost + */ + if (valuelen == 4 && nullInt == *(int32_t *)value) { *payload = TSDB_DATA_BINARY_NULL; } else { /* truncate too long string */ @@ -312,8 +314,8 @@ int32_t tsParseOneColumnData(SSchema* pSchema, char* value, int valuelen, char* break; case TSDB_DATA_TYPE_NCHAR: - if (valuelen == 4 && nullInt == *(int32_t*)value) { - *(uint32_t*)payload = TSDB_DATA_NCHAR_NULL; + if (valuelen == 4 && nullInt == *(int32_t *)value) { + *(uint32_t *)payload = TSDB_DATA_NCHAR_NULL; } else { if (!taosMbsToUcs4(value, valuelen, payload, pSchema->bytes)) { sprintf(msg, "%s", strerror(errno)); @@ -323,17 +325,17 @@ int32_t tsParseOneColumnData(SSchema* pSchema, char* value, int valuelen, char* break; case TSDB_DATA_TYPE_TIMESTAMP: { - if (valuelen == 4 && nullInt == *(int32_t*)value) { + if (valuelen == 4 && nullInt == *(int32_t *)value) { if (primaryKey) { - *((int64_t*)payload) = 0; + *((int64_t *)payload) = 0; } else { - *((int64_t*)payload) = TSDB_DATA_BIGINT_NULL; + *((int64_t *)payload) = TSDB_DATA_BIGINT_NULL; } } else { if (tsParseTime(value, valuelen, &temp, str, msg, timePrec) != TSDB_CODE_SUCCESS) { return TSDB_CODE_INVALID_SQL; } - *((int64_t*)payload) = temp; + *((int64_t *)payload) = temp; } break; @@ -344,7 +346,7 @@ int32_t tsParseOneColumnData(SSchema* pSchema, char* value, int valuelen, char* } // todo merge the error msg function with tSQLParser -static void setErrMsg(char* msg, char* sql) { +static void setErrMsg(char *msg, char *sql) { const char* msgFormat = "near \"%s\" syntax error"; const int32_t BACKWARD_CHAR_STEP = 15; @@ -354,16 +356,18 @@ static void setErrMsg(char* msg, char* sql) { sprintf(msg, msgFormat, buf); } -int tsParseOneRowData(char** str, char* payload, SSchema schema[], SParsedDataColInfo* spd, char* error, +int tsParseOneRowData(char **str, STableDataBlocks *pDataBlocks, SSchema schema[], SParsedDataColInfo *spd, char *error, int16_t timePrec) { - char* value = NULL; + char *value = NULL; int valuelen = 0; + char *payload = pDataBlocks->pData + pDataBlocks->size; + /* 1. set the parsed value from sql string */ int32_t rowSize = 0; for (int i = 0; i < spd->numOfAssignedCols; ++i) { /* the start position in data block buffer of current value in sql */ - char* start = payload + spd->elems[i].offset; + char * start = payload + spd->elems[i].offset; int16_t colIndex = spd->elems[i].colIndex; rowSize += schema[colIndex].bytes; @@ -394,19 +398,19 @@ int tsParseOneRowData(char** str, char* payload, SSchema schema[], SParsedDataCo } // once the data block is disordered, we do NOT keep previous timestamp any more - if (colIndex == PRIMARYKEY_TIMESTAMP_COL_INDEX && spd->ordered) { - TSKEY k = *(TSKEY*)start; - if (k < spd->prevTimestamp) { - spd->ordered = false; + if (colIndex == PRIMARYKEY_TIMESTAMP_COL_INDEX && pDataBlocks->ordered) { + TSKEY k = *(TSKEY *)start; + if (k <= pDataBlocks->prevTS) { + pDataBlocks->ordered = false; } - spd->prevTimestamp = k; + pDataBlocks->prevTS = k; } } /*2. set the null value for the rest columns */ if (spd->numOfAssignedCols < spd->numOfCols) { - char* ptr = payload; + char *ptr = payload; for (int32_t i = 0; i < spd->numOfCols; ++i) { if (!spd->hasVal[i]) { @@ -423,39 +427,42 @@ int tsParseOneRowData(char** str, char* payload, SSchema schema[], SParsedDataCo return rowSize; } -static int32_t rowDataCompar(const void* lhs, const void* rhs) { - TSKEY left = GET_INT64_VAL(lhs); - TSKEY right = GET_INT64_VAL(rhs); - DEFAULT_COMP(left, right); +static int32_t rowDataCompar(const void *lhs, const void *rhs) { + TSKEY left = *(TSKEY *)lhs; + TSKEY right = *(TSKEY *)rhs; + + if (left == right) { + return 0; + } else { + return left > right ? 1 : -1; + } } -int tsParseValues(char** str, SInsertedDataBlocks* pDataBlock, SMeterMeta* pMeterMeta, int maxRows, - SParsedDataColInfo* spd, char* error) { - char* token; +int tsParseValues(char **str, STableDataBlocks *pDataBlock, SMeterMeta *pMeterMeta, int maxRows, + SParsedDataColInfo *spd, char *error) { + char *token; int tokenlen; - SSchema* pSchema = tsGetSchema(pMeterMeta); - int16_t numOfRows = 0; - pDataBlock->size += sizeof(SShellSubmitBlock); - if (!spd->hasVal[0]) { + SSchema *pSchema = tsGetSchema(pMeterMeta); + int32_t precision = pMeterMeta->precision; + + if (spd->hasVal[0] == false) { sprintf(error, "primary timestamp column can not be null"); return -1; } while (1) { - char* tmp = tscGetToken(*str, &token, &tokenlen); + char *tmp = tscGetToken(*str, &token, &tokenlen); if (tokenlen == 0 || *token != '(') break; *str = tmp; - if (numOfRows >= maxRows || - pDataBlock->size + pMeterMeta->rowSize + sizeof(SShellSubmitBlock) >= pDataBlock->nAllocSize) { + if (numOfRows >= maxRows || pDataBlock->size + pMeterMeta->rowSize >= pDataBlock->nAllocSize) { maxRows += tscAllocateMemIfNeed(pDataBlock, pMeterMeta->rowSize); } - int32_t len = - tsParseOneRowData(str, pDataBlock->pData + pDataBlock->size, pSchema, spd, error, pMeterMeta->precision); + int32_t len = tsParseOneRowData(str, pDataBlock, pSchema, spd, error, precision); if (len <= 0) { setErrMsg(error, *str); return -1; @@ -474,26 +481,25 @@ int tsParseValues(char** str, SInsertedDataBlocks* pDataBlock, SMeterMeta* pMete if (numOfRows <= 0) { strcpy(error, "no any data points"); + return -1; + } else { + return numOfRows; } - - return numOfRows; } -static void appendDataBlock(SDataBlockList* pList, SInsertedDataBlocks* pBlocks) { +void tscAppendDataBlock(SDataBlockList *pList, STableDataBlocks *pBlocks) { if (pList->nSize >= pList->nAlloc) { pList->nAlloc = pList->nAlloc << 1; - pList->pData = realloc(pList->pData, (size_t)pList->nAlloc); + pList->pData = realloc(pList->pData, sizeof(void *) * (size_t)pList->nAlloc); // reset allocated memory - memset(pList->pData + pList->nSize, 0, POINTER_BYTES * (pList->nAlloc - pList->nSize)); + memset(pList->pData + pList->nSize, 0, sizeof(void *) * (pList->nAlloc - pList->nSize)); } pList->pData[pList->nSize++] = pBlocks; } -static void tscSetAssignedColumnInfo(SParsedDataColInfo* spd, SSchema* pSchema, int16_t numOfCols) { - spd->ordered = true; - spd->prevTimestamp = INT64_MIN; +static void tscSetAssignedColumnInfo(SParsedDataColInfo *spd, SSchema *pSchema, int32_t numOfCols) { spd->numOfCols = numOfCols; spd->numOfAssignedCols = numOfCols; @@ -507,106 +513,125 @@ static void tscSetAssignedColumnInfo(SParsedDataColInfo* spd, SSchema* pSchema, } } -int32_t tscAllocateMemIfNeed(SInsertedDataBlocks* pDataBlock, int32_t rowSize) { +int32_t tscAllocateMemIfNeed(STableDataBlocks *pDataBlock, int32_t rowSize) { size_t remain = pDataBlock->nAllocSize - pDataBlock->size; + const int factor = 5; // expand the allocated size - if (remain <= sizeof(SShellSubmitBlock) + rowSize) { - int32_t oldSize = pDataBlock->nAllocSize; + if (remain < rowSize * factor) { + while (remain < rowSize * factor) { + pDataBlock->nAllocSize = (uint32_t) (pDataBlock->nAllocSize * 1.5); + remain = pDataBlock->nAllocSize - pDataBlock->size; + } - pDataBlock->nAllocSize = (uint32_t)(oldSize * 1.5); - - char* tmp = realloc(pDataBlock->pData, (size_t)pDataBlock->nAllocSize); + char *tmp = realloc(pDataBlock->pData, (size_t)pDataBlock->nAllocSize); if (tmp != NULL) { pDataBlock->pData = tmp; + memset(pDataBlock->pData + pDataBlock->size, 0, pDataBlock->nAllocSize - pDataBlock->size); } else { + assert(false); // do nothing } } - return (int32_t)(pDataBlock->nAllocSize - pDataBlock->size - sizeof(SShellSubmitBlock)) / rowSize; + return (int32_t)(pDataBlock->nAllocSize - pDataBlock->size) / rowSize; } -void tsSetBlockInfo(SShellSubmitBlock* pBlocks, const SMeterMeta* pMeterMeta, int32_t numOfRows) { - pBlocks->sid = htonl(pMeterMeta->sid); - pBlocks->uid = htobe64(pMeterMeta->uid); - pBlocks->sversion = htonl(pMeterMeta->sversion); - pBlocks->numOfRows = htons(numOfRows); +static void tsSetBlockInfo(SShellSubmitBlock *pBlocks, const SMeterMeta *pMeterMeta, int32_t numOfRows) { + pBlocks->sid = pMeterMeta->sid; + pBlocks->uid = pMeterMeta->uid; + pBlocks->sversion = pMeterMeta->sversion; + pBlocks->numOfRows += numOfRows; } -static int32_t doParseInsertStatement(SSqlCmd* pCmd, SSqlRes* pRes, void* pDataBlockHashList, char** str, - SParsedDataColInfo* spd) { - SMeterMeta* pMeterMeta = pCmd->pMeterMeta; - int32_t numOfRows = 0; +// data block is disordered, sort it in ascending order +void sortRemoveDuplicates(STableDataBlocks *dataBuf) { + SShellSubmitBlock* pBlocks = (SShellSubmitBlock*)dataBuf->pData; - SInsertedDataBlocks** pData = (SInsertedDataBlocks**)taosGetIntHashData(pDataBlockHashList, pMeterMeta->vgid); - SInsertedDataBlocks* dataBuf = NULL; + // size is less than the total size, since duplicated rows may be removed yet. + assert(pBlocks->numOfRows * dataBuf->rowSize + sizeof(SShellSubmitBlock) == dataBuf->size); - /* no data in hash list */ - if (pData == NULL) { - dataBuf = tscCreateDataBlock(TSDB_PAYLOAD_SIZE); + if (!dataBuf->ordered) { + char *pBlockData = pBlocks->payLoad; + qsort(pBlockData, pBlocks->numOfRows, dataBuf->rowSize, rowDataCompar); - /* here we only keep the pointer of chunk of buffer, not the whole buffer */ - dataBuf = *(SInsertedDataBlocks**)taosAddIntHash(pDataBlockHashList, pCmd->pMeterMeta->vgid, (char*)&dataBuf); + int32_t i = 0; + int32_t j = 1; - dataBuf->size = tsInsertHeadSize; - strncpy(dataBuf->meterId, pCmd->name, tListLen(pCmd->name)); - appendDataBlock(pCmd->pDataBlocks, dataBuf); - } else { - dataBuf = *pData; + while (j < pBlocks->numOfRows) { + TSKEY ti = *(TSKEY *)(pBlockData + dataBuf->rowSize * i); + TSKEY tj = *(TSKEY *)(pBlockData + dataBuf->rowSize * j); + + if (ti == tj) { + ++j; + continue; + } + + int32_t nextPos = (++i); + if (nextPos != j) { + memmove(pBlockData + dataBuf->rowSize * nextPos, pBlockData + dataBuf->rowSize * j, dataBuf->rowSize); + } + + ++j; + } + + dataBuf->ordered = true; + + pBlocks->numOfRows = i + 1; + dataBuf->size = sizeof(SShellSubmitBlock) + dataBuf->rowSize*pBlocks->numOfRows; } +} + +static int32_t doParseInsertStatement(SSqlObj *pSql, void *pTableHashList, char **str, SParsedDataColInfo *spd, + int32_t *totalNum) { + SSqlCmd * pCmd = &pSql->cmd; + SMeterMeta *pMeterMeta = pCmd->pMeterMeta; + + STableDataBlocks *dataBuf = + tscGetDataBlockFromList(pTableHashList, pCmd->pDataBlocks, pMeterMeta->uid, TSDB_DEFAULT_PAYLOAD_SIZE, + sizeof(SShellSubmitBlock), pMeterMeta->rowSize, pCmd->name); int32_t maxNumOfRows = tscAllocateMemIfNeed(dataBuf, pMeterMeta->rowSize); - int64_t startPos = dataBuf->size; - numOfRows = tsParseValues(str, dataBuf, pMeterMeta, maxNumOfRows, spd, pCmd->payload); + int32_t numOfRows = tsParseValues(str, dataBuf, pMeterMeta, maxNumOfRows, spd, pCmd->payload); if (numOfRows <= 0) { return TSDB_CODE_INVALID_SQL; } - // data block is disordered, sort it in ascending order - if (!spd->ordered) { - char* pBlockData = dataBuf->pData + startPos + sizeof(SShellSubmitBlock); - qsort(pBlockData, numOfRows, pMeterMeta->rowSize, rowDataCompar); - spd->ordered = true; - } - - SShellSubmitBlock* pBlocks = (SShellSubmitBlock*)(dataBuf->pData + startPos); + SShellSubmitBlock *pBlocks = (SShellSubmitBlock *)(dataBuf->pData); tsSetBlockInfo(pBlocks, pMeterMeta, numOfRows); - dataBuf->numOfMeters += 1; + + dataBuf->vgid = pMeterMeta->vgid; + dataBuf->numOfMeters = 1; /* - * the value of pRes->numOfRows does not affect the true result of AFFECTED ROWS, which is - * actually returned from server. - * - * NOTE: - * The better way is to use a local variable to store the number of rows that - * has been extracted from sql expression string, and avoid to do the invalid write check + * the value of pRes->numOfRows does not affect the true result of AFFECTED ROWS, + * which is actually returned from server. */ - pRes->numOfRows += numOfRows; + *totalNum += numOfRows; return TSDB_CODE_SUCCESS; } -static int32_t tscParseSqlForCreateTableOnDemand(char** sqlstr, SSqlObj* pSql) { - char* id = NULL; +static int32_t tscParseSqlForCreateTableOnDemand(char **sqlstr, SSqlObj *pSql) { + char * id = NULL; int32_t idlen = 0; int32_t code = TSDB_CODE_SUCCESS; - SSqlCmd* pCmd = &pSql->cmd; - char* sql = *sqlstr; + SSqlCmd *pCmd = &pSql->cmd; + char * sql = *sqlstr; sql = tscGetToken(sql, &id, &idlen); /* build the token of specified table */ SSQLToken tableToken = {.z = id, .n = idlen, .type = TK_ID}; - char* cstart = NULL; - char* cend = NULL; + char *cstart = NULL; + char *cend = NULL; /* skip possibly exists column list */ sql = tscGetToken(sql, &id, &idlen); int32_t numOfColList = 0; - bool createTable = false; + bool createTable = false; if (id[0] == '(' && idlen == 1) { cstart = &id[0]; @@ -630,7 +655,7 @@ static int32_t tscParseSqlForCreateTableOnDemand(char** sqlstr, SSqlObj* pSql) { if (strncmp(id, "using", idlen) == 0 && idlen == 5) { /* create table if not exists */ sql = tscGetToken(sql, &id, &idlen); - STagData* pTag = (STagData*)pCmd->payload; + STagData *pTag = (STagData *)pCmd->payload; memset(pTag, 0, sizeof(STagData)); SSQLToken token1 = {idlen, TK_ID, id}; @@ -643,8 +668,14 @@ static int32_t tscParseSqlForCreateTableOnDemand(char** sqlstr, SSqlObj* pSql) { return code; } - char* tagVal = pTag->data; - SSchema* pTagSchema = tsGetTagSchema(pCmd->pMeterMeta); + if (!UTIL_METER_IS_METRIC(pCmd)) { + const char* msg = "create table only from super table is allowed"; + sprintf(pCmd->payload, "%s", msg); + return TSDB_CODE_INVALID_SQL; + } + + char * tagVal = pTag->data; + SSchema *pTagSchema = tsGetTagSchema(pCmd->pMeterMeta); sql = tscGetToken(sql, &id, &idlen); if (!(strncmp(id, "tags", idlen) == 0 && idlen == 4)) { @@ -722,11 +753,11 @@ static int32_t tscParseSqlForCreateTableOnDemand(char** sqlstr, SSqlObj* pSql) { return code; } -int validateTableName(char* tblName, int len) { +int validateTableName(char *tblName, int len) { char buf[TSDB_METER_ID_LEN] = {0}; - memcpy(buf, tblName, len); + strncpy(buf, tblName, len); - SSQLToken token = {len, TK_ID, buf}; + SSQLToken token = {.n = len, .type = TK_ID, .z = buf}; tSQLGetToken(buf, &token.type); return tscValidateName(&token); @@ -742,18 +773,21 @@ int validateTableName(char* tblName, int len) { * @param pSql * @return */ -int tsParseInsertStatement(SSqlCmd* pCmd, char* str, char* acct, char* db, SSqlObj* pSql) { +int tsParseInsertStatement(SSqlObj *pSql, char *str, char *acct, char *db) { + SSqlCmd *pCmd = &pSql->cmd; + pCmd->command = TSDB_SQL_INSERT; pCmd->isInsertFromFile = -1; pCmd->count = 0; pSql->res.numOfRows = 0; + int32_t totalNum = 0; if (!pSql->pTscObj->writeAuth) { return TSDB_CODE_NO_RIGHTS; } - char* id; + char *id; int idlen; int code = TSDB_CODE_INVALID_SQL; @@ -766,15 +800,21 @@ int tsParseInsertStatement(SSqlCmd* pCmd, char* str, char* acct, char* db, SSqlO return code; } - void* pDataBlockHashList = taosInitIntHash(4, POINTER_BYTES, taosHashInt); + void *pTableHashList = taosInitIntHash(128, sizeof(void *), taosHashInt); pSql->cmd.pDataBlocks = tscCreateBlockArrayList(); tscTrace("%p create data block list for submit data, %p", pSql, pSql->cmd.pDataBlocks); while (1) { tscGetToken(str, &id, &idlen); + if (idlen == 0) { - if ((pSql->res.numOfRows > 0) || (1 == pCmd->isInsertFromFile)) { + // parse file, do not release the STableDataBlock + if (pCmd->isInsertFromFile == 1) { + goto _clean; + } + + if (totalNum > 0) { break; } else { // no data in current sql string, error code = TSDB_CODE_INVALID_SQL; @@ -782,10 +822,7 @@ int tsParseInsertStatement(SSqlCmd* pCmd, char* str, char* acct, char* db, SSqlO } } - /* - * Check the validity of the table name - * - */ + // Check if the table name available or not if (validateTableName(id, idlen) != TSDB_CODE_SUCCESS) { code = TSDB_CODE_INVALID_SQL; sprintf(pCmd->payload, "table name is invalid"); @@ -797,7 +834,7 @@ int tsParseInsertStatement(SSqlCmd* pCmd, char* str, char* acct, char* db, SSqlO goto _error_clean; } - void* fp = pSql->fp; + void *fp = pSql->fp; if ((code = tscParseSqlForCreateTableOnDemand(&str, pSql)) != TSDB_CODE_SUCCESS) { if (fp != NULL) { goto _clean; @@ -826,7 +863,7 @@ int tsParseInsertStatement(SSqlCmd* pCmd, char* str, char* acct, char* db, SSqlO if (strncmp(id, "values", 6) == 0 && idlen == 6) { SParsedDataColInfo spd = {0}; - SSchema* pSchema = tsGetSchema(pCmd->pMeterMeta); + SSchema * pSchema = tsGetSchema(pCmd->pMeterMeta); tscSetAssignedColumnInfo(&spd, pSchema, pCmd->pMeterMeta->numOfColumns); @@ -844,7 +881,7 @@ int tsParseInsertStatement(SSqlCmd* pCmd, char* str, char* acct, char* db, SSqlO * app here insert data in different vnodes, so we need to set the following * data in another submit procedure using async insert routines */ - code = doParseInsertStatement(pCmd, &pSql->res, pDataBlockHashList, &str, &spd); + code = doParseInsertStatement(pSql, pTableHashList, &str, &spd, &totalNum); if (code != TSDB_CODE_SUCCESS) { goto _error_clean; } @@ -867,34 +904,29 @@ int tsParseInsertStatement(SSqlCmd* pCmd, char* str, char* acct, char* db, SSqlO goto _error_clean; } - char* fname = calloc(1, idlen + 1); - memcpy(fname, id, idlen); + char fname[PATH_MAX] = {0}; + strncpy(fname, id, idlen); + strdequote(fname); wordexp_t full_path; if (wordexp(fname, &full_path, 0) != 0) { code = TSDB_CODE_INVALID_SQL; sprintf(pCmd->payload, "invalid filename"); - free(fname); goto _error_clean; } - strcpy(fname, full_path.we_wordv[0]); wordfree(&full_path); - SInsertedDataBlocks* dataBuf = tscCreateDataBlock(strlen(fname) + sizeof(SInsertedDataBlocks) + 1); - strcpy(dataBuf->filename, fname); - - dataBuf->size = strlen(fname) + 1; - free(fname); - - strcpy(dataBuf->meterId, pCmd->name); - appendDataBlock(pCmd->pDataBlocks, dataBuf); + STableDataBlocks* pDataBlock = tscCreateDataBlockEx(PATH_MAX, pCmd->pMeterMeta->rowSize, sizeof(SShellSubmitBlock), + pCmd->name); + tscAppendDataBlock(pCmd->pDataBlocks, pDataBlock); + strcpy(pDataBlock->filename, fname); str = id + idlen; } else if (idlen == 1 && id[0] == '(') { /* insert into tablename(col1, col2,..., coln) values(v1, v2,... vn); */ - SMeterMeta* pMeterMeta = pCmd->pMeterMeta; - SSchema* pSchema = tsGetSchema(pMeterMeta); + SMeterMeta *pMeterMeta = pCmd->pMeterMeta; + SSchema * pSchema = tsGetSchema(pMeterMeta); if (pCmd->isInsertFromFile == -1) { pCmd->isInsertFromFile = 0; @@ -905,8 +937,6 @@ int tsParseInsertStatement(SSqlCmd* pCmd, char* str, char* acct, char* db, SSqlO } SParsedDataColInfo spd = {0}; - spd.ordered = true; - spd.prevTimestamp = INT64_MIN; spd.numOfCols = pMeterMeta->numOfColumns; int16_t offset[TSDB_MAX_COLUMNS] = {0}; @@ -925,7 +955,7 @@ int tsParseInsertStatement(SSqlCmd* pCmd, char* str, char* acct, char* db, SSqlO // todo speedup by using hash list for (int32_t t = 0; t < pMeterMeta->numOfColumns; ++t) { if (strncmp(id, pSchema[t].name, idlen) == 0 && strlen(pSchema[t].name) == idlen) { - SParsedColElem* pElem = &spd.elems[spd.numOfAssignedCols++]; + SParsedColElem *pElem = &spd.elems[spd.numOfAssignedCols++]; pElem->offset = offset[t]; pElem->colIndex = t; @@ -961,7 +991,7 @@ int tsParseInsertStatement(SSqlCmd* pCmd, char* str, char* acct, char* db, SSqlO goto _error_clean; } - code = doParseInsertStatement(pCmd, &pSql->res, pDataBlockHashList, &str, &spd); + code = doParseInsertStatement(pSql, pTableHashList, &str, &spd, &totalNum); if (code != TSDB_CODE_SUCCESS) { goto _error_clean; } @@ -972,53 +1002,49 @@ int tsParseInsertStatement(SSqlCmd* pCmd, char* str, char* acct, char* db, SSqlO } } - /* submit to more than one vnode */ + // submit to more than one vnode if (pCmd->pDataBlocks->nSize > 0) { - // lihui: if import file, only malloc the size of file name - if (1 != pCmd->isInsertFromFile) { - tscFreeUnusedDataBlocks(pCmd->pDataBlocks); + // merge according to vgid + tscMergeTableDataBlocks(pSql, pCmd->pDataBlocks); - SInsertedDataBlocks* pDataBlock = pCmd->pDataBlocks->pData[0]; - if ((code = tscCopyDataBlockToPayload(pSql, pDataBlock)) != TSDB_CODE_SUCCESS) { - goto _error_clean; - } + STableDataBlocks *pDataBlock = pCmd->pDataBlocks->pData[0]; + if ((code = tscCopyDataBlockToPayload(pSql, pDataBlock)) != TSDB_CODE_SUCCESS) { + goto _error_clean; } pCmd->vnodeIdx = 1; // set the next sent data vnode index in data block arraylist } else { - tscDestroyBlockArrayList(&pCmd->pDataBlocks); + pCmd->pDataBlocks = tscDestroyBlockArrayList(pCmd->pDataBlocks); } code = TSDB_CODE_SUCCESS; goto _clean; _error_clean: - tscDestroyBlockArrayList(&pCmd->pDataBlocks); + pCmd->pDataBlocks = tscDestroyBlockArrayList(pCmd->pDataBlocks); _clean: - taosCleanUpIntHash(pDataBlockHashList); + taosCleanUpIntHash(pTableHashList); return code; } -int tsParseImportStatement(SSqlObj* pSql, char* str, char* acct, char* db) { - SSqlCmd* pCmd = &pSql->cmd; +int tsParseImportStatement(SSqlObj *pSql, char *str, char *acct, char *db) { + SSqlCmd *pCmd = &pSql->cmd; pCmd->order.order = TSQL_SO_ASC; - return tsParseInsertStatement(pCmd, str, acct, db, pSql); + return tsParseInsertStatement(pSql, str, acct, db); } -int tsParseInsertSql(SSqlObj* pSql, char* sql, char* acct, char* db) { - char* verb; +int tsParseInsertSql(SSqlObj *pSql, char *sql, char *acct, char *db) { + char *verb; int verblen; int code = TSDB_CODE_INVALID_SQL; - SSqlCmd* pCmd = &pSql->cmd; - tscCleanSqlCmd(pCmd); - + SSqlCmd *pCmd = &pSql->cmd; sql = tscGetToken(sql, &verb, &verblen); if (verblen) { if (strncmp(verb, "insert", 6) == 0 && verblen == 6) { - code = tsParseInsertStatement(pCmd, sql, acct, db, pSql); + code = tsParseInsertStatement(pSql, sql, acct, db); } else if (strncmp(verb, "import", 6) == 0 && verblen == 6) { code = tsParseImportStatement(pSql, sql, acct, db); } else { @@ -1029,27 +1055,24 @@ int tsParseInsertSql(SSqlObj* pSql, char* sql, char* acct, char* db) { sprintf(pCmd->payload, "no any keywords"); } - // sql object has not been released in async model - if (pSql->signature == pSql) { - pSql->res.numOfRows = 0; - } - return code; } -int tsParseSql(SSqlObj* pSql, char* acct, char* db, bool multiVnodeInsertion) { +int tsParseSql(SSqlObj *pSql, char *acct, char *db, bool multiVnodeInsertion) { int32_t ret = TSDB_CODE_SUCCESS; + tscCleanSqlCmd(&pSql->cmd); if (tscIsInsertOrImportData(pSql->sqlstr)) { /* - * only for async multi-vnode insertion Set the fp before parse the sql string, in case of getmetermeta failed, - * in which the error handle callback function can rightfully restore the user defined function (fp) + * only for async multi-vnode insertion + * Set the fp before parse the sql string, in case of getmetermeta failed, in which + * the error handle callback function can rightfully restore the user defined function (fp) */ if (pSql->fp != NULL && multiVnodeInsertion) { assert(pSql->fetchFp == NULL); pSql->fetchFp = pSql->fp; - /* replace user defined callback function with multi-insert proxy function*/ + /* replace user defined callback function with multi-insert proxy function */ pSql->fp = tscAsyncInsertMultiVnodesProxy; } @@ -1057,6 +1080,9 @@ int tsParseSql(SSqlObj* pSql, char* acct, char* db, bool multiVnodeInsertion) { } else { SSqlInfo SQLInfo = {0}; tSQLParse(&SQLInfo, pSql->sqlstr); + + tscAllocPayloadWithSize(&pSql->cmd, TSDB_DEFAULT_PAYLOAD_SIZE); + ret = tscToSQLCmd(pSql, &SQLInfo); SQLInfoDestroy(&SQLInfo); } @@ -1072,31 +1098,55 @@ int tsParseSql(SSqlObj* pSql, char* acct, char* db, bool multiVnodeInsertion) { return ret; } -static int tscInsertDataFromFile(SSqlObj* pSql, FILE* fp) { - // TODO : import data from file - int readLen = 0; - char* line = NULL; - size_t n = 0; - int len = 0; - uint32_t maxRows = 0; - SSqlCmd* pCmd = &pSql->cmd; - char* pStart = pCmd->payload + tsInsertHeadSize; - SMeterMeta* pMeterMeta = pCmd->pMeterMeta; - int numOfRows = 0; - uint32_t rowSize = pMeterMeta->rowSize; - char error[128] = "\0"; - SShellSubmitBlock* pBlock = (SShellSubmitBlock*)(pStart); - pStart += sizeof(SShellSubmitBlock); - int nrows = 0; +static int doPackSendDataBlock(SSqlObj* pSql, int32_t numOfRows, STableDataBlocks* pTableDataBlocks) { + int32_t code = TSDB_CODE_SUCCESS; + SSqlCmd* pCmd = &pSql->cmd; - const int32_t RESERVED_SIZE = 1024; + SMeterMeta* pMeterMeta = pCmd->pMeterMeta; - maxRows = (TSDB_PAYLOAD_SIZE - RESERVED_SIZE - sizeof(SShellSubmitBlock)) / rowSize; + SShellSubmitBlock *pBlocks = (SShellSubmitBlock *)(pTableDataBlocks->pData); + tsSetBlockInfo(pBlocks, pMeterMeta, numOfRows); + + tscMergeTableDataBlocks(pSql, pCmd->pDataBlocks); + + // the pDataBlock is different from the pTableDataBlocks + STableDataBlocks *pDataBlock = pCmd->pDataBlocks->pData[0]; + if ((code = tscCopyDataBlockToPayload(pSql, pDataBlock)) != TSDB_CODE_SUCCESS) { + return code; + } + + if ((code = tscProcessSql(pSql)) != TSDB_CODE_SUCCESS) { + return code; + } + + return TSDB_CODE_SUCCESS; +} + +static int tscInsertDataFromFile(SSqlObj *pSql, FILE *fp) { + int readLen = 0; + char * line = NULL; + size_t n = 0; + int len = 0; + uint32_t maxRows = 0; + SSqlCmd * pCmd = &pSql->cmd; + SMeterMeta *pMeterMeta = pCmd->pMeterMeta; + int numOfRows = 0; + int32_t rowSize = pMeterMeta->rowSize; + int32_t code = 0; + int nrows = 0; + + pCmd->pDataBlocks = tscCreateBlockArrayList(); + STableDataBlocks* pTableDataBlock = tscCreateDataBlockEx(TSDB_PAYLOAD_SIZE, pMeterMeta->rowSize, + sizeof(SShellSubmitBlock), pCmd->name); + + tscAppendDataBlock(pCmd->pDataBlocks, pTableDataBlock); + + maxRows = tscAllocateMemIfNeed(pTableDataBlock, rowSize); if (maxRows < 1) return -1; int count = 0; - SParsedDataColInfo spd = {0}; - SSchema* pSchema = tsGetSchema(pCmd->pMeterMeta); + SParsedDataColInfo spd = {.numOfCols = pCmd->pMeterMeta->numOfColumns}; + SSchema * pSchema = tsGetSchema(pCmd->pMeterMeta); tscSetAssignedColumnInfo(&spd, pSchema, pCmd->pMeterMeta->numOfColumns); @@ -1105,43 +1155,42 @@ static int tscInsertDataFromFile(SSqlObj* pSql, FILE* fp) { if (('\r' == line[readLen - 1]) || ('\n' == line[readLen - 1])) line[--readLen] = 0; if (readLen <= 0) continue; - char* lineptr = line; + char *lineptr = line; strtolower(line, line); - len = tsParseOneRowData(&lineptr, pStart, pSchema, &spd, error, pCmd->pMeterMeta->precision); + + len = tsParseOneRowData(&lineptr, pTableDataBlock, pSchema, &spd, pCmd->payload, pMeterMeta->precision); if (len <= 0) return -1; - pStart += len; + + pTableDataBlock->size += len; count++; nrows++; if (count >= maxRows) { - pCmd->payloadLen = (pStart - pCmd->payload); - pBlock->sid = htonl(pMeterMeta->sid); - pBlock->numOfRows = htons(count); - pSql->res.numOfRows = 0; - if (tscProcessSql(pSql) != 0) { - return -1; + if ((code = doPackSendDataBlock(pSql, count, pTableDataBlock)) != TSDB_CODE_SUCCESS) { + return -code; } + + pTableDataBlock = pCmd->pDataBlocks->pData[0]; + pTableDataBlock->size = sizeof(SShellSubmitBlock); + pTableDataBlock->rowSize = pMeterMeta->rowSize; + numOfRows += pSql->res.numOfRows; + pSql->res.numOfRows = 0; count = 0; - memset(pCmd->payload, 0, TSDB_PAYLOAD_SIZE); - pStart = pCmd->payload + tsInsertHeadSize; - pBlock = (SShellSubmitBlock*)(pStart); - pStart += sizeof(SShellSubmitBlock); } } if (count > 0) { - pCmd->payloadLen = (pStart - pCmd->payload); - pBlock->sid = htonl(pMeterMeta->sid); - pBlock->numOfRows = htons(count); - pSql->res.numOfRows = 0; - if (tscProcessSql(pSql) != 0) { - return -1; + if ((code = doPackSendDataBlock(pSql, count, pTableDataBlock)) != TSDB_CODE_SUCCESS) { + return -code; } + numOfRows += pSql->res.numOfRows; + pSql->res.numOfRows = 0; } if (line) tfree(line); + return numOfRows; } @@ -1151,20 +1200,21 @@ static int tscInsertDataFromFile(SSqlObj* pSql, FILE* fp) { * 2019.05.10 lihui * Remove the code for importing records from files */ -void tscProcessMultiVnodesInsert(SSqlObj* pSql) { - SSqlCmd* pCmd = &pSql->cmd; +void tscProcessMultiVnodesInsert(SSqlObj *pSql) { + SSqlCmd *pCmd = &pSql->cmd; if (pCmd->command != TSDB_SQL_INSERT) { return; } - SInsertedDataBlocks* pDataBlock = NULL; - int32_t code = TSDB_CODE_SUCCESS; + STableDataBlocks *pDataBlock = NULL; + int32_t affected_rows = 0; + int32_t code = TSDB_CODE_SUCCESS; /* the first block has been sent to server in processSQL function */ assert(pCmd->isInsertFromFile != -1 && pCmd->vnodeIdx >= 1 && pCmd->pDataBlocks != NULL); if (pCmd->vnodeIdx < pCmd->pDataBlocks->nSize) { - SDataBlockList* pDataBlocks = pCmd->pDataBlocks; + SDataBlockList *pDataBlocks = pCmd->pDataBlocks; for (int32_t i = pCmd->vnodeIdx; i < pDataBlocks->nSize; ++i) { pDataBlock = pDataBlocks->pData[i]; @@ -1182,59 +1232,70 @@ void tscProcessMultiVnodesInsert(SSqlObj* pSql) { } // all data have been submit to vnode, release data blocks - tscDestroyBlockArrayList(&pCmd->pDataBlocks); + pCmd->pDataBlocks = tscDestroyBlockArrayList(pCmd->pDataBlocks); } /* multi-vnodes insertion in sync query model */ -void tscProcessMultiVnodesInsertForFile(SSqlObj* pSql) { - SSqlCmd* pCmd = &pSql->cmd; +void tscProcessMultiVnodesInsertForFile(SSqlObj *pSql) { + SSqlCmd *pCmd = &pSql->cmd; if (pCmd->command != TSDB_SQL_INSERT) { return; } - SInsertedDataBlocks* pDataBlock = NULL; - int32_t affected_rows = 0; + STableDataBlocks *pDataBlock = NULL; + int32_t affected_rows = 0; - assert(pCmd->isInsertFromFile == 1 && pCmd->vnodeIdx >= 1 && pCmd->pDataBlocks != NULL); + assert(pCmd->isInsertFromFile == 1 && pCmd->pDataBlocks != NULL); + SDataBlockList *pDataBlockList = pCmd->pDataBlocks; + pCmd->pDataBlocks = NULL; - SDataBlockList* pDataBlocks = pCmd->pDataBlocks; + char path[PATH_MAX] = {0}; - pCmd->isInsertFromFile = 0; // for tscProcessSql() - - pSql->res.numOfRows = 0; - for (int32_t i = 0; i < pDataBlocks->nSize; ++i) { - pDataBlock = pDataBlocks->pData[i]; + for (int32_t i = 0; i < pDataBlockList->nSize; ++i) { + pDataBlock = pDataBlockList->pData[i]; if (pDataBlock == NULL) { continue; } tscAllocPayloadWithSize(pCmd, TSDB_PAYLOAD_SIZE); - pCmd->count = 1; - FILE* fp = fopen(pDataBlock->filename, "r"); + strncpy(path, pDataBlock->filename, PATH_MAX); + + FILE *fp = fopen(path, "r"); if (fp == NULL) { - tscError("%p Failed to open file %s to insert data from file", pSql, pDataBlock->filename); + tscError("%p failed to open file %s to load data from file, reason:%s", pSql, path, + strerror(errno)); continue; } strcpy(pCmd->name, pDataBlock->meterId); - tscGetMeterMeta(pSql, pCmd->name); - int nrows = tscInsertDataFromFile(pSql, fp); - if (nrows < 0) { - fclose(fp); - tscTrace("%p There is no record in file %s", pSql, pDataBlock->filename); + memset(pDataBlock->pData, 0, pDataBlock->nAllocSize); + + int32_t ret = tscGetMeterMeta(pSql, pCmd->name); + if (ret != TSDB_CODE_SUCCESS) { + tscError("%p get meter meta failed, abort", pSql); continue; } - fclose(fp); + int nrows = tscInsertDataFromFile(pSql, fp); + pCmd->pDataBlocks = tscDestroyBlockArrayList(pCmd->pDataBlocks); + + if (nrows < 0) { + fclose(fp); + tscTrace("%p no records in file %s", pSql, path); + continue; + } + + fclose(fp); affected_rows += nrows; - tscTrace("%p Insert data %d records from file %s", pSql, nrows, pDataBlock->filename); + tscTrace("%p Insert data %d records from file %s", pSql, nrows, path); } pSql->res.numOfRows = affected_rows; // all data have been submit to vnode, release data blocks - tscDestroyBlockArrayList(&pCmd->pDataBlocks); + pCmd->pDataBlocks = tscDestroyBlockArrayList(pCmd->pDataBlocks); + tscDestroyBlockArrayList(pDataBlockList); } diff --git a/src/client/src/tscSQLParser.c b/src/client/src/tscSQLParser.c index 33f699660d..cf43aad94b 100644 --- a/src/client/src/tscSQLParser.c +++ b/src/client/src/tscSQLParser.c @@ -143,9 +143,6 @@ int32_t tscToSQLCmd(SSqlObj* pSql, struct SSqlInfo* pInfo) { return TSDB_CODE_INVALID_SQL; } - tscCleanSqlCmd(pCmd); - tscAllocPayloadWithSize(pCmd, TSDB_DEFAULT_PAYLOAD_SIZE); - // transfer pInfo into select operation switch (pInfo->sqlType) { case DROP_TABLE: @@ -785,7 +782,8 @@ int32_t tscToSQLCmd(SSqlObj* pSql, struct SSqlInfo* pInfo) { // set sliding value SSQLToken* pSliding = &pQuerySql->sliding; if (pSliding->n != 0) { - if (!tscEmbedded) { + // pCmd->count == 1 means sql in stream function + if (!tscEmbedded && pCmd->count == 0) { const char* msg = "not support sliding in query"; setErrMsg(pCmd, msg); return TSDB_CODE_INVALID_SQL; diff --git a/src/client/src/tscSQLParserImpl.c b/src/client/src/tscSQLParserImpl.c index ebb6fecbd2..a0d2f09ae7 100644 --- a/src/client/src/tscSQLParserImpl.c +++ b/src/client/src/tscSQLParserImpl.c @@ -140,12 +140,10 @@ tSQLExpr *tSQLExprIdValueCreate(SSQLToken *pToken, int32_t optrType) { nodePtr->val.nType = TSDB_DATA_TYPE_BIGINT; nodePtr->nSQLOptr = TK_TIMESTAMP; } else { // must be field id if not numbers - if (pToken != NULL) { - assert(optrType == TK_ID); - /* it must be the column name (tk_id) */ + assert(optrType == TK_ALL || optrType == TK_ID); + + if (pToken != NULL) { // it must be the column name (tk_id) nodePtr->colInfo = *pToken; - } else { - assert(optrType == TK_ALL); } nodePtr->nSQLOptr = optrType; diff --git a/src/client/src/tscSecondaryMerge.c b/src/client/src/tscSecondaryMerge.c index b912264c4e..d4745f6525 100644 --- a/src/client/src/tscSecondaryMerge.c +++ b/src/client/src/tscSecondaryMerge.c @@ -19,25 +19,26 @@ #include #include "tlosertree.h" -#include "tsclient.h" +#include "tlosertree.h" #include "tscSecondaryMerge.h" #include "tscUtil.h" +#include "tsclient.h" #include "tutil.h" typedef struct SCompareParam { - SLocalDataSrc ** pLocalData; - tOrderDescriptor *pDesc; - int32_t numOfElems; - int32_t groupOrderType; + SLocalDataSource **pLocalData; + tOrderDescriptor * pDesc; + int32_t numOfElems; + int32_t groupOrderType; } SCompareParam; int32_t treeComparator(const void *pLeft, const void *pRight, void *param) { int32_t pLeftIdx = *(int32_t *)pLeft; int32_t pRightIdx = *(int32_t *)pRight; - SCompareParam * pParam = (SCompareParam *)param; - tOrderDescriptor *pDesc = pParam->pDesc; - SLocalDataSrc ** pLocalData = pParam->pLocalData; + SCompareParam * pParam = (SCompareParam *)param; + tOrderDescriptor * pDesc = pParam->pDesc; + SLocalDataSource **pLocalData = pParam->pLocalData; /* this input is exhausted, set the special value to denote this */ if (pLocalData[pLeftIdx]->rowIdx == -1) { @@ -105,7 +106,7 @@ static void tscInitSqlContext(SSqlCmd *pCmd, SSqlRes *pRes, SLocalReducer *pRedu } /* - * todo error process with async process + * todo release allocated memory process with async process */ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrderDescriptor *pDesc, tColModel *finalmodel, SSqlCmd *pCmd, SSqlRes *pRes) { @@ -133,32 +134,32 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd if (numOfFlush == 0 || numOfBuffer == 0) { tscLocalReducerEnvDestroy(pMemBuffer, pDesc, finalmodel, numOfBuffer); - tscTrace("%p retrieved no data", pSqlObjAddr); + return; } if (pDesc->pSchema->maxCapacity >= pMemBuffer[0]->nPageSize) { - tscLocalReducerEnvDestroy(pMemBuffer, pDesc, finalmodel, numOfBuffer); - tscError("%p Invalid value of buffer capacity %d and page size %d ", pSqlObjAddr, pDesc->pSchema->maxCapacity, pMemBuffer[0]->nPageSize); + + tscLocalReducerEnvDestroy(pMemBuffer, pDesc, finalmodel, numOfBuffer); pRes->code = TSDB_CODE_APP_ERROR; return; } - size_t nReducerSize = sizeof(SLocalReducer) + POINTER_BYTES * numOfFlush; + size_t nReducerSize = sizeof(SLocalReducer) + sizeof(void *) * numOfFlush; SLocalReducer *pReducer = (SLocalReducer *)calloc(1, nReducerSize); if (pReducer == NULL) { - tscLocalReducerEnvDestroy(pMemBuffer, pDesc, finalmodel, numOfBuffer); - tscError("%p failed to create merge structure", pSqlObjAddr); + + tscLocalReducerEnvDestroy(pMemBuffer, pDesc, finalmodel, numOfBuffer); pRes->code = TSDB_CODE_CLI_OUT_OF_MEMORY; return; } pReducer->pExtMemBuffer = pMemBuffer; - pReducer->pLocalDataSrc = (SLocalDataSrc **)&pReducer[1]; + pReducer->pLocalDataSrc = (SLocalDataSource **)&pReducer[1]; assert(pReducer->pLocalDataSrc != NULL); pReducer->numOfBuffer = numOfFlush; @@ -172,7 +173,7 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd int32_t numOfFlushoutInFile = pMemBuffer[i]->fileMeta.flushoutData.nLength; for (int32_t j = 0; j < numOfFlushoutInFile; ++j) { - SLocalDataSrc *pDS = (SLocalDataSrc *)malloc(sizeof(SLocalDataSrc) + pMemBuffer[0]->nPageSize); + SLocalDataSource *pDS = (SLocalDataSource *)malloc(sizeof(SLocalDataSource) + pMemBuffer[0]->nPageSize); if (pDS == NULL) { tscError("%p failed to create merge structure", pSqlObjAddr); pRes->code = TSDB_CODE_CLI_OUT_OF_MEMORY; @@ -468,9 +469,7 @@ static int32_t createOrderDescriptor(tOrderDescriptor **pOrderDesc, SSqlCmd *pCm } if (pCmd->nAggTimeInterval != 0) { - /* - * the first column is the timestamp, handles queries like "interval(10m) group by tags" - */ + //the first column is the timestamp, handles queries like "interval(10m) group by tags" orderIdx[numOfGroupByCols - 1] = PRIMARYKEY_TIMESTAMP_COL_INDEX; } } @@ -485,29 +484,32 @@ static int32_t createOrderDescriptor(tOrderDescriptor **pOrderDesc, SSqlCmd *pCm } } -bool isSameGroupOfPrev(SSqlCmd *pCmd, SLocalReducer *pReducer, char *pPrev, tFilePage *tmpPage) { +bool isSameGroup(SSqlCmd *pCmd, SLocalReducer *pReducer, char *pPrev, tFilePage *tmpBuffer) { int16_t functionId = tscSqlExprGet(pCmd, 0)->sqlFuncId; - if (functionId == TSDB_FUNC_PRJ || functionId == TSDB_FUNC_ARITHM) { // column projection query - return false; // disable merge procedure + + // disable merge procedure for column projection query + if (functionId == TSDB_FUNC_PRJ || functionId == TSDB_FUNC_ARITHM) { + return false; } tOrderDescriptor *pOrderDesc = pReducer->pDesc; int32_t numOfCols = pOrderDesc->orderIdx.numOfOrderedCols; - if (numOfCols > 0) { - if (pOrderDesc->orderIdx.pData[numOfCols - 1] == PRIMARYKEY_TIMESTAMP_COL_INDEX) { //<= 0 - /* metric interval query */ - assert(pCmd->nAggTimeInterval > 0); - pOrderDesc->orderIdx.numOfOrderedCols -= 1; - } else { /* simple group by query */ - assert(pCmd->nAggTimeInterval == 0); - } - } else { + // no group by columns, all data belongs to one group + if (numOfCols <= 0) { return true; } + if (pOrderDesc->orderIdx.pData[numOfCols - 1] == PRIMARYKEY_TIMESTAMP_COL_INDEX) { //<= 0 + // super table interval query + assert(pCmd->nAggTimeInterval > 0); + pOrderDesc->orderIdx.numOfOrderedCols -= 1; + } else { // simple group by query + assert(pCmd->nAggTimeInterval == 0); + } + // only one row exists - int32_t ret = compare_a(pOrderDesc, 1, 0, pPrev, 1, 0, tmpPage->data); + int32_t ret = compare_a(pOrderDesc, 1, 0, pPrev, 1, 0, tmpBuffer->data); pOrderDesc->orderIdx.numOfOrderedCols = numOfCols; return (ret == 0); @@ -602,7 +604,7 @@ void tscLocalReducerEnvDestroy(tExtMemBuffer **pMemBuffer, tOrderDescriptor *pDe * @param treeList * @return the number of remain input source. if ret == 0, all data has been handled */ -int32_t loadNewDataFromDiskFor(SLocalReducer *pLocalReducer, SLocalDataSrc *pOneInterDataSrc, +int32_t loadNewDataFromDiskFor(SLocalReducer *pLocalReducer, SLocalDataSource *pOneInterDataSrc, bool *needAdjustLoserTree) { pOneInterDataSrc->rowIdx = 0; pOneInterDataSrc->pageId += 1; @@ -629,8 +631,8 @@ int32_t loadNewDataFromDiskFor(SLocalReducer *pLocalReducer, SLocalDataSrc *pOne return pLocalReducer->numOfBuffer; } -void loadDataIntoMemAndAdjustLoserTree(SLocalReducer *pLocalReducer, SLocalDataSrc *pOneInterDataSrc, - SLoserTreeInfo *pTree) { +void adjustLoserTreeFromNewData(SLocalReducer *pLocalReducer, SLocalDataSource *pOneInterDataSrc, + SLoserTreeInfo *pTree) { /* * load a new data page into memory for intermediate dataset source, * since it's last record in buffer has been chosen to be processed, as the winner of loser-tree @@ -662,10 +664,10 @@ void loadDataIntoMemAndAdjustLoserTree(SLocalReducer *pLocalReducer, SLocalDataS } } -void savePrevRecordAndSetupInterpoInfo(SLocalReducer *pLocalReducer, SSqlCmd *pCmd, - SInterpolationInfo *pInterpoInfo) { // discard following dataset in the - // same group and reset the - // interpolation information +void savePrevRecordAndSetupInterpoInfo( + SLocalReducer *pLocalReducer, SSqlCmd *pCmd, + SInterpolationInfo + *pInterpoInfo) { // discard following dataset in the same group and reset the interpolation information int64_t stime = (pCmd->stime < pCmd->etime) ? pCmd->stime : pCmd->etime; int64_t revisedSTime = taosGetIntervalStartTimestamp(stime, pCmd->nAggTimeInterval, pCmd->intervalTimeUnit); @@ -749,7 +751,7 @@ static void doInterpolateResult(SSqlObj *pSql, SLocalReducer *pLocalReducer, boo tColModelErase(pLocalReducer->resColModel, pFinalDataPage, prevSize, 0, pCmd->limit.offset - 1); /* remove the hole in column model */ - tColModelCompress(pLocalReducer->resColModel, pFinalDataPage, prevSize); + tColModelCompact(pLocalReducer->resColModel, pFinalDataPage, prevSize); pRes->numOfRows -= pCmd->limit.offset; pRes->numOfTotal -= pCmd->limit.offset; @@ -772,7 +774,7 @@ static void doInterpolateResult(SSqlObj *pSql, SLocalReducer *pLocalReducer, boo pRes->numOfRows -= overFlow; pFinalDataPage->numOfElems -= overFlow; - tColModelCompress(pLocalReducer->resColModel, pFinalDataPage, prevSize); + tColModelCompact(pLocalReducer->resColModel, pFinalDataPage, prevSize); /* set remain data to be discarded, and reset the interpolation information */ savePrevRecordAndSetupInterpoInfo(pLocalReducer, pCmd, &pLocalReducer->interpolationInfo); @@ -892,21 +894,21 @@ static void doInterpolateResult(SSqlObj *pSql, SLocalReducer *pLocalReducer, boo free(srcData); } -static void savePrevRecord(SLocalReducer *pLocalReducer, tFilePage *tmpPages) { +static void savePreviousRow(SLocalReducer *pLocalReducer, tFilePage *tmpBuffer) { tColModel *pColModel = pLocalReducer->pDesc->pSchema; - assert(pColModel->maxCapacity == 1 && tmpPages->numOfElems == 1); + assert(pColModel->maxCapacity == 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], tmpPages->data + pColModel->colOffset[i], + memcpy(pLocalReducer->prevRowOfInput + pColModel->colOffset[i], tmpBuffer->data + pColModel->colOffset[i], pColModel->pFields[i].bytes); } - tmpPages->numOfElems = 0; + tmpBuffer->numOfElems = 0; pLocalReducer->hasPrevRow = true; } -static void handleUnprocessedRow(SLocalReducer *pLocalReducer, SSqlCmd *pCmd, tFilePage *tmpPages) { +static void handleUnprocessedRow(SLocalReducer *pLocalReducer, SSqlCmd *pCmd, tFilePage *tmpBuffer) { if (pLocalReducer->hasUnprocessedRow) { for (int32_t j = 0; j < pCmd->fieldsInfo.numOfOutputCols; ++j) { SSqlExpr *pExpr = tscSqlExprGet(pCmd, j); @@ -922,7 +924,7 @@ static void handleUnprocessedRow(SLocalReducer *pLocalReducer, SSqlCmd *pCmd, tF pLocalReducer->hasUnprocessedRow = false; // copy to previous temp buffer - savePrevRecord(pLocalReducer, tmpPages); + savePreviousRow(pLocalReducer, tmpBuffer); } } @@ -1005,7 +1007,7 @@ int32_t finalizeRes(SSqlCmd *pCmd, SLocalReducer *pLocalReducer) { * results generated by simple aggregation function, we merge them all into one points * *Exception*: column projection query, required no merge procedure */ -bool needToMerge(SSqlCmd *pCmd, SLocalReducer *pLocalReducer, tFilePage *tmpPages) { +bool needToMerge(SSqlCmd *pCmd, SLocalReducer *pLocalReducer, tFilePage *tmpBuffer) { int32_t ret = 0; // merge all result by default int16_t functionId = tscSqlExprGet(pCmd, 0)->sqlFuncId; @@ -1016,9 +1018,9 @@ bool needToMerge(SSqlCmd *pCmd, SLocalReducer *pLocalReducer, tFilePage *tmpPage if (pDesc->orderIdx.numOfOrderedCols > 0) { if (pDesc->tsOrder == TSQL_SO_ASC) { // asc // todo refactor comparator - ret = compare_a(pLocalReducer->pDesc, 1, 0, pLocalReducer->prevRowOfInput, 1, 0, tmpPages->data); + ret = compare_a(pLocalReducer->pDesc, 1, 0, pLocalReducer->prevRowOfInput, 1, 0, tmpBuffer->data); } else { // desc - ret = compare_d(pLocalReducer->pDesc, 1, 0, pLocalReducer->prevRowOfInput, 1, 0, tmpPages->data); + ret = compare_d(pLocalReducer->pDesc, 1, 0, pLocalReducer->prevRowOfInput, 1, 0, tmpBuffer->data); } } } @@ -1027,23 +1029,55 @@ bool needToMerge(SSqlCmd *pCmd, SLocalReducer *pLocalReducer, tFilePage *tmpPage return (ret == 0); } -void savePreGroupNumOfRes(SSqlRes *pRes) { - // pRes->numOfGroups += 1; - // pRes->pGroupRec = realloc(pRes->pGroupRec, - // pRes->numOfGroups*sizeof(SResRec)); - // - // pRes->pGroupRec[pRes->numOfGroups-1].numOfRows = pRes->numOfRows; - // pRes->pGroupRec[pRes->numOfGroups-1].numOfTotal = pRes->numOfTotal; +static bool reachGroupResultLimit(SSqlCmd *pCmd, SSqlRes *pRes) { + return (pRes->numOfGroups >= pCmd->glimit.limit && pCmd->glimit.limit >= 0); } -void doGenerateFinalResults(SSqlObj *pSql, SLocalReducer *pLocalReducer, - bool doneOuput) { // there are merged results in buffer, flush to client +static bool saveGroupResultInfo(SSqlObj *pSql) { + SSqlCmd *pCmd = &pSql->cmd; + SSqlRes *pRes = &pSql->res; + + pRes->numOfGroups += 1; + + // the output group is limited by the glimit clause + if (reachGroupResultLimit(pCmd, pRes)) { + return true; + } + + // pRes->pGroupRec = realloc(pRes->pGroupRec, pRes->numOfGroups*sizeof(SResRec)); + // pRes->pGroupRec[pRes->numOfGroups-1].numOfRows = pRes->numOfRows; + // pRes->pGroupRec[pRes->numOfGroups-1].numOfTotal = pRes->numOfTotal; + + return false; +} + +/** + * + * @param pSql + * @param pLocalReducer + * @param noMoreCurrentGroupRes + * @return if current group is skipped, return false, and do NOT record it into pRes->numOfGroups + */ +bool doGenerateFinalResults(SSqlObj *pSql, SLocalReducer *pLocalReducer, bool noMoreCurrentGroupRes) { SSqlCmd * pCmd = &pSql->cmd; SSqlRes * pRes = &pSql->res; tFilePage *pResBuf = pLocalReducer->pResultBuf; tColModel *pModel = pLocalReducer->resColModel; - tColModelCompress(pModel, pResBuf, pModel->maxCapacity); + pRes->code = TSDB_CODE_SUCCESS; + + /* + * ignore the output of the current group since this group is skipped by user + * We set the numOfRows to be 0 and discard the possible remain results. + */ + if (pCmd->glimit.offset > 0) { + pRes->numOfRows = 0; + pCmd->glimit.offset -= 1; + pLocalReducer->discard = !noMoreCurrentGroupRes; + return false; + } + + tColModelCompact(pModel, pResBuf, pModel->maxCapacity); memcpy(pLocalReducer->pBufForInterpo, pResBuf->data, pLocalReducer->nResultBufSize); #ifdef _DEBUG_VIEW @@ -1061,9 +1095,9 @@ void doGenerateFinalResults(SSqlObj *pSql, SLocalReducer *pLocalReducer, } taosInterpoSetStartInfo(&pLocalReducer->interpolationInfo, pResBuf->numOfElems, pCmd->interpoType); - doInterpolateResult(pSql, pLocalReducer, doneOuput); + doInterpolateResult(pSql, pLocalReducer, noMoreCurrentGroupRes); - pRes->code = TSDB_CODE_SUCCESS; + return true; } void resetOutputBuf(SSqlCmd *pCmd, SLocalReducer *pLocalReducer) { // reset output buffer to the beginning @@ -1075,10 +1109,8 @@ void resetOutputBuf(SSqlCmd *pCmd, SLocalReducer *pLocalReducer) { // reset out memset(pLocalReducer->pResultBuf, 0, pLocalReducer->nResultBufSize + sizeof(tFilePage)); } -static void setUpForNewGroupRes(SSqlRes *pRes, SSqlCmd *pCmd, SLocalReducer *pLocalReducer) { - /* - * In handling data in other groups, we need to reset the interpolation information for a new group data - */ +static void resetEnvForNewResultset(SSqlRes *pRes, SSqlCmd *pCmd, SLocalReducer *pLocalReducer) { + //In handling data in other groups, we need to reset the interpolation information for a new group data pRes->numOfRows = 0; pRes->numOfTotal = 0; pCmd->limit.offset = pLocalReducer->offset; @@ -1093,6 +1125,124 @@ static void setUpForNewGroupRes(SSqlRes *pRes, SSqlCmd *pCmd, SLocalReducer *pLo } } +static bool isAllSourcesCompleted(SLocalReducer *pLocalReducer) { + return (pLocalReducer->numOfBuffer == pLocalReducer->numOfCompleted); +} + +static bool doInterpolationForCurrentGroup(SSqlObj *pSql) { + SSqlCmd *pCmd = &pSql->cmd; + SSqlRes *pRes = &pSql->res; + + SLocalReducer * pLocalReducer = pRes->pLocalReducer; + SInterpolationInfo *pInterpoInfo = &pLocalReducer->interpolationInfo; + + if (taosHasRemainsDataForInterpolation(pInterpoInfo)) { + assert(pCmd->interpoType != TSDB_INTERPO_NONE); + + tFilePage *pFinalDataBuf = pLocalReducer->pResultBuf; + int64_t etime = *(int64_t *)(pFinalDataBuf->data + TSDB_KEYSIZE * (pInterpoInfo->numOfRawDataInRows - 1)); + + int32_t remain = taosNumOfRemainPoints(pInterpoInfo); + TSKEY ekey = taosGetRevisedEndKey(etime, pCmd->order.order, pCmd->nAggTimeInterval, pCmd->intervalTimeUnit); + int32_t rows = taosGetNumOfResultWithInterpo(pInterpoInfo, (TSKEY *)pLocalReducer->pBufForInterpo, remain, + pCmd->nAggTimeInterval, ekey, pLocalReducer->resColModel->maxCapacity); + if (rows > 0) { // do interpo + doInterpolateResult(pSql, pLocalReducer, false); + } + + return true; + } else { + return false; + } +} + +static bool doHandleLastRemainData(SSqlObj *pSql) { + SSqlCmd *pCmd = &pSql->cmd; + SSqlRes *pRes = &pSql->res; + + SLocalReducer * pLocalReducer = pRes->pLocalReducer; + SInterpolationInfo *pInterpoInfo = &pLocalReducer->interpolationInfo; + + bool prevGroupCompleted = (!pLocalReducer->discard) && pLocalReducer->hasUnprocessedRow; + + if ((isAllSourcesCompleted(pLocalReducer) && !pLocalReducer->hasPrevRow) || pLocalReducer->pLocalDataSrc[0] == NULL || + prevGroupCompleted) { + // if interpoType == TSDB_INTERPO_NONE, return directly + if (pCmd->interpoType != TSDB_INTERPO_NONE) { + int64_t etime = (pCmd->stime < pCmd->etime) ? pCmd->etime : pCmd->stime; + + etime = taosGetRevisedEndKey(etime, pCmd->order.order, pCmd->nAggTimeInterval, pCmd->intervalTimeUnit); + int32_t rows = taosGetNumOfResultWithInterpo(pInterpoInfo, NULL, 0, pCmd->nAggTimeInterval, etime, + pLocalReducer->resColModel->maxCapacity); + if (rows > 0) { // do interpo + doInterpolateResult(pSql, pLocalReducer, true); + } + } + + /* + * 1. numOfRows == 0, means no interpolation results are generated. + * 2. if all local data sources are consumed, and no un-processed rows exist. + * + * No results will be generated and query completed. + */ + if (pRes->numOfRows > 0 || (isAllSourcesCompleted(pLocalReducer) && (!pLocalReducer->hasUnprocessedRow))) { + return true; + } + + // start to process result for a new group and save the result info of previous group + if (saveGroupResultInfo(pSql)) { + return true; + } + + resetEnvForNewResultset(pRes, pCmd, pLocalReducer); + } + + return false; +} + +static void doMergeWithPrevRows(SSqlObj *pSql, int32_t numOfRes) { + SSqlCmd * pCmd = &pSql->cmd; + SSqlRes * pRes = &pSql->res; + SLocalReducer *pLocalReducer = pRes->pLocalReducer; + + for (int32_t k = 0; k < pCmd->fieldsInfo.numOfOutputCols; ++k) { + SSqlExpr *pExpr = tscSqlExprGet(pCmd, k); + + pLocalReducer->pCtx[k].aOutputBuf += pLocalReducer->pCtx[k].outputBytes * numOfRes; + + // set the correct output timestamp column position + if (pExpr->sqlFuncId == TSDB_FUNC_TOP_DST || pExpr->sqlFuncId == TSDB_FUNC_BOTTOM_DST) { + pLocalReducer->pCtx[k].ptsOutputBuf = ((char *)pLocalReducer->pCtx[k].ptsOutputBuf + TSDB_KEYSIZE * numOfRes); + } + + /* set the parameters for the SQLFunctionCtx */ + tVariantAssign(&pLocalReducer->pCtx[k].param[0], &pExpr->param[0]); + + aAggs[pExpr->sqlFuncId].init(&pLocalReducer->pCtx[k]); + pLocalReducer->pCtx[k].currentStage = SECONDARY_STAGE_MERGE; + aAggs[pExpr->sqlFuncId].distSecondaryMergeFunc(&pLocalReducer->pCtx[k]); + } +} + +static void doExecuteSecondaryMerge(SSqlObj *pSql) { + SSqlCmd * pCmd = &pSql->cmd; + SSqlRes * pRes = &pSql->res; + SLocalReducer *pLocalReducer = pRes->pLocalReducer; + + for (int32_t j = 0; j < pCmd->fieldsInfo.numOfOutputCols; ++j) { + SSqlExpr *pExpr = tscSqlExprGet(pCmd, j); + + tVariantAssign(&pLocalReducer->pCtx[j].param[0], &pExpr->param[0]); + pLocalReducer->pCtx[j].numOfIteratedElems = 0; + pLocalReducer->pCtx[j].currentStage = 0; + + aAggs[pExpr->sqlFuncId].init(&pLocalReducer->pCtx[j]); + pLocalReducer->pCtx[j].currentStage = SECONDARY_STAGE_MERGE; + + aAggs[pExpr->sqlFuncId].distSecondaryMergeFunc(&pLocalReducer->pCtx[j]); + } +} + int32_t tscLocalDoReduce(SSqlObj *pSql) { SSqlCmd *pCmd = &pSql->cmd; SSqlRes *pRes = &pSql->res; @@ -1111,82 +1261,35 @@ int32_t tscLocalDoReduce(SSqlObj *pSql) { SLocalReducer *pLocalReducer = pRes->pLocalReducer; - // set the local reduce in progress + // set the data merge in progress int32_t prevStatus = __sync_val_compare_and_swap_32(&pLocalReducer->status, TSC_LOCALREDUCE_READY, TSC_LOCALREDUCE_IN_PROGRESS); if (prevStatus != TSC_LOCALREDUCE_READY || pLocalReducer == NULL) { assert(prevStatus == TSC_LOCALREDUCE_TOBE_FREED); /* it is in tscDestroyLocalReducer function already */ - return 0; + return TSDB_CODE_SUCCESS; } - SInterpolationInfo *pInterpoInfo = &pLocalReducer->interpolationInfo; - tFilePage * tmpPages = pLocalReducer->pTempBuffer; + tFilePage *tmpBuffer = pLocalReducer->pTempBuffer; - bool prevGroupDone = (!pLocalReducer->discard) && pLocalReducer->hasUnprocessedRow; - - if ((pLocalReducer->numOfBuffer == pLocalReducer->numOfCompleted && !pLocalReducer->hasPrevRow) || - pLocalReducer->pLocalDataSrc[0] == NULL || prevGroupDone) { - /* if interpoType == TSDB_INTERPO_NONE, return directly */ - if (pCmd->interpoType != TSDB_INTERPO_NONE) { - int64_t etime = (pCmd->stime < pCmd->etime) ? pCmd->etime : pCmd->stime; - - etime = taosGetRevisedEndKey(etime, pCmd->order.order, pCmd->nAggTimeInterval, pCmd->intervalTimeUnit); - int32_t rows = taosGetNumOfResultWithInterpo(pInterpoInfo, NULL, 0, pCmd->nAggTimeInterval, etime, - pLocalReducer->resColModel->maxCapacity); - if (rows > 0) { // do interpo - doInterpolateResult(pSql, pLocalReducer, true); - } - } - - /* numOfRows == 0, means no interpolation results are generated yet */ - if (pRes->numOfRows == 0) { - /* local reduce is completed */ - if ((pLocalReducer->numOfBuffer == pLocalReducer->numOfCompleted) && (!pLocalReducer->hasUnprocessedRow)) { - pLocalReducer->status = TSC_LOCALREDUCE_READY; - // set the flag, taos_free_result can release this result. - return 0; - } else { - /* start for process result for a new group */ - savePreGroupNumOfRes(pRes); - setUpForNewGroupRes(pRes, pCmd, pLocalReducer); - } - } else { - pLocalReducer->status = TSC_LOCALREDUCE_READY; - // set the flag, taos_free_result can release this result. - return 0; - } + if (doHandleLastRemainData(pSql)) { + pLocalReducer->status = TSC_LOCALREDUCE_READY; // set the flag, taos_free_result can release this result. + return TSDB_CODE_SUCCESS; } - if (taosHasNoneInterpoPoints(pInterpoInfo)) { - assert(pCmd->interpoType != TSDB_INTERPO_NONE); - - tFilePage *pFinalDataPage = pLocalReducer->pResultBuf; - int64_t etime = *(int64_t *)(pFinalDataPage->data + TSDB_KEYSIZE * (pInterpoInfo->numOfRawDataInRows - 1)); - - int32_t remain = taosNumOfRemainPoints(pInterpoInfo); - TSKEY ekey = taosGetRevisedEndKey(etime, pCmd->order.order, pCmd->nAggTimeInterval, pCmd->intervalTimeUnit); - int32_t rows = taosGetNumOfResultWithInterpo(pInterpoInfo, (TSKEY *)pLocalReducer->pBufForInterpo, remain, - pCmd->nAggTimeInterval, ekey, pLocalReducer->resColModel->maxCapacity); - if (rows > 0) { // do interpo - doInterpolateResult(pSql, pLocalReducer, false); - } - - pLocalReducer->status = TSC_LOCALREDUCE_READY; - // set the flag, taos_free_result can release this result. - return 0; + if (doInterpolationForCurrentGroup(pSql)) { + pLocalReducer->status = TSC_LOCALREDUCE_READY; // set the flag, taos_free_result can release this result. + return TSDB_CODE_SUCCESS; } SLoserTreeInfo *pTree = pLocalReducer->pLoserTree; // clear buffer - handleUnprocessedRow(pLocalReducer, pCmd, tmpPages); + handleUnprocessedRow(pLocalReducer, pCmd, tmpBuffer); tColModel *pModel = pLocalReducer->pDesc->pSchema; while (1) { - _reduce_retrieve: - if (pLocalReducer->numOfBuffer == pLocalReducer->numOfCompleted) { - pRes->numOfRows = 0; + if (isAllSourcesCompleted(pLocalReducer)) { break; } @@ -1194,12 +1297,12 @@ int32_t tscLocalDoReduce(SSqlObj *pSql) { printf("chosen data in pTree[0] = %d\n", pTree->pNode[0].index); #endif assert((pTree->pNode[0].index < pLocalReducer->numOfBuffer) && (pTree->pNode[0].index >= 0) && - tmpPages->numOfElems == 0); + tmpBuffer->numOfElems == 0); // chosen from loser tree - SLocalDataSrc *pOneDataSrc = pLocalReducer->pLocalDataSrc[pTree->pNode[0].index]; + SLocalDataSource *pOneDataSrc = pLocalReducer->pLocalDataSrc[pTree->pNode[0].index]; - tColModelAppend(pModel, tmpPages, pOneDataSrc->filePage.data, pOneDataSrc->rowIdx, 1, + tColModelAppend(pModel, tmpBuffer, pOneDataSrc->filePage.data, pOneDataSrc->rowIdx, 1, pOneDataSrc->pMemBuffer->pColModel->maxCapacity); #if defined(_DEBUG_VIEW) @@ -1207,35 +1310,42 @@ int32_t tscLocalDoReduce(SSqlObj *pSql) { SSrcColumnInfo colInfo[256] = {0}; tscGetSrcColumnInfo(colInfo, pCmd); - tColModelDisplayEx(pModel, tmpPages->data, tmpPages->numOfElems, pModel->maxCapacity, colInfo); + tColModelDisplayEx(pModel, tmpBuffer->data, tmpBuffer->numOfElems, pModel->maxCapacity, colInfo); #endif + if (pLocalReducer->discard) { assert(pLocalReducer->hasUnprocessedRow == false); /* current record belongs to the same group of previous record, need to discard it */ - if (isSameGroupOfPrev(pCmd, pLocalReducer, pLocalReducer->discardData->data, tmpPages)) { - tmpPages->numOfElems = 0; + if (isSameGroup(pCmd, pLocalReducer, pLocalReducer->discardData->data, tmpBuffer)) { + tmpBuffer->numOfElems = 0; pOneDataSrc->rowIdx += 1; - loadDataIntoMemAndAdjustLoserTree(pLocalReducer, pOneDataSrc, pTree); - /* all inputs are exhausted, abort current process */ - if (pLocalReducer->numOfBuffer == pLocalReducer->numOfCompleted) { + adjustLoserTreeFromNewData(pLocalReducer, pOneDataSrc, pTree); + + // all inputs are exhausted, abort current process + if (isAllSourcesCompleted(pLocalReducer)) { break; } - /* since it belongs to the same group, ignore following records */ + // data belongs to the same group needs to be discarded continue; } else { pLocalReducer->discard = false; pLocalReducer->discardData->numOfElems = 0; - savePreGroupNumOfRes(pRes); - setUpForNewGroupRes(pRes, pCmd, pLocalReducer); + if (saveGroupResultInfo(pSql)) { + pLocalReducer->status = TSC_LOCALREDUCE_READY; + return TSDB_CODE_SUCCESS; + } + + resetEnvForNewResultset(pRes, pCmd, pLocalReducer); } } if (pLocalReducer->hasPrevRow) { - if (needToMerge(pCmd, pLocalReducer, tmpPages)) { // belong to the group of the previous row + if (needToMerge(pCmd, pLocalReducer, tmpBuffer)) { + // belong to the group of the previous row, continue process it for (int32_t j = 0; j < pCmd->fieldsInfo.numOfOutputCols; ++j) { SSqlExpr *pExpr = tscSqlExprGet(pCmd, j); tVariantAssign(&pLocalReducer->pCtx[j].param[0], &pExpr->param[0]); @@ -1244,109 +1354,86 @@ int32_t tscLocalDoReduce(SSqlObj *pSql) { } // copy to buffer - savePrevRecord(pLocalReducer, tmpPages); - } else { // reduce the previous is completed, start a new one + savePreviousRow(pLocalReducer, tmpBuffer); + } else { + /* + * current row does not belong to the group of previous row. + * so the processing of previous group is completed. + */ int32_t numOfRes = finalizeRes(pCmd, pLocalReducer); - bool sameGroup = isSameGroupOfPrev(pCmd, pLocalReducer, pLocalReducer->prevRowOfInput, tmpPages); + bool sameGroup = isSameGroup(pCmd, pLocalReducer, pLocalReducer->prevRowOfInput, tmpBuffer); tFilePage *pResBuf = pLocalReducer->pResultBuf; /* - * if the previous group does NOTE generate any result - * (pResBuf->numOfElems == 0), + * if the previous group does NOT generate any result (pResBuf->numOfElems == 0), * continue to process results instead of return results. */ if ((!sameGroup && pResBuf->numOfElems > 0) || (pResBuf->numOfElems == pLocalReducer->resColModel->maxCapacity)) { // does not belong to the same group - assert(pResBuf->numOfElems > 0); - - doGenerateFinalResults(pSql, pLocalReducer, !sameGroup); + bool notSkipped = doGenerateFinalResults(pSql, pLocalReducer, !sameGroup); + // this row needs to discard, since it belongs to the group of previous if (pLocalReducer->discard && sameGroup) { - /* this row needs to discard, since it belongs to the group of previous */ pLocalReducer->hasUnprocessedRow = false; - tmpPages->numOfElems = 0; + tmpBuffer->numOfElems = 0; } else { + // current row does not belongs to the previous group, so it is not be handled yet. pLocalReducer->hasUnprocessedRow = true; } resetOutputBuf(pCmd, pLocalReducer); pOneDataSrc->rowIdx += 1; - /* here we do not check the return value */ - loadDataIntoMemAndAdjustLoserTree(pLocalReducer, pOneDataSrc, pTree); + // here we do not check the return value + adjustLoserTreeFromNewData(pLocalReducer, pOneDataSrc, pTree); assert(pLocalReducer->status == TSC_LOCALREDUCE_IN_PROGRESS); if (pRes->numOfRows == 0) { - handleUnprocessedRow(pLocalReducer, pCmd, tmpPages); + handleUnprocessedRow(pLocalReducer, pCmd, tmpBuffer); if (!sameGroup) { - /* previous group is done, we start a new one by continuing to - * retrieve data */ - savePreGroupNumOfRes(pRes); - setUpForNewGroupRes(pRes, pCmd, pLocalReducer); - } + /* + * previous group is done, prepare for the next group + * If previous group is not skipped, keep it in pRes->numOfGroups + */ + if (notSkipped && saveGroupResultInfo(pSql)) { + pLocalReducer->status = TSC_LOCALREDUCE_READY; + return TSDB_CODE_SUCCESS; + } - goto _reduce_retrieve; + resetEnvForNewResultset(pRes, pCmd, pLocalReducer); + } } else { /* * if next record belongs to a new group, we do not handle this record here. * We start the process in a new round. */ if (sameGroup) { - handleUnprocessedRow(pLocalReducer, pCmd, tmpPages); + handleUnprocessedRow(pLocalReducer, pCmd, tmpBuffer); } } - pLocalReducer->status = TSC_LOCALREDUCE_READY; - // set the flag, taos_free_result can release this result. - return 0; + // current group has no result, + if (pRes->numOfRows == 0) { + continue; + } else { + pLocalReducer->status = TSC_LOCALREDUCE_READY; // set the flag, taos_free_result can release this result. + return TSDB_CODE_SUCCESS; + } } else { // result buffer is not full - for (int32_t k = 0; k < pCmd->fieldsInfo.numOfOutputCols; ++k) { - SSqlExpr *pExpr = tscSqlExprGet(pCmd, k); - - pLocalReducer->pCtx[k].aOutputBuf += pLocalReducer->pCtx[k].outputBytes * numOfRes; - if (pExpr->sqlFuncId == TSDB_FUNC_TOP_DST || pExpr->sqlFuncId == TSDB_FUNC_BOTTOM_DST) { - pLocalReducer->pCtx[k].ptsOutputBuf = - ((char *)pLocalReducer->pCtx[k].ptsOutputBuf + TSDB_KEYSIZE * numOfRes); - } - - /* set the parameters for the SQLFunctionCtx */ - tVariantAssign(&pLocalReducer->pCtx[k].param[0], &pExpr->param[0]); - - aAggs[pExpr->sqlFuncId].init(&pLocalReducer->pCtx[k]); - pLocalReducer->pCtx[k].currentStage = SECONDARY_STAGE_MERGE; - aAggs[pExpr->sqlFuncId].distSecondaryMergeFunc(&pLocalReducer->pCtx[k]); - } - - savePrevRecord(pLocalReducer, tmpPages); + doMergeWithPrevRows(pSql, numOfRes); + savePreviousRow(pLocalReducer, tmpBuffer); } } - } else { // put to previous input row for comparision - for (int32_t j = 0; j < pCmd->fieldsInfo.numOfOutputCols; ++j) { - SSqlExpr *pExpr = tscSqlExprGet(pCmd, j); - - tVariantAssign(&pLocalReducer->pCtx[j].param[0], &pExpr->param[0]); - pLocalReducer->pCtx[j].numOfIteratedElems = 0; - pLocalReducer->pCtx[j].currentStage = 0; - - aAggs[pExpr->sqlFuncId].init(&pLocalReducer->pCtx[j]); - pLocalReducer->pCtx[j].currentStage = SECONDARY_STAGE_MERGE; - - aAggs[pExpr->sqlFuncId].distSecondaryMergeFunc(&pLocalReducer->pCtx[j]); - } - - // copy to buffer - savePrevRecord(pLocalReducer, tmpPages); + } else { + doExecuteSecondaryMerge(pSql); + savePreviousRow(pLocalReducer, tmpBuffer); // copy the processed row to buffer } pOneDataSrc->rowIdx += 1; - - loadDataIntoMemAndAdjustLoserTree(pLocalReducer, pOneDataSrc, pTree); - if (pLocalReducer->numOfCompleted == pLocalReducer->numOfBuffer) { - break; - } + adjustLoserTreeFromNewData(pLocalReducer, pOneDataSrc, pTree); } if (pLocalReducer->hasPrevRow) { @@ -1358,8 +1445,7 @@ int32_t tscLocalDoReduce(SSqlObj *pSql) { } assert(pLocalReducer->status == TSC_LOCALREDUCE_IN_PROGRESS && pRes->row == 0); - pLocalReducer->status = TSC_LOCALREDUCE_READY; - // set the flag, taos_free_result can release this result. + pLocalReducer->status = TSC_LOCALREDUCE_READY; // set the flag, taos_free_result can release this result. return TSDB_CODE_SUCCESS; } @@ -1378,7 +1464,8 @@ void tscInitResObjForLocalQuery(SSqlObj *pObj, int32_t numOfRes, int32_t rowLen) pRes->pLocalReducer = (SLocalReducer *)calloc(1, sizeof(SLocalReducer)); /* - * we need one additional byte space the sprintf function needs one additional space to put '\0' at the end of string + * we need one additional byte space + * the sprintf function needs one additional space to put '\0' at the end of string */ size_t allocSize = numOfRes * rowLen + sizeof(tFilePage) + 1; pRes->pLocalReducer->pResultBuf = (tFilePage *)calloc(1, allocSize); diff --git a/src/client/src/tscServer.c b/src/client/src/tscServer.c index ba645c53b2..9f58fe1390 100644 --- a/src/client/src/tscServer.c +++ b/src/client/src/tscServer.c @@ -287,14 +287,20 @@ void *tscProcessMsgFromServer(char *msg, void *ahandle, void *thandle) { pSql->thandle = NULL; taosAddConnIntoCache(tscConnCache, thandle, pSql->ip, pSql->vnode, pObj->user); - if (UTIL_METER_IS_METRIC(pCmd) && - (pMsg->content[0] == TSDB_CODE_INVALID_SESSION_ID || pMsg->content[0] == TSDB_CODE_NOT_ACTIVE_SESSION)) { + if (UTIL_METER_IS_METRIC(pCmd) && pMsg->content[0] == TSDB_CODE_NOT_ACTIVE_SESSION) { /* * for metric query, in case of any meter missing during query, sub-query of metric query will failed, * causing metric query failed, and return TSDB_CODE_METRICMETA_EXPIRED code to app */ tscTrace("%p invalid meters id cause metric query failed, code:%d", pSql, pMsg->content[0]); code = TSDB_CODE_METRICMETA_EXPIRED; + } else if ((pCmd->command == TSDB_SQL_INSERT || pCmd->command == TSDB_SQL_SELECT) && + pMsg->content[0] == TSDB_CODE_INVALID_SESSION_ID) { + /* + * session id is invalid(e.g., less than 0 or larger than maximum session per + * vnode) in submit/query msg, no retry + */ + code = TSDB_CODE_INVALID_QUERY_MSG; } else if (pCmd->command == TSDB_SQL_CONNECT) { code = TSDB_CODE_NETWORK_UNAVAIL; } else if (pCmd->command == TSDB_SQL_HB) { @@ -358,14 +364,17 @@ void *tscProcessMsgFromServer(char *msg, void *ahandle, void *thandle) { pRes->code = TSDB_CODE_SUCCESS; } - tscTrace("%p cmd:%d code:%d rsp len:%d", pSql, pCmd->command, pRes->code, pRes->rspLen); - /* * There is not response callback function for submit response. * The actual inserted number of points is the first number. */ if (pMsg->msgType == TSDB_MSG_TYPE_SUBMIT_RSP) { pRes->numOfRows += *(int32_t *)pRes->pRsp; + + tscTrace("%p cmd:%d code:%d, inserted rows:%d, rsp len:%d", pSql, pCmd->command, pRes->code, + *(int32_t *)pRes->pRsp, pRes->rspLen); + } else { + tscTrace("%p cmd:%d code:%d rsp len:%d", pSql, pCmd->command, pRes->code, pRes->rspLen); } } @@ -421,7 +430,7 @@ void *tscProcessMsgFromServer(char *msg, void *ahandle, void *thandle) { return ahandle; } -static SSqlObj* tscCreateSqlObjForSubquery(SSqlObj *pSql, SRetrieveSupport *trsupport, SSqlObj* pOld); +static SSqlObj* tscCreateSqlObjForSubquery(SSqlObj *pSql, SRetrieveSupport *trsupport, SSqlObj* prevSqlObj); static int tscLaunchMetricSubQueries(SSqlObj *pSql); int tscProcessSql(SSqlObj *pSql) { @@ -430,12 +439,6 @@ int tscProcessSql(SSqlObj *pSql) { tscTrace("%p SQL cmd:%d will be processed, name:%s", pSql, pSql->cmd.command, pSql->cmd.name); - // whether don't judge 'isInsertFromFile' ? - if (pSql->cmd.command == TSDB_SQL_INSERT && pCmd->isInsertFromFile == 1) { - // pCmd->isInsertFromFile = 0; // lihui: can not clear the flag - return 0; - } - pSql->retry = 0; if (pSql->cmd.command < TSDB_SQL_MGMT) { pSql->maxRetry = 2; @@ -595,7 +598,6 @@ int tscLaunchMetricSubQueries(SSqlObj *pSql) { SSqlObj *pNew = tscCreateSqlObjForSubquery(pSql, trs, NULL); tscTrace("%p sub:%p launch subquery.orderOfSub:%d", pSql, pNew, pNew->cmd.vnodeIdx); - tscProcessSql(pNew); } @@ -665,7 +667,6 @@ static void tscHandleSubRetrievalError(SRetrieveSupport *trsupport, SSqlObj *pSq tscError("%p sub:%p abort further retrieval due to other queries failure,orderOfSub:%d,code:%d", pPObj, pSql, idx, *trsupport->code); } else { - if (trsupport->numOfRetry++ < MAX_NUM_OF_SUBQUERY_RETRY && *(trsupport->code) == TSDB_CODE_SUCCESS) { /* * current query failed, and the retry count is less than the available count, @@ -675,11 +676,12 @@ static void tscHandleSubRetrievalError(SRetrieveSupport *trsupport, SSqlObj *pSq // clear local saved number of results trsupport->localBuffer->numOfElems = 0; + pthread_mutex_unlock(&trsupport->queryMutex); SSqlObj *pNew = tscCreateSqlObjForSubquery(trsupport->pParentSqlObj, trsupport, pSql); tscTrace("%p sub:%p retrieve failed, code:%d, orderOfSub:%d, retry:%d, new SqlObj:%p", - trsupport->pParentSqlObj, pSql, numOfRows, idx, trsupport->numOfRetry, pNew); + trsupport->pParentSqlObj, pSql, numOfRows, idx, trsupport->numOfRetry, pNew); tscProcessSql(pNew); return; @@ -689,7 +691,6 @@ static void tscHandleSubRetrievalError(SRetrieveSupport *trsupport, SSqlObj *pSq tscError("%p sub:%p retrieve failed,code:%d,orderOfSub:%d failed.no more retry,set global code:%d", pPObj, pSql, numOfRows, idx, *trsupport->code); } - } if (__sync_add_and_fetch_32(trsupport->numOfFinished, 1) < trsupport->numOfVnodes) { @@ -778,7 +779,7 @@ void tscRetrieveFromVnodeCallBack(void *param, TAOS_RES *tres, int numOfRows) { tscTrace("%p sub:%p all data retrieved from ip:%u,vid:%d, numOfRows:%d, orderOfSub:%d", pPObj, pSql, pSvd->ip, pSvd->vnode, numOfRowsFromVnode, idx); - tColModelCompress(pDesc->pSchema, trsupport->localBuffer, pDesc->pSchema->maxCapacity); + tColModelCompact(pDesc->pSchema, trsupport->localBuffer, pDesc->pSchema->maxCapacity); #ifdef _DEBUG_VIEW printf("%ld rows data flushed to disk:\n", trsupport->localBuffer->numOfElems); @@ -877,7 +878,7 @@ void tscKillMetricQuery(SSqlObj *pSql) { tscTrace("%p metric query is cancelled", pSql); } -static SSqlObj* tscCreateSqlObjForSubquery(SSqlObj *pSql, SRetrieveSupport *trsupport, SSqlObj* prevSqlObj) { +SSqlObj* tscCreateSqlObjForSubquery(SSqlObj *pSql, SRetrieveSupport *trsupport, SSqlObj* prevSqlObj) { SSqlCmd *pCmd = &pSql->cmd; SSqlObj *pNew = (SSqlObj *)calloc(1, sizeof(SSqlObj)); @@ -1032,8 +1033,6 @@ int tscBuildSubmitMsg(SSqlObj *pSql) { pShellMsg->vnode = htons(pMeterMeta->vpeerDesc[pMeterMeta->index].vnode); pShellMsg->numOfSid = htonl(pSql->cmd.count); /* number of meters to be inserted */ - pMsg += sizeof(SShellSubmitMsg); - /* * pSql->cmd.payloadLen is set during parse sql routine, so we do not use it here */ @@ -2264,8 +2263,6 @@ int tscBuildMetricMetaMsg(SSqlObj *pSql) { SSqlGroupbyExpr *pGroupby = &pCmd->groupbyExpr; - pMetaMsg->limit = htobe64(pCmd->glimit.limit); - pMetaMsg->offset = htobe64(pCmd->glimit.offset); pMetaMsg->numOfTags = htons(pCmd->numOfReqTags); pMetaMsg->numOfGroupbyCols = htons(pGroupby->numOfGroupbyCols); @@ -2750,7 +2747,6 @@ static int32_t tscDoGetMeterMeta(SSqlObj *pSql, char *meterId) { } else { pNew->fp = tscMeterMetaCallBack; pNew->param = pSql; - pNew->sqlstr = strdup(pSql->sqlstr); code = tscProcessSql(pNew); diff --git a/src/client/src/tscSql.c b/src/client/src/tscSql.c index a58238624f..d55a765d3b 100644 --- a/src/client/src/tscSql.c +++ b/src/client/src/tscSql.c @@ -72,7 +72,7 @@ TAOS *taos_connect_imp(char *ip, char *user, char *pass, char *db, int port, voi pObj->signature = pObj; strncpy(pObj->user, user, TSDB_USER_LEN); - taosEncryptPass(pass, strlen(pass), pObj->pass); + taosEncryptPass((uint8_t *)pass, strlen(pass), pObj->pass); pObj->mgmtPort = port ? port : tsMgmtShellPort; if (db) { diff --git a/src/client/src/tscStream.c b/src/client/src/tscStream.c index 84741f098e..ea3d1e86e5 100644 --- a/src/client/src/tscStream.c +++ b/src/client/src/tscStream.c @@ -145,13 +145,14 @@ static void tscProcessStreamQueryCallback(void *param, TAOS_RES *tres, int numOf static void tscSetTimestampForRes(SSqlStream *pStream, SSqlObj *pSql, int32_t numOfRows) { SSqlRes *pRes = &pSql->res; - int64_t timestamp = *(int64_t *)pRes->data; - if (timestamp != pStream->stime) { + int64_t timestamp = *(int64_t *)pRes->data; + int64_t actualTimestamp = pStream->stime - pStream->interval; + + if (timestamp != actualTimestamp) { // reset the timestamp of each agg point by using start time of each interval - *((int64_t *)pRes->data) = pStream->stime - pStream->interval; - tscWarn("%p stream:%p, timestamp of points is:%lld, reset to %lld", pSql, pStream, timestamp, - pStream->stime - pStream->interval); + *((int64_t *)pRes->data) = actualTimestamp; + tscWarn("%p stream:%p, timestamp of points is:%lld, reset to %lld", pSql, pStream, timestamp, actualTimestamp); } } @@ -397,7 +398,7 @@ static int64_t tscGetStreamStartTimestamp(SSqlObj *pSql, SSqlStream *pStream, in } else { // timewindow based aggregation stream if (stime == 0) { // no data in meter till now stime = ((int64_t)taosGetTimestamp(pStream->precision) / pStream->interval) * pStream->interval; - tscWarn("%p stream:%p, last timestamp:0, reset to:%lld", pSql, pStream, stime, stime); + tscWarn("%p stream:%p, last timestamp:0, reset to:%lld", pSql, pStream, stime); } else { int64_t newStime = (stime / pStream->interval) * pStream->interval; if (newStime != stime) { @@ -435,13 +436,25 @@ static int64_t tscGetLaunchTimestamp(const SSqlStream *pStream) { return (pStream->precision == TSDB_TIME_PRECISION_MICRO) ? timer / 1000L : timer; } +static void setErrorInfo(STscObj* pObj, int32_t code, char* info) { + if (pObj == NULL) { + return; + } + + SSqlCmd* pCmd = &pObj->pSql->cmd; + + pObj->pSql->res.code = code; + strncpy(pCmd->payload, info, pCmd->payloadLen); +} + TAOS_STREAM *taos_open_stream(TAOS *taos, char *sqlstr, void (*fp)(void *param, TAOS_RES *, TAOS_ROW row), int64_t stime, void *param, void (*callback)(void *)) { STscObj *pObj = (STscObj *)taos; if (pObj == NULL || pObj->signature != pObj) return NULL; SSqlObj *pSql = (SSqlObj *)calloc(1, sizeof(SSqlObj)); - if (pSql == NULL) { // todo set corect error msg + if (pSql == NULL) { + setErrorInfo(pObj, TSDB_CODE_CLI_OUT_OF_MEMORY, NULL); return NULL; } @@ -451,22 +464,31 @@ TAOS_STREAM *taos_open_stream(TAOS *taos, char *sqlstr, void (*fp)(void *param, SSqlRes *pRes = &pSql->res; tscAllocPayloadWithSize(pCmd, TSDB_DEFAULT_PAYLOAD_SIZE); - pSql->sqlstr = calloc(1, strlen(sqlstr) + 1); - if (pSql->sqlstr == NULL) { // todo set corect error msg + pSql->sqlstr = strdup(sqlstr); + if (pSql->sqlstr == NULL) { + setErrorInfo(pObj, TSDB_CODE_CLI_OUT_OF_MEMORY, NULL); + tfree(pSql); return NULL; } - strcpy(pSql->sqlstr, sqlstr); sem_init(&pSql->rspSem, 0, 0); sem_init(&pSql->emptyRspSem, 0, 1); SSqlInfo SQLInfo = {0}; tSQLParse(&SQLInfo, pSql->sqlstr); + + tscCleanSqlCmd(&pSql->cmd); + tscAllocPayloadWithSize(&pSql->cmd, TSDB_DEFAULT_PAYLOAD_SIZE); + + //todo refactor later + pSql->cmd.count = 1; pRes->code = tscToSQLCmd(pSql, &SQLInfo); SQLInfoDestroy(&SQLInfo); if (pRes->code != TSDB_CODE_SUCCESS) { + setErrorInfo(pObj, pRes->code, pCmd->payload); + tscError("%p open stream failed, sql:%s, reason:%s, code:%d", pSql, sqlstr, pCmd->payload, pRes->code); tscFreeSqlObj(pSql); return NULL; @@ -474,6 +496,8 @@ TAOS_STREAM *taos_open_stream(TAOS *taos, char *sqlstr, void (*fp)(void *param, SSqlStream *pStream = (SSqlStream *)calloc(1, sizeof(SSqlStream)); if (pStream == NULL) { + setErrorInfo(pObj, TSDB_CODE_CLI_OUT_OF_MEMORY, NULL); + tscError("%p open stream failed, sql:%s, reason:%s, code:%d", pSql, sqlstr, pCmd->payload, pRes->code); tscFreeSqlObj(pSql); return NULL; diff --git a/src/client/src/tscUtil.c b/src/client/src/tscUtil.c index ea70292435..4c9c180f6a 100644 --- a/src/client/src/tscUtil.c +++ b/src/client/src/tscUtil.c @@ -17,6 +17,7 @@ #include #include +#include "ihash.h" #include "taosmsg.h" #include "tcache.h" #include "tkey.h" @@ -31,9 +32,10 @@ /* * the detailed information regarding metric meta key is: - * fullmetername + '.' + querycond + '.' + [tagId1, tagId2,...] + '.' + group_orderType + '.' + limit + '.' + offset + * fullmetername + '.' + querycond + '.' + [tagId1, tagId2,...] + '.' + group_orderType + * * if querycond is null, its format is: - * fullmetername + '.' + '(nil)' + '.' + [tagId1, tagId2,...] + '.' + group_orderType + '.' + limit + '.' + offset + * fullmetername + '.' + '(nil)' + '.' + [tagId1, tagId2,...] + '.' + group_orderType */ void tscGetMetricMetaCacheKey(SSqlCmd* pCmd, char* keyStr) { char* pTagCondStr = NULL; @@ -60,8 +62,7 @@ void tscGetMetricMetaCacheKey(SSqlCmd* pCmd, char* keyStr) { pTagCondStr = strdup(tsGetMetricQueryCondPos(&pCmd->tagCond)); } - int32_t keyLen = sprintf(keyStr, "%s.%s.[%s].%d.%lld.%lld", pCmd->name, pTagCondStr, tagIdBuf, - pCmd->groupbyExpr.orderType, pCmd->glimit.limit, pCmd->glimit.offset); + int32_t keyLen = sprintf(keyStr, "%s.%s.[%s].%d", pCmd->name, pTagCondStr, tagIdBuf, pCmd->groupbyExpr.orderType); free(pTagCondStr); assert(keyLen <= TSDB_MAX_TAGS_LEN); @@ -142,8 +143,7 @@ bool tscProjectionQueryOnMetric(SSqlObj* pSql) { /* * In following cases, return false for project query on metric - * 1. failed to get metermeta from server; 2. not a metric; 3. limit 0; 4. - * show query, instead of a select query + * 1. failed to get metermeta from server; 2. not a metric; 3. limit 0; 4. show query, instead of a select query */ if (pCmd->pMeterMeta == NULL || !UTIL_METER_IS_METRIC(pCmd) || pCmd->command == TSDB_SQL_RETRIEVE_EMPTY_RESULT || pCmd->exprsInfo.numOfExprs == 0) { @@ -252,7 +252,7 @@ void tscDestroyResPointerInfo(SSqlRes* pRes) { } void tscfreeSqlCmdData(SSqlCmd* pCmd) { - tscDestroyBlockArrayList(&pCmd->pDataBlocks); + pCmd->pDataBlocks = tscDestroyBlockArrayList(pCmd->pDataBlocks); tscTagCondRelease(&pCmd->tagCond); tscClearFieldInfo(pCmd); @@ -334,20 +334,22 @@ void tscFreeSqlObj(SSqlObj* pSql) { free(pSql); } -SInsertedDataBlocks* tscCreateDataBlock(int32_t size) { - SInsertedDataBlocks* dataBuf = (SInsertedDataBlocks*)calloc(1, sizeof(SInsertedDataBlocks)); - dataBuf->nAllocSize = (uint32_t) size; +STableDataBlocks* tscCreateDataBlock(int32_t size) { + STableDataBlocks* dataBuf = (STableDataBlocks*)calloc(1, sizeof(STableDataBlocks)); + dataBuf->nAllocSize = (uint32_t)size; dataBuf->pData = calloc(1, dataBuf->nAllocSize); + dataBuf->ordered = true; + dataBuf->prevTS = INT64_MIN; return dataBuf; } -void tscDestroyDataBlock(SInsertedDataBlocks** pDataBlock) { - if (*pDataBlock == NULL) { +void tscDestroyDataBlock(STableDataBlocks* pDataBlock) { + if (pDataBlock == NULL) { return; } - tfree((*pDataBlock)->pData); - tfree(*pDataBlock); + tfree(pDataBlock->pData); + tfree(pDataBlock); } SDataBlockList* tscCreateBlockArrayList() { @@ -360,29 +362,31 @@ SDataBlockList* tscCreateBlockArrayList() { return pDataBlockArrayList; } -void tscDestroyBlockArrayList(SDataBlockList** pList) { - if (*pList == NULL) { - return; +void* tscDestroyBlockArrayList(SDataBlockList* pList) { + if (pList == NULL) { + return NULL; } - for (int32_t i = 0; i < (*pList)->nSize; i++) { - tscDestroyDataBlock(&(*pList)->pData[i]); + for (int32_t i = 0; i < pList->nSize; i++) { + tscDestroyDataBlock(pList->pData[i]); } - tfree((*pList)->pData); - tfree(*pList); + tfree(pList->pData); + tfree(pList); + + return NULL; } -int32_t tscCopyDataBlockToPayload(SSqlObj* pSql, SInsertedDataBlocks* pDataBlock) { +int32_t tscCopyDataBlockToPayload(SSqlObj* pSql, STableDataBlocks* pDataBlock) { SSqlCmd* pCmd = &pSql->cmd; pCmd->count = pDataBlock->numOfMeters; - strcpy(pCmd->name, pDataBlock->meterId); + strncpy(pCmd->name, pDataBlock->meterId, TSDB_METER_ID_LEN); tscAllocPayloadWithSize(pCmd, pDataBlock->nAllocSize); memcpy(pCmd->payload, pDataBlock->pData, pDataBlock->nAllocSize); - /* set the message length */ + // set the message length pCmd->payloadLen = pDataBlock->nAllocSize; return tscGetMeterMeta(pSql, pCmd->name); } @@ -390,12 +394,91 @@ int32_t tscCopyDataBlockToPayload(SSqlObj* pSql, SInsertedDataBlocks* pDataBlock void tscFreeUnusedDataBlocks(SDataBlockList* pList) { /* release additional memory consumption */ for (int32_t i = 0; i < pList->nSize; ++i) { - SInsertedDataBlocks* pDataBlock = pList->pData[i]; - pDataBlock->pData = realloc(pDataBlock->pData, (size_t) pDataBlock->size); - pDataBlock->nAllocSize = (uint32_t) pDataBlock->size; + STableDataBlocks* pDataBlock = pList->pData[i]; + pDataBlock->pData = realloc(pDataBlock->pData, pDataBlock->size); + pDataBlock->nAllocSize = (uint32_t)pDataBlock->size; } } +STableDataBlocks* tscCreateDataBlockEx(size_t size, int32_t rowSize, int32_t startOffset, char* name) { + STableDataBlocks* dataBuf = tscCreateDataBlock(size); + + dataBuf->rowSize = rowSize; + dataBuf->size = startOffset; + strncpy(dataBuf->meterId, name, TSDB_METER_ID_LEN); + return dataBuf; +} + +STableDataBlocks* tscGetDataBlockFromList(void* pHashList, SDataBlockList* pDataBlockList, int64_t id, int32_t size, + int32_t startOffset, int32_t rowSize, char* tableId) { + STableDataBlocks* dataBuf = NULL; + + STableDataBlocks** t1 = (STableDataBlocks**)taosGetIntHashData(pHashList, id); + if (t1 != NULL) { + dataBuf = *t1; + } + + if (dataBuf == NULL) { + dataBuf = tscCreateDataBlockEx((size_t)size, rowSize, startOffset, tableId); + dataBuf = *(STableDataBlocks**)taosAddIntHash(pHashList, id, (char*)&dataBuf); + tscAppendDataBlock(pDataBlockList, dataBuf); + } + + return dataBuf; +} + +void tscMergeTableDataBlocks(SSqlObj* pSql, SDataBlockList* pTableDataBlockList) { + SSqlCmd* pCmd = &pSql->cmd; + void* pVnodeDataBlockHashList = taosInitIntHash(8, sizeof(void*), taosHashInt); + SDataBlockList* pVnodeDataBlockList = tscCreateBlockArrayList(); + + for (int32_t i = 0; i < pTableDataBlockList->nSize; ++i) { + STableDataBlocks* pOneTableBlock = pTableDataBlockList->pData[i]; + STableDataBlocks* dataBuf = + tscGetDataBlockFromList(pVnodeDataBlockHashList, pVnodeDataBlockList, pOneTableBlock->vgid, TSDB_PAYLOAD_SIZE, + tsInsertHeadSize, 0, pOneTableBlock->meterId); + + int64_t destSize = dataBuf->size + pOneTableBlock->size; + if (dataBuf->nAllocSize < destSize) { + while (dataBuf->nAllocSize < destSize) { + dataBuf->nAllocSize = dataBuf->nAllocSize * 1.5; + } + + char* tmp = realloc(dataBuf->pData, dataBuf->nAllocSize); + if (tmp != NULL) { + dataBuf->pData = tmp; + memset(dataBuf->pData + dataBuf->size, 0, dataBuf->nAllocSize - dataBuf->size); + } else { + // to do handle error + } + } + + SShellSubmitBlock* pBlocks = (SShellSubmitBlock*)pOneTableBlock->pData; + sortRemoveDuplicates(pOneTableBlock); + + tscTrace("%p meterId:%s, sid:%d, rows:%d, sversion:%d", pSql, pOneTableBlock->meterId, pBlocks->sid, + pBlocks->numOfRows, pBlocks->sversion); + + pBlocks->sid = htonl(pBlocks->sid); + pBlocks->uid = htobe64(pBlocks->uid); + pBlocks->sversion = htonl(pBlocks->sversion); + pBlocks->numOfRows = htons(pBlocks->numOfRows); + + memcpy(dataBuf->pData + dataBuf->size, pOneTableBlock->pData, pOneTableBlock->size); + + dataBuf->size += pOneTableBlock->size; + dataBuf->numOfMeters += 1; + } + + tscDestroyBlockArrayList(pTableDataBlockList); + + // free the table data blocks; + pCmd->pDataBlocks = pVnodeDataBlockList; + + tscFreeUnusedDataBlocks(pCmd->pDataBlocks); + taosCleanUpIntHash(pVnodeDataBlockHashList); +} + void tscCloseTscObj(STscObj* pObj) { pObj->signature = NULL; SSqlObj* pSql = pObj->pSql; @@ -802,7 +885,7 @@ static int32_t validateQuoteToken(SSQLToken* pToken) { if (pToken->type == TK_STRING) { return tscValidateName(pToken); - } + } if (k != pToken->n || pToken->type != TK_ID) { return TSDB_CODE_INVALID_SQL; @@ -821,17 +904,20 @@ int32_t tscValidateName(SSQLToken* pToken) { pToken->n = strdequote(pToken->z); strtrim(pToken->z); pToken->n = (uint32_t)strlen(pToken->z); - int len = tSQLGetToken(pToken->z, &pToken->type); - if (len == pToken->n){ + + int len = tSQLGetToken(pToken->z, &pToken->type); + + // single token, validate it + if (len == pToken->n) { return validateQuoteToken(pToken); - } - else { - sep = strnchrNoquote(pToken->z, TS_PATH_DELIMITER[0], pToken->n); - if (sep == NULL) { - return TSDB_CODE_INVALID_SQL; - } + } else { + sep = strnchrNoquote(pToken->z, TS_PATH_DELIMITER[0], pToken->n); + if (sep == NULL) { + return TSDB_CODE_INVALID_SQL; + } + return tscValidateName(pToken); - } + } } else { if (isNumber(pToken)) { return TSDB_CODE_INVALID_SQL; @@ -843,7 +929,7 @@ int32_t tscValidateName(SSQLToken* pToken) { if (pToken->type == TK_SPACE) { strtrim(pToken->z); - pToken->n = (uint32_t)strlen(pToken->z); + pToken->n = (uint32_t)strlen(pToken->z); } pToken->n = tSQLGetToken(pToken->z, &pToken->type); @@ -965,8 +1051,7 @@ void tscSetFreeHeatBeat(STscObj* pObj) { SSqlObj* pHeatBeat = pObj->pHb; assert(pHeatBeat == pHeatBeat->signature); - pHeatBeat->cmd.type = 1; // to denote the heart-beat timer close connection - // and free all allocated resources + pHeatBeat->cmd.type = 1; // to denote the heart-beat timer close connection and free all allocated resources } bool tscShouldFreeHeatBeat(SSqlObj* pHb) { @@ -1052,7 +1137,6 @@ void tscDoQuery(SSqlObj* pSql) { if (pCmd->command > TSDB_SQL_LOCAL) { tscProcessLocalCmd(pSql); } else { - // add to sql list, so that the show queries could get the query info if (pCmd->command == TSDB_SQL_SELECT) { tscAddIntoSqlList(pSql); } @@ -1061,18 +1145,19 @@ void tscDoQuery(SSqlObj* pSql) { pSql->cmd.vnodeIdx += 1; } - if (pSql->fp == NULL) { - if (0 == pCmd->isInsertFromFile) { - tscProcessSql(pSql); - tscProcessMultiVnodesInsert(pSql); // handle the multi-vnode insertion - } else if (1 == pCmd->isInsertFromFile) { - tscProcessMultiVnodesInsertForFile(pSql); - } else { - assert(false); - } - } else { - tscProcessSql(pSql); - } + void* fp = pSql->fp; + if (pCmd->isInsertFromFile == 1) { + tscProcessMultiVnodesInsertForFile(pSql); + } else { + // pSql may be released in this function if it is a async insertion. + tscProcessSql(pSql); + + // handle the multi-vnode insertion for sync model + if (fp == NULL) { + assert(pSql->signature == pSql); + tscProcessMultiVnodesInsert(pSql); + } + } } } diff --git a/src/connector/jdbc/pom.xml b/src/connector/jdbc/pom.xml index 05989178c7..d29700fe8a 100755 --- a/src/connector/jdbc/pom.xml +++ b/src/connector/jdbc/pom.xml @@ -3,7 +3,7 @@ 4.0.0 com.taosdata.jdbc taos-jdbcdriver - 1.0.0 + 1.0.1 JDBCDriver TDengine JDBC Driver diff --git a/src/connector/jdbc/src/main/java/com/taosdata/jdbc/TSDBConstants.java b/src/connector/jdbc/src/main/java/com/taosdata/jdbc/TSDBConstants.java index ba0c6d939e..3940e80930 100644 --- a/src/connector/jdbc/src/main/java/com/taosdata/jdbc/TSDBConstants.java +++ b/src/connector/jdbc/src/main/java/com/taosdata/jdbc/TSDBConstants.java @@ -24,6 +24,8 @@ public abstract class TSDBConstants { public static final String INVALID_VARIABLES = "invalid variables"; public static Map DATATYPE_MAP = null; + public static final long JNI_NULL_POINTER = 0L; + public static final int JNI_SUCCESS = 0; public static final int JNI_TDENGINE_ERROR = -1; public static final int JNI_CONNECTION_NULL = -2; diff --git a/src/connector/jdbc/src/main/java/com/taosdata/jdbc/TSDBJNIConnector.java b/src/connector/jdbc/src/main/java/com/taosdata/jdbc/TSDBJNIConnector.java index b1c3891424..6f305cc2c7 100755 --- a/src/connector/jdbc/src/main/java/com/taosdata/jdbc/TSDBJNIConnector.java +++ b/src/connector/jdbc/src/main/java/com/taosdata/jdbc/TSDBJNIConnector.java @@ -19,7 +19,6 @@ import java.sql.SQLWarning; import java.util.List; public class TSDBJNIConnector { - static final long INVALID_CONNECTION_POINTER_VALUE = 0l; static volatile Boolean isInitialized = false; static { @@ -29,7 +28,12 @@ public class TSDBJNIConnector { /** * Connection pointer used in C */ - private long taos = INVALID_CONNECTION_POINTER_VALUE; + private long taos = TSDBConstants.JNI_NULL_POINTER; + + /** + * Result set pointer for the current connection + */ + private long taosResultSetPointer = TSDBConstants.JNI_NULL_POINTER; /** * result set status in current connection @@ -41,7 +45,7 @@ public class TSDBJNIConnector { * Whether the connection is closed */ public boolean isClosed() { - return this.taos == INVALID_CONNECTION_POINTER_VALUE; + return this.taos == TSDBConstants.JNI_NULL_POINTER; } /** @@ -86,13 +90,13 @@ public class TSDBJNIConnector { * @throws SQLException */ public boolean connect(String host, int port, String dbName, String user, String password) throws SQLException { - if (this.taos != INVALID_CONNECTION_POINTER_VALUE) { + if (this.taos != TSDBConstants.JNI_NULL_POINTER) { this.closeConnectionImp(this.taos); - this.taos = INVALID_CONNECTION_POINTER_VALUE; + this.taos = TSDBConstants.JNI_NULL_POINTER; } this.taos = this.connectImp(host, port, dbName, user, password); - if (this.taos == INVALID_CONNECTION_POINTER_VALUE) { + if (this.taos == TSDBConstants.JNI_NULL_POINTER) { throw new SQLException(TSDBConstants.WrapErrMsg(this.getErrMsg()), "", this.getErrCode()); } @@ -108,13 +112,7 @@ public class TSDBJNIConnector { */ public int executeQuery(String sql) throws SQLException { if (!this.isResultsetClosed) { - //throw new RuntimeException(TSDBConstants.WrapErrMsg("Connection already has an open result set")); - long resultSetPointer = this.getResultSet(); - if (resultSetPointer == TSDBConstants.JNI_CONNECTION_NULL) { - //do nothing - } else { - this.freeResultSet(resultSetPointer); - } + freeResultSet(taosResultSetPointer); } int code; @@ -133,7 +131,14 @@ public class TSDBJNIConnector { throw new SQLException(TSDBConstants.FixErrMsg(code), "", this.getErrCode()); } } - + + // Try retrieving result set for the executed SQLusing the current connection pointer. If the executed + // SQL is a DML/DDL which doesn't return a result set, then taosResultSetPointer should be 0L. Otherwise, + // taosResultSetPointer should be a non-zero value. + taosResultSetPointer = this.getResultSetImp(this.taos); + if (taosResultSetPointer != TSDBConstants.JNI_NULL_POINTER) { + isResultsetClosed = false; + } return code; } @@ -162,8 +167,7 @@ public class TSDBJNIConnector { * Each connection should have a single open result set at a time */ public long getResultSet() { - long res = this.getResultSetImp(this.taos); - return res; + return taosResultSetPointer; } private native long getResultSetImp(long connection); @@ -172,11 +176,31 @@ public class TSDBJNIConnector { * Free resultset operation from C to release resultset pointer by JNI */ public int freeResultSet(long result) { - int res = this.freeResultSetImp(this.taos, result); - this.isResultsetClosed = true; // reset resultSetPointer to 0 after freeResultSetImp() return - return res; + int res = TSDBConstants.JNI_SUCCESS; + if (result != taosResultSetPointer && taosResultSetPointer != TSDBConstants.JNI_NULL_POINTER) { + throw new RuntimeException("Invalid result set pointer"); + } else if (taosResultSetPointer != TSDBConstants.JNI_NULL_POINTER){ + res = this.freeResultSetImp(this.taos, result); + isResultsetClosed = true; // reset resultSetPointer to 0 after freeResultSetImp() return + taosResultSetPointer = TSDBConstants.JNI_NULL_POINTER; + } + return res; } + /** + * Close the open result set which is associated to the current connection. If the result set is already + * closed, return 0 for success. + * @return + */ + public int freeResultSet() { + int resCode = TSDBConstants.JNI_SUCCESS; + if (!isResultsetClosed) { + resCode = this.freeResultSetImp(this.taos, this.taosResultSetPointer); + taosResultSetPointer = TSDBConstants.JNI_NULL_POINTER; + } + return resCode; + } + private native int freeResultSetImp(long connection, long result); /** @@ -220,7 +244,7 @@ public class TSDBJNIConnector { if (code < 0) { throw new SQLException(TSDBConstants.FixErrMsg(code), "", this.getErrCode()); } else if (code == 0){ - this.taos = INVALID_CONNECTION_POINTER_VALUE; + this.taos = TSDBConstants.JNI_NULL_POINTER; } else { throw new SQLException("Undefined error code returned by TDengine when closing a connection"); } diff --git a/src/connector/jdbc/src/main/java/com/taosdata/jdbc/TSDBPreparedStatement.java b/src/connector/jdbc/src/main/java/com/taosdata/jdbc/TSDBPreparedStatement.java index d613c252f0..bd45fbfb7a 100644 --- a/src/connector/jdbc/src/main/java/com/taosdata/jdbc/TSDBPreparedStatement.java +++ b/src/connector/jdbc/src/main/java/com/taosdata/jdbc/TSDBPreparedStatement.java @@ -244,7 +244,7 @@ public class TSDBPreparedStatement extends TSDBStatement implements PreparedStat @Override public boolean execute() throws SQLException { - return executeUpdate(getNativeSql()) == 0; + return super.execute(getNativeSql()); } @Override diff --git a/src/connector/jdbc/src/main/java/com/taosdata/jdbc/TSDBStatement.java b/src/connector/jdbc/src/main/java/com/taosdata/jdbc/TSDBStatement.java index 7d4f4c4691..455b73bb92 100644 --- a/src/connector/jdbc/src/main/java/com/taosdata/jdbc/TSDBStatement.java +++ b/src/connector/jdbc/src/main/java/com/taosdata/jdbc/TSDBStatement.java @@ -27,8 +27,14 @@ public class TSDBStatement implements Statement { /** Timeout for a query */ protected int queryTimeout = 0; + /** + * Status of current statement + */ + private boolean isClosed = true; + TSDBStatement(TSDBJNIConnector connecter) { this.connecter = connecter; + this.isClosed = false; } public T unwrap(Class iface) throws SQLException { @@ -40,13 +46,16 @@ public class TSDBStatement implements Statement { } public ResultSet executeQuery(String sql) throws SQLException { + if (isClosed) { + throw new SQLException("Invalid method call on a closed statement."); + } this.connecter.executeQuery(sql); long resultSetPointer = this.connecter.getResultSet(); if (resultSetPointer == TSDBConstants.JNI_CONNECTION_NULL) { throw new SQLException(TSDBConstants.FixErrMsg(TSDBConstants.JNI_CONNECTION_NULL)); - } else if (resultSetPointer == 0) { + } else if (resultSetPointer == TSDBConstants.JNI_NULL_POINTER) { return null; } else { return new TSDBResultSet(this.connecter, resultSetPointer); @@ -54,7 +63,20 @@ public class TSDBStatement implements Statement { } public int executeUpdate(String sql) throws SQLException { - return this.connecter.executeQuery(sql); + if (isClosed) { + throw new SQLException("Invalid method call on a closed statement."); + } + int res = this.connecter.executeQuery(sql); + long resultSetPointer = this.connecter.getResultSet(); + + if (resultSetPointer == TSDBConstants.JNI_CONNECTION_NULL) { + throw new SQLException(TSDBConstants.FixErrMsg(TSDBConstants.JNI_CONNECTION_NULL)); + } else if (resultSetPointer != TSDBConstants.JNI_NULL_POINTER) { + this.connecter.freeResultSet(); + throw new SQLException("The executed SQL is not a DML or a DDL"); + } else { + return res; + } } public String getErrorMsg() { @@ -62,6 +84,12 @@ public class TSDBStatement implements Statement { } public void close() throws SQLException { + if (!isClosed) { + if (!this.connecter.isResultsetClosed()) { + this.connecter.freeResultSet(); + } + isClosed = true; + } } public int getMaxFieldSize() throws SQLException { @@ -110,19 +138,38 @@ public class TSDBStatement implements Statement { } public boolean execute(String sql) throws SQLException { - return executeUpdate(sql) == 0; + if (isClosed) { + throw new SQLException("Invalid method call on a closed statement."); + } + boolean res = true; + this.connecter.executeQuery(sql); + long resultSetPointer = this.connecter.getResultSet(); + + if (resultSetPointer == TSDBConstants.JNI_CONNECTION_NULL) { + throw new SQLException(TSDBConstants.FixErrMsg(TSDBConstants.JNI_CONNECTION_NULL)); + } else if (resultSetPointer == TSDBConstants.JNI_NULL_POINTER) { + // no result set is retrieved + res = false; + } + return res; } public ResultSet getResultSet() throws SQLException { + if (isClosed) { + throw new SQLException("Invalid method call on a closed statement."); + } long resultSetPointer = connecter.getResultSet(); TSDBResultSet resSet = null; - if (resultSetPointer != 0l) { + if (resultSetPointer != TSDBConstants.JNI_NULL_POINTER) { resSet = new TSDBResultSet(connecter, resultSetPointer); } return resSet; } public int getUpdateCount() throws SQLException { + if (isClosed) { + throw new SQLException("Invalid method call on a closed statement."); + } return this.connecter.getAffectedRows(); } @@ -171,6 +218,9 @@ public class TSDBStatement implements Statement { } public int[] executeBatch() throws SQLException { + if (isClosed) { + throw new SQLException("Invalid method call on a closed statement."); + } if (batchedArgs == null) { throw new SQLException(TSDBConstants.WrapErrMsg("Batch is empty!")); } else { @@ -223,7 +273,7 @@ public class TSDBStatement implements Statement { } public boolean isClosed() throws SQLException { - throw new SQLException(TSDBConstants.UNSUPPORT_METHOD_EXCEPTIONZ_MSG); + return isClosed; } public void setPoolable(boolean poolable) throws SQLException { diff --git a/src/inc/textbuffer.h b/src/inc/textbuffer.h index e601a0c129..f86e14ec8b 100644 --- a/src/inc/textbuffer.h +++ b/src/inc/textbuffer.h @@ -184,7 +184,7 @@ void tColModelDisplayEx(tColModel *pModel, void *pData, int32_t numOfRows, int32 /* * compress data into consecutive block without hole in data */ -void tColModelCompress(tColModel *pModel, tFilePage *inputBuffer, int32_t maxElemsCapacity); +void tColModelCompact(tColModel *pModel, tFilePage *inputBuffer, int32_t maxElemsCapacity); void tColModelErase(tColModel *pModel, tFilePage *inputBuffer, int32_t maxCapacity, int32_t s, int32_t e); diff --git a/src/inc/tinterpolation.h b/src/inc/tinterpolation.h index 3592664804..e8bbd66692 100644 --- a/src/inc/tinterpolation.h +++ b/src/inc/tinterpolation.h @@ -69,7 +69,7 @@ int32_t taosGetNumOfResWithoutLimit(SInterpolationInfo *pInterpoInfo, int64_t *p * @param pInterpoInfo * @return */ -bool taosHasNoneInterpoPoints(SInterpolationInfo *pInterpoInfo); +bool taosHasRemainsDataForInterpolation(SInterpolationInfo *pInterpoInfo); int32_t taosNumOfRemainPoints(SInterpolationInfo *pInterpoInfo); diff --git a/src/kit/shell/src/shellEngine.c b/src/kit/shell/src/shellEngine.c index 1842a42f6c..ce1d41eec8 100644 --- a/src/kit/shell/src/shellEngine.c +++ b/src/kit/shell/src/shellEngine.c @@ -626,6 +626,7 @@ void source_file(TAOS *con, char *fptr) { } while ((read_len = getline(&line, &line_len, f)) != -1) { + if (read_len >= MAX_COMMAND_SIZE) continue; line[--read_len] = '\0'; if (read_len == 0 || isCommentLine(line)) { // line starts with # diff --git a/src/modules/http/src/gcHandle.c b/src/modules/http/src/gcHandle.c index dfea677ffc..eb0c6223a7 100644 --- a/src/modules/http/src/gcHandle.c +++ b/src/modules/http/src/gcHandle.c @@ -64,14 +64,12 @@ bool gcProcessLoginRequest(HttpContext* pContext) { //[{ // "refId": "A", // "alias" : "taosd", -// "sql" : "select first(taosd) from sys.mem where ts > now-6h and ts < now -// interval(20000a)" +// "sql" : "select first(taosd) from sys.mem where ts > now-6h and ts < now interval(20000a)" //}, //{ // "refId": "B", // "alias" : "system", -// "sql" : "select first(taosd) from sys.mem where ts > now-6h and ts < now -// interval(20000a)" +// "sql" : "select first(taosd) from sys.mem where ts > now-6h and ts < now interval(20000a)" //}] // output //[{ diff --git a/src/modules/http/src/httpCode.c b/src/modules/http/src/httpCode.c index e541872e2b..a1d66f32b4 100644 --- a/src/modules/http/src/httpCode.c +++ b/src/modules/http/src/httpCode.c @@ -20,7 +20,8 @@ char* httpMsg[] = { "http method parse error", // 3 "http version should be 1.0, 1.1 or 1.2", // 4 "http head parse error", // 5 - "request size is too big", "http body size invalid", + "request size is too big", + "http body size invalid", "http chunked body parse error", // 8 "http url parse error", // 9 "invalid type of Authorization", @@ -52,7 +53,8 @@ char* httpMsg[] = { "tags not find", "tags size is 0", "tags size too long", // 36 - "tag is null", "tag name is null", + "tag is null", + "tag name is null", "tag name length too long", // 39 "tag value type should be number or string", "tag value is null", diff --git a/src/system/inc/vnode.h b/src/system/inc/vnode.h index 9cbc7e664b..fa118cd1ec 100644 --- a/src/system/inc/vnode.h +++ b/src/system/inc/vnode.h @@ -69,11 +69,12 @@ enum _sync_cmd { }; enum _meter_state { - TSDB_METER_STATE_READY, - TSDB_METER_STATE_IMPORTING, - TSDB_METER_STATE_UPDATING, - TSDB_METER_STATE_DELETING, - TSDB_METER_STATE_DELETED, + TSDB_METER_STATE_READY = 0x00, + TSDB_METER_STATE_INSERT = 0x01, + TSDB_METER_STATE_IMPORTING = 0x02, + TSDB_METER_STATE_UPDATING = 0x04, + TSDB_METER_STATE_DELETING = 0x10, + TSDB_METER_STATE_DELETED = 0x18, }; typedef struct { @@ -184,10 +185,10 @@ typedef struct _meter_obj { short sqlLen; char searchAlgorithm : 4; char compAlgorithm : 4; - char state : 5; // deleted or added, 1: added - char status : 3; // 0: ok, 1: stop stream computing + char status; // 0: ok, 1: stop stream computing char reserved[16]; + int state; int numOfQueries; char * pSql; void * pStream; @@ -418,10 +419,6 @@ void vnodeCommitOver(SVnodeObj *pVnode); TSKEY vnodeGetFirstKey(int vnode); -int vnodeSyncRetrieveCache(int vnode, int fd); - -int vnodeSyncRestoreCache(int vnode, int fd); - pthread_t vnodeCreateCommitThread(SVnodeObj *pVnode); void vnodeCancelCommit(SVnodeObj *pVnode); @@ -447,10 +444,6 @@ void *vnodeCommitToFile(void *param); void *vnodeCommitMultiToFile(SVnodeObj *pVnode, int ssid, int esid); -int vnodeSyncRetrieveFile(int vnode, int fd, uint32_t fileId, uint64_t *fmagic); - -int vnodeSyncRestoreFile(int vnode, int sfd); - int vnodeWriteBlockToFile(SMeterObj *pObj, SCompBlock *pBlock, SData *data[], SData *cdata[], int pointsRead); int vnodeSearchPointInFile(SMeterObj *pObj, SQuery *pQuery); @@ -476,14 +469,8 @@ void *vnodeGetMeterPeerConnection(SMeterObj *pObj, int index); int vnodeForwardToPeer(SMeterObj *pObj, char *msg, int msgLen, char action, int sversion); -void vnodeCloseAllSyncFds(int vnode); - void vnodeConfigVPeers(int vnode, int numOfPeers, SVPeerDesc peerDesc[]); -void vnodeStartSyncProcess(SVnodeObj *pVnode); - -void vnodeCancelSync(int vnode); - void vnodeListPeerStatus(char *buffer); void vnodeCheckOwnStatus(SVnodeObj *pVnode); @@ -499,7 +486,7 @@ int vnodeInitStore(); void vnodeCleanUpVnodes(); -void vnodeRemoveVnode(int vnode); +int vnodeRemoveVnode(int vnode); int vnodeCreateVnode(int vnode, SVnodeCfg *pCfg, SVPeerDesc *pDesc); diff --git a/src/system/inc/vnodeUtil.h b/src/system/inc/vnodeUtil.h index 9713b4da0d..2f619c85aa 100644 --- a/src/system/inc/vnodeUtil.h +++ b/src/system/inc/vnodeUtil.h @@ -75,6 +75,12 @@ int32_t vnodeIncQueryRefCount(SQueryMeterMsg *pQueryMsg, SMeterSidExtInfo **pSid void vnodeDecQueryRefCount(SQueryMeterMsg *pQueryMsg, SMeterObj **pMeterObjList, int32_t numOfInc); +int32_t vnodeTransferMeterState(SMeterObj* pMeterObj, int32_t state); +void vnodeClearMeterState(SMeterObj* pMeterObj, int32_t state); +bool vnodeIsMeterState(SMeterObj* pMeterObj, int32_t state); +void vnodeSetMeterDeleting(SMeterObj* pMeterObj); +bool vnodeIsSafeToDeleteMeter(SVnodeObj* pVnode, int32_t sid); + #ifdef __cplusplus } #endif diff --git a/src/system/src/dnodeMgmt.c b/src/system/src/dnodeMgmt.c index 0c6822159c..28e723f3f0 100644 --- a/src/system/src/dnodeMgmt.c +++ b/src/system/src/dnodeMgmt.c @@ -113,7 +113,7 @@ int vnodeProcessCreateMeterRequest(char *pMsg) { pVnode = vnodeList + vid; if (pVnode->cfg.maxSessions <= 0) { dError("vid:%d, not activated", vid); - code = TSDB_CODE_INVALID_SESSION_ID; + code = TSDB_CODE_NOT_ACTIVE_SESSION; goto _over; } @@ -215,7 +215,7 @@ int vnodeProcessCreateMeterMsg(char *pMsg) { if (pVnode->pCachePool == NULL) { dError("vid:%d is not activated yet", pCreate->vnode); vnodeSendVpeerCfgMsg(pCreate->vnode); - code = TSDB_CODE_INVALID_SESSION_ID; + code = TSDB_CODE_NOT_ACTIVE_SESSION; goto _create_over; } @@ -445,7 +445,8 @@ int vnodeProcessFreeVnodeRequest(char *pMsg) { } dTrace("vid:%d receive free vnode message", pFree->vnode); - vnodeRemoveVnode(pFree->vnode); + int32_t code = vnodeRemoveVnode(pFree->vnode); + assert(code == TSDB_CODE_SUCCESS || code == TSDB_CODE_ACTION_IN_PROGRESS); pStart = (char *)malloc(128); if (pStart == NULL) return 0; @@ -453,7 +454,7 @@ int vnodeProcessFreeVnodeRequest(char *pMsg) { *pStart = TSDB_MSG_TYPE_FREE_VNODE_RSP; pMsg = pStart + 1; - *pMsg = 0; + *pMsg = code; vnodeSendMsgToMgmt(pStart); return 0; diff --git a/src/system/src/dnodeSystem.c b/src/system/src/dnodeSystem.c index 7d7c3ee6aa..bc08d5b3c2 100644 --- a/src/system/src/dnodeSystem.c +++ b/src/system/src/dnodeSystem.c @@ -44,7 +44,7 @@ void dnodeInitModules() { tsModule[TSDB_MOD_HTTP].cleanUpFp = httpCleanUpSystem; tsModule[TSDB_MOD_HTTP].startFp = httpStartSystem; tsModule[TSDB_MOD_HTTP].stopFp = httpStopSystem; - tsModule[TSDB_MOD_HTTP].num = tsEnableHttpModule ? -1 : 0; + tsModule[TSDB_MOD_HTTP].num = (tsEnableHttpModule == 1) ? -1 : 0; tsModule[TSDB_MOD_HTTP].curNum = 0; tsModule[TSDB_MOD_HTTP].equalVnodeNum = 0; @@ -53,7 +53,7 @@ void dnodeInitModules() { tsModule[TSDB_MOD_MONITOR].cleanUpFp = monitorCleanUpSystem; tsModule[TSDB_MOD_MONITOR].startFp = monitorStartSystem; tsModule[TSDB_MOD_MONITOR].stopFp = monitorStopSystem; - tsModule[TSDB_MOD_MONITOR].num = tsEnableMonitorModule ? -1 : 0; + tsModule[TSDB_MOD_MONITOR].num = (tsEnableMonitorModule == 1) ? -1 : 0; tsModule[TSDB_MOD_MONITOR].curNum = 0; tsModule[TSDB_MOD_MONITOR].equalVnodeNum = 0; } diff --git a/src/system/src/mgmtMeter.c b/src/system/src/mgmtMeter.c index f8ac13061d..bf09c90c7d 100644 --- a/src/system/src/mgmtMeter.c +++ b/src/system/src/mgmtMeter.c @@ -1140,54 +1140,13 @@ static void mgmtReorganizeMetersInMetricMeta(STabObj *pMetric, SMetricMetaMsg *p startPos[1] = (int32_t)pRes->num; } - /* if pInfo->limit == 0, the query will be intercepted by sdk, and wont be - * sent to mnode */ - assert(pInfo->limit == -1 || pInfo->limit > 0); - - int32_t numOfTotal = 0; - if (pInfo->offset >= numOfSubset) { - numOfTotal = 0; - } else if (numOfSubset == 1) { - // no 'groupBy' clause, all tables returned - numOfTotal = pRes->num; - } else { - /* there is a offset value of group */ - int32_t start = 0; - int32_t end = 0; - - if (pInfo->orderType == TSQL_SO_ASC) { - start = startPos[pInfo->offset]; - - if (pInfo->limit + pInfo->offset >= numOfSubset || pInfo->limit == -1) { - /* all results are required */ - end = startPos[numOfSubset]; - } else { - end = startPos[pInfo->limit + pInfo->offset]; - } - } else { - end = startPos[numOfSubset - pInfo->offset]; - - if (pInfo->limit + pInfo->offset >= numOfSubset || pInfo->limit == -1) { - start = startPos[0]; - } else { - start = startPos[numOfSubset - pInfo->limit - pInfo->offset]; - } - } - - numOfTotal = end - start; - assert(numOfTotal > 0); - - memmove(pRes->pRes, pRes->pRes + start, numOfTotal * POINTER_BYTES); - } - /* * sort the result according to vgid to ensure meters with the same vgid is * continuous in the result list */ __compar_fn_t functor = (pRes->nodeType == TAST_NODE_TYPE_METER_PTR) ? tabObjVGIDComparator : nodeVGIDComparator; - qsort(pRes->pRes, numOfTotal, POINTER_BYTES, functor); + qsort(pRes->pRes, (size_t) pRes->num, POINTER_BYTES, functor); - pRes->num = numOfTotal; free(descriptor->pTagSchema); free(descriptor); free(startPos); diff --git a/src/system/src/vnodeCache.c b/src/system/src/vnodeCache.c index 81ce27fd83..4c166136d0 100644 --- a/src/system/src/vnodeCache.c +++ b/src/system/src/vnodeCache.c @@ -340,19 +340,33 @@ void vnodeCommitOver(SVnodeObj *pVnode) { pthread_mutex_unlock(&pPool->vmutex); } +static void vnodeWaitForCommitComplete(SVnodeObj *pVnode) { + SCachePool *pPool = (SCachePool *)(pVnode->pCachePool); + + // wait for 100s at most + const int32_t totalCount = 1000; + int32_t count = 0; + + // all meter is marked as dropped, so the commit will abort very quickly + while(count++ < totalCount) { + int32_t commitInProcess = 0; + + pthread_mutex_lock(&pPool->vmutex); + commitInProcess = pPool->commitInProcess; + pthread_mutex_unlock(&pPool->vmutex); + + if (commitInProcess) { + dWarn("vid:%d still in commit, wait for completed", pVnode->vnode); + taosMsleep(10); + } + } +} + void vnodeCancelCommit(SVnodeObj *pVnode) { SCachePool *pPool = (SCachePool *)(pVnode->pCachePool); if (pPool == NULL) return; - pthread_mutex_lock(&pPool->vmutex); - - if (pPool->commitInProcess) { - pPool->commitInProcess = 0; - pthread_cancel(pVnode->commitThread); - } - - pthread_mutex_unlock(&pPool->vmutex); - + vnodeWaitForCommitComplete(pVnode); taosTmrReset(vnodeProcessCommitTimer, pVnode->cfg.commitTime * 1000, pVnode, vnodeTmrCtrl, &pVnode->commitTimer); } diff --git a/src/system/src/vnodeCommit.c b/src/system/src/vnodeCommit.c index d2c6d82556..64de3bee7f 100644 --- a/src/system/src/vnodeCommit.c +++ b/src/system/src/vnodeCommit.c @@ -26,6 +26,7 @@ #include "tsdb.h" #include "vnode.h" +#include "vnodeUtil.h" typedef struct { int sversion; @@ -160,13 +161,17 @@ size_t vnodeRestoreDataFromLog(int vnode, char *fileName, uint64_t *firstV) { if (*(int *)(cont+head.contLen) != simpleCheck) break; SMeterObj *pObj = pVnode->meterList[head.sid]; if (pObj == NULL) { - dError( - "vid:%d, sid:%d not exists, ignore data in commit log, " - "contLen:%d action:%d", + dError("vid:%d, sid:%d not exists, ignore data in commit log, contLen:%d action:%d", vnode, head.sid, head.contLen, head.action); continue; } + if (vnodeIsMeterState(pObj, TSDB_METER_STATE_DELETING)) { + dWarn("vid:%d sid:%d id:%s, meter is dropped, ignore data in commit log, contLen:%d action:%d", + vnode, head.sid, head.contLen, head.action); + continue; + } + int32_t numOfPoints = 0; (*vnodeProcessAction[head.action])(pObj, cont, head.contLen, TSDB_DATA_SOURCE_LOG, NULL, head.sversion, &numOfPoints); diff --git a/src/system/src/vnodeFile.c b/src/system/src/vnodeFile.c index 82af0cf145..98bc482d3a 100644 --- a/src/system/src/vnodeFile.c +++ b/src/system/src/vnodeFile.c @@ -408,7 +408,6 @@ void vnodeCloseCommitFiles(SVnodeObj *pVnode) { char dpath[TSDB_FILENAME_LEN] = "\0"; int fileId; int ret; - int file_removed = 0; close(pVnode->nfd); pVnode->nfd = 0; @@ -449,14 +448,15 @@ void vnodeCloseCommitFiles(SVnodeObj *pVnode) { dTrace("vid:%d, %s and %s is saved", pVnode->vnode, pVnode->cfn, pVnode->lfn); - if (pVnode->numOfFiles > pVnode->maxFiles) { - fileId = pVnode->fileId - pVnode->numOfFiles + 1; + // Retention policy here + fileId = pVnode->fileId - pVnode->numOfFiles + 1; + int cfile = taosGetTimestamp(pVnode->cfg.precision)/pVnode->cfg.daysPerFile/tsMsPerDay[pVnode->cfg.precision]; + while (fileId <= cfile - pVnode->maxFiles) { vnodeRemoveFile(pVnode->vnode, fileId); pVnode->numOfFiles--; - file_removed = 1; + fileId++; } - if (!file_removed) vnodeUpdateFileMagic(pVnode->vnode, pVnode->commitFileId); vnodeSaveAllMeterObjToFile(pVnode->vnode); return; @@ -577,8 +577,20 @@ _again: // read compInfo for (sid = 0; sid < pCfg->maxSessions; ++sid) { + if (pVnode->meterList == NULL) { // vnode is being freed, abort + goto _over; + } + pObj = (SMeterObj *)(pVnode->meterList[sid]); - if (pObj == NULL) continue; + if (pObj == NULL) { + continue; + } + + // meter is going to be deleted, abort + if (vnodeIsMeterState(pObj, TSDB_METER_STATE_DELETING)) { + dWarn("vid:%d sid:%d is dropped, ignore this meter", vnode, sid); + continue; + } pMeter = meterInfo + sid; pHeader = ((SCompHeader *)tmem) + sid; @@ -672,8 +684,9 @@ _again: pointsReadLast = pMeter->lastBlock.numOfPoints; query.over = 0; headInfo.totalStorage -= (pointsReadLast * pObj->bytesPerPoint); + dTrace("vid:%d sid:%d id:%s, points:%d in last block will be merged to new block", - pObj->vnode, pObj->sid, pObj->meterId, pointsReadLast); + pObj->vnode, pObj->sid, pObj->meterId, pointsReadLast); } pMeter->changed = 1; @@ -717,8 +730,8 @@ _again: } dTrace("vid:%d sid:%d id:%s, %d points are committed, lastKey:%lld slot:%d pos:%d newNumOfBlocks:%d", - pObj->vnode, pObj->sid, pObj->meterId, pMeter->committedPoints, pObj->lastKeyOnFile, query.slot, query.pos, - pMeter->newNumOfBlocks); + pObj->vnode, pObj->sid, pObj->meterId, pMeter->committedPoints, pObj->lastKeyOnFile, query.slot, query.pos, + pMeter->newNumOfBlocks); if (pMeter->committedPoints > 0) { pMeter->commitSlot = query.slot; diff --git a/src/system/src/vnodeImport.c b/src/system/src/vnodeImport.c index f9874edd36..a49c9a5619 100644 --- a/src/system/src/vnodeImport.c +++ b/src/system/src/vnodeImport.c @@ -24,6 +24,7 @@ #include "vnode.h" #include "vnodeMgmt.h" #include "vnodeShell.h" +#include "vnodeShell.h" #include "vnodeUtil.h" #pragma GCC diagnostic ignored "-Wpointer-sign" #pragma GCC diagnostic ignored "-Wint-conversion" @@ -281,14 +282,32 @@ void vnodeProcessImportTimer(void *param, void *tmrId) { SShellObj * pShell = pImport->pShell; pImport->retry++; - pObj->state = TSDB_METER_STATE_IMPORTING; + //slow query will block the import operation + int32_t state = vnodeTransferMeterState(pObj, TSDB_METER_STATE_IMPORTING); + if (state >= TSDB_METER_STATE_DELETING) { + dError("vid:%d sid:%d id:%s, meter is deleted, failed to import, state:%d", + pObj->vnode, pObj->sid, pObj->meterId, state); + return; + } + + int32_t num = 0; + pthread_mutex_lock(&pVnode->vmutex); + num = pObj->numOfQueries; + pthread_mutex_unlock(&pVnode->vmutex); + + //if the num == 0, it will never be increased before state is set to TSDB_METER_STATE_READY + int32_t commitInProcess = 0; pthread_mutex_lock(&pPool->vmutex); - if (pPool->commitInProcess || pObj->numOfQueries > 0) { + if (((commitInProcess = pPool->commitInProcess) == 1) || num > 0 || state != TSDB_METER_STATE_READY) { pthread_mutex_unlock(&pPool->vmutex); - pObj->state = TSDB_METER_STATE_READY; + vnodeClearMeterState(pObj, TSDB_METER_STATE_IMPORTING); + if (pImport->retry < 1000) { - dTrace("vid:%d sid:%d id:%s, commit in process, try to import later", pObj->vnode, pObj->sid, pObj->meterId); + dTrace("vid:%d sid:%d id:%s, import failed, retry later. commit in process or queries on it, or not ready." + "commitInProcess:%d, numOfQueries:%d, state:%d", pObj->vnode, pObj->sid, pObj->meterId, + commitInProcess, num, state); + taosTmrStart(vnodeProcessImportTimer, 10, pImport, vnodeTmrCtrl); return; } else { @@ -304,7 +323,8 @@ void vnodeProcessImportTimer(void *param, void *tmrId) { } } - pObj->state = TSDB_METER_STATE_READY; + vnodeClearMeterState(pObj, TSDB_METER_STATE_IMPORTING); + pVnode->version++; // send response back to shell @@ -850,10 +870,13 @@ int vnodeImportPoints(SMeterObj *pObj, char *cont, int contLen, char source, voi } payload = pSubmit->payLoad; - if (pVnode->lastKeyOnFile > pVnode->cfg.daysToKeep * tsMsPerDay[pVnode->cfg.precision] + *((TSKEY *)(payload))) { - dError("vid:%d sid:%d id:%s, vnode lastKeyOnFile:%lld, data is too old to import, key:%lld", - pObj->vnode, pObj->sid, pObj->meterId, pVnode->lastKeyOnFile, *(TSKEY *)(payload)); - return TSDB_CODE_OTHERS; + int firstId = (*(TSKEY *)payload)/pVnode->cfg.daysPerFile/tsMsPerDay[pVnode->cfg.precision]; + int lastId = (*(TSKEY *)(payload+pObj->bytesPerPoint*(rows-1)))/pVnode->cfg.daysPerFile/tsMsPerDay[pVnode->cfg.precision]; + int cfile = taosGetTimestamp(pVnode->cfg.precision)/pVnode->cfg.daysPerFile/tsMsPerDay[pVnode->cfg.precision]; + if ((firstId <= cfile - pVnode->maxFiles) || (firstId > cfile + 1) || (lastId <= cfile - pVnode->maxFiles) || (lastId > cfile + 1)) { + dError("vid:%d sid:%d id:%s, invalid timestamp to import, firstKey: %ld lastKey: %ld", + pObj->vnode, pObj->sid, pObj->meterId, *(TSKEY *)(payload), *(TSKEY *)(payload+pObj->bytesPerPoint*(rows-1))); + return TSDB_CODE_TIMESTAMP_OUT_OF_RANGE; } if ( pVnode->cfg.commitLog && source != TSDB_DATA_SOURCE_LOG) { @@ -862,15 +885,19 @@ int vnodeImportPoints(SMeterObj *pObj, char *cont, int contLen, char source, voi } if (*((TSKEY *)(pSubmit->payLoad + (rows - 1) * pObj->bytesPerPoint)) > pObj->lastKey) { + vnodeClearMeterState(pObj, TSDB_METER_STATE_IMPORTING); + vnodeTransferMeterState(pObj, TSDB_METER_STATE_INSERT); code = vnodeInsertPoints(pObj, cont, contLen, TSDB_DATA_SOURCE_LOG, NULL, pObj->sversion, &pointsImported); + if (pShell) { pShell->code = code; pShell->numOfTotalPoints += pointsImported; } + + vnodeClearMeterState(pObj, TSDB_METER_STATE_INSERT); } else { SImportInfo *pNew, import; - pObj->state = TSDB_METER_STATE_IMPORTING; dTrace("vid:%d sid:%d id:%s, import %d rows data", pObj->vnode, pObj->sid, pObj->meterId, rows); memset(&import, 0, sizeof(import)); import.firstKey = *((TSKEY *)(payload)); @@ -880,10 +907,16 @@ int vnodeImportPoints(SMeterObj *pObj, char *cont, int contLen, char source, voi import.payload = payload; import.rows = rows; + int32_t num = 0; + pthread_mutex_lock(&pVnode->vmutex); + num = pObj->numOfQueries; + pthread_mutex_unlock(&pVnode->vmutex); + + int32_t commitInProcess = 0; + pthread_mutex_lock(&pPool->vmutex); - if (pPool->commitInProcess || pObj->numOfQueries > 0) { + if (((commitInProcess = pPool->commitInProcess) == 1) || num > 0) { pthread_mutex_unlock(&pPool->vmutex); - pObj->state = TSDB_METER_STATE_READY; pNew = (SImportInfo *)malloc(sizeof(SImportInfo)); memcpy(pNew, &import, sizeof(SImportInfo)); @@ -892,8 +925,9 @@ int vnodeImportPoints(SMeterObj *pObj, char *cont, int contLen, char source, voi pNew->payload = malloc(payloadLen); memcpy(pNew->payload, payload, payloadLen); - dTrace("vid:%d sid:%d id:%s, commit/query:%d in process, import later, ", pObj->vnode, pObj->sid, pObj->meterId, - pObj->numOfQueries); + dTrace("vid:%d sid:%d id:%s, import later, commit in process:%d, numOfQueries:%d", pObj->vnode, pObj->sid, + pObj->meterId, commitInProcess, pObj->numOfQueries); + taosTmrStart(vnodeProcessImportTimer, 10, pNew, vnodeTmrCtrl); return 0; } else { @@ -907,7 +941,6 @@ int vnodeImportPoints(SMeterObj *pObj, char *cont, int contLen, char source, voi } } - pObj->state = TSDB_METER_STATE_READY; pVnode->version++; if (pShell) { @@ -918,6 +951,7 @@ int vnodeImportPoints(SMeterObj *pObj, char *cont, int contLen, char source, voi return 0; } +//todo abort from the procedure if the meter is going to be dropped int vnodeImportData(SMeterObj *pObj, SImportInfo *pImport) { int code = 0; diff --git a/src/system/src/vnodeMeter.c b/src/system/src/vnodeMeter.c index 04dc22f7d6..b8961e98ec 100644 --- a/src/system/src/vnodeMeter.c +++ b/src/system/src/vnodeMeter.c @@ -47,6 +47,8 @@ void vnodeFreeMeterObj(SMeterObj *pObj) { if (vnodeList[pObj->vnode].meterList != NULL) { vnodeList[pObj->vnode].meterList[pObj->sid] = NULL; } + + memset(pObj->meterId, 0, tListLen(pObj->meterId)); tfree(pObj); } @@ -143,7 +145,7 @@ int vnodeSaveMeterObjToFile(SMeterObj *pObj) { memcpy(buffer, pObj, offsetof(SMeterObj, reserved)); memcpy(buffer + offsetof(SMeterObj, reserved), pObj->schema, pObj->numOfColumns * sizeof(SColumn)); memcpy(buffer + offsetof(SMeterObj, reserved) + pObj->numOfColumns * sizeof(SColumn), pObj->pSql, pObj->sqlLen); - taosCalcChecksumAppend(0, buffer, new_length); + taosCalcChecksumAppend(0, (uint8_t *)buffer, new_length); if (offset == 0 || length < new_length) { // New, append to file end fseek(fp, 0, SEEK_END); @@ -208,7 +210,7 @@ int vnodeSaveAllMeterObjToFile(int vnode) { memcpy(buffer, pObj, offsetof(SMeterObj, reserved)); memcpy(buffer + offsetof(SMeterObj, reserved), pObj->schema, pObj->numOfColumns * sizeof(SColumn)); memcpy(buffer + offsetof(SMeterObj, reserved) + pObj->numOfColumns * sizeof(SColumn), pObj->pSql, pObj->sqlLen); - taosCalcChecksumAppend(0, buffer, new_length); + taosCalcChecksumAppend(0, (uint8_t *)buffer, new_length); if (offset == 0 || length > new_length) { // New, append to file end new_offset = fseek(fp, 0, SEEK_END); @@ -391,7 +393,7 @@ int vnodeOpenMetersVnode(int vnode) { fseek(fp, offset, SEEK_SET); if (fread(buffer, length, 1, fp) <= 0) break; - if (taosCheckChecksumWhole(buffer, length)) { + if (taosCheckChecksumWhole((uint8_t *)buffer, length)) { vnodeRestoreMeterObj(buffer, length - sizeof(TSCKSUM)); } else { dError("meter object file is broken since checksum mismatch, vnode: %d sid: %d, try to recover", vnode, sid); @@ -440,7 +442,7 @@ int vnodeCreateMeterObj(SMeterObj *pNew, SConnSec *pSec) { } dTrace("vid:%d sid:%d id:%s, update schema", pNew->vnode, pNew->sid, pNew->meterId); - if (pObj->state != TSDB_METER_STATE_UPDATING) vnodeUpdateMeter(pNew, NULL); + if (!vnodeIsMeterState(pObj, TSDB_METER_STATE_UPDATING)) vnodeUpdateMeter(pNew, NULL); return TSDB_CODE_SUCCESS; } @@ -483,27 +485,20 @@ int vnodeRemoveMeterObj(int vnode, int sid) { if (vnodeList[vnode].meterList == NULL) return 0; pObj = vnodeList[vnode].meterList[sid]; - if ((pObj == NULL) || (pObj->state == TSDB_METER_STATE_DELETED)) return 0; - if (pObj->state == TSDB_METER_STATE_IMPORTING) return TSDB_CODE_ACTION_IN_PROGRESS; - - int32_t retFlag = 0; - pthread_mutex_lock(&vnodeList[vnode].vmutex); - pObj->state = TSDB_METER_STATE_DELETING; - if (pObj->numOfQueries > 0) { - retFlag = TSDB_CODE_ACTION_IN_PROGRESS; - dWarn("vid:%d sid:%d id:%s %d queries executing on it, wait query to be finished", - vnode, pObj->sid, pObj->meterId, pObj->numOfQueries); + if (pObj == NULL) { + return TSDB_CODE_SUCCESS; } - pthread_mutex_unlock(&vnodeList[vnode].vmutex); - if (retFlag != 0) return retFlag; - // after remove this meter, change its stat to DELETED + if (!vnodeIsSafeToDeleteMeter(&vnodeList[vnode], sid)) { + return TSDB_CODE_ACTION_IN_PROGRESS; + } + + // after remove this meter, change its state to DELETED pObj->state = TSDB_METER_STATE_DELETED; pObj->timeStamp = taosGetTimestampMs(); vnodeList[vnode].lastRemove = pObj->timeStamp; vnodeRemoveStream(pObj); - pObj->meterId[0] = 0; vnodeSaveMeterObjToFile(pObj); vnodeFreeMeterObj(pObj); @@ -517,6 +512,7 @@ int vnodeInsertPoints(SMeterObj *pObj, char *cont, int contLen, char source, voi SSubmitMsg *pSubmit = (SSubmitMsg *)cont; char * pData; TSKEY tsKey; + int cfile; int points = 0; int code = TSDB_CODE_SUCCESS; SVnodeObj * pVnode = vnodeList + pObj->vnode; @@ -533,6 +529,7 @@ int vnodeInsertPoints(SMeterObj *pObj, char *cont, int contLen, char source, voi // to guarantee time stamp is the same for all vnodes pData = pSubmit->payLoad; tsKey = taosGetTimestamp(pVnode->cfg.precision); + cfile = tsKey/pVnode->cfg.daysPerFile/tsMsPerDay[pVnode->cfg.precision]; if (*((TSKEY *)pData) == 0) { for (i = 0; i < numOfPoints; ++i) { *((TSKEY *)pData) = tsKey++; @@ -575,13 +572,24 @@ int vnodeInsertPoints(SMeterObj *pObj, char *cont, int contLen, char source, voi code = 0; TSKEY firstKey = *((TSKEY *)pData); - if (pVnode->lastKeyOnFile > pVnode->cfg.daysToKeep * tsMsPerDay[pVnode->cfg.precision] + firstKey) { - dError("vid:%d sid:%d id:%s, vnode lastKeyOnFile:%lld, data is too old to insert, key:%lld", pObj->vnode, pObj->sid, - pObj->meterId, pVnode->lastKeyOnFile, firstKey); - return TSDB_CODE_OTHERS; + int firstId = firstKey/pVnode->cfg.daysPerFile/tsMsPerDay[pVnode->cfg.precision]; + int lastId = (*(TSKEY *)(pData + pObj->bytesPerPoint * (numOfPoints - 1)))/pVnode->cfg.daysPerFile/tsMsPerDay[pVnode->cfg.precision]; + if ((firstId <= cfile - pVnode->maxFiles) || (firstId > cfile + 1) || (lastId <= cfile - pVnode->maxFiles) || (lastId > cfile + 1)) { + dError("vid:%d sid:%d id:%s, invalid timestamp to insert, firstKey: %ld lastKey: %ld ", pObj->vnode, pObj->sid, + pObj->meterId, firstKey, (*(TSKEY *)(pData + pObj->bytesPerPoint * (numOfPoints - 1)))); + return TSDB_CODE_TIMESTAMP_OUT_OF_RANGE; } for (i = 0; i < numOfPoints; ++i) { + // meter will be dropped, abort current insertion + if (pObj->state >= TSDB_METER_STATE_DELETING) { + dWarn("vid:%d sid:%d id:%s, meter is dropped, abort insert, state:%d", pObj->vnode, pObj->sid, pObj->meterId, + pObj->state); + + code = TSDB_CODE_NOT_ACTIVE_SESSION; + break; + } + if (*((TSKEY *)pData) <= pObj->lastKey) { dWarn("vid:%d sid:%d id:%s, received key:%ld not larger than lastKey:%ld", pObj->vnode, pObj->sid, pObj->meterId, *((TSKEY *)pData), pObj->lastKey); @@ -632,9 +640,11 @@ void vnodeProcessUpdateSchemaTimer(void *param, void *tmrId) { pthread_mutex_lock(&pPool->vmutex); if (pPool->commitInProcess) { - dTrace("vid:%d sid:%d mid:%s, commiting in process, commit later", pObj->vnode, pObj->sid, pObj->meterId); - if (taosTmrStart(vnodeProcessUpdateSchemaTimer, 10, pObj, vnodeTmrCtrl) == NULL) - pObj->state = TSDB_METER_STATE_READY; + dTrace("vid:%d sid:%d mid:%s, committing in process, commit later", pObj->vnode, pObj->sid, pObj->meterId); + if (taosTmrStart(vnodeProcessUpdateSchemaTimer, 10, pObj, vnodeTmrCtrl) == NULL) { + vnodeClearMeterState(pObj, TSDB_METER_STATE_UPDATING); + } + pthread_mutex_unlock(&pPool->vmutex); return; } @@ -649,41 +659,54 @@ void vnodeUpdateMeter(void *param, void *tmrId) { SMeterObj *pNew = (SMeterObj *)param; if (pNew == NULL || pNew->vnode < 0 || pNew->sid < 0) return; - if (vnodeList[pNew->vnode].meterList == NULL) { + SVnodeObj* pVnode = &vnodeList[pNew->vnode]; + + if (pVnode->meterList == NULL) { dTrace("vid:%d sid:%d id:%s, vnode is deleted, abort update schema", pNew->vnode, pNew->sid, pNew->meterId); free(pNew->schema); free(pNew); return; } - SMeterObj *pObj = vnodeList[pNew->vnode].meterList[pNew->sid]; - if (pObj == NULL) { + SMeterObj *pObj = pVnode->meterList[pNew->sid]; + if (pObj == NULL || vnodeIsMeterState(pObj, TSDB_METER_STATE_DELETING)) { dTrace("vid:%d sid:%d id:%s, meter is deleted, abort update schema", pNew->vnode, pNew->sid, pNew->meterId); free(pNew->schema); free(pNew); return; } - pObj->state = TSDB_METER_STATE_UPDATING; + int32_t state = vnodeTransferMeterState(pObj, TSDB_METER_STATE_UPDATING); + if (state >= TSDB_METER_STATE_DELETING) { + dError("vid:%d sid:%d id:%s, meter is deleted, failed to update, state:%d", + pObj->vnode, pObj->sid, pObj->meterId, state); + return; + } - if (pObj->numOfQueries > 0) { + int32_t num = 0; + pthread_mutex_lock(&pVnode->vmutex); + num = pObj->numOfQueries; + pthread_mutex_unlock(&pVnode->vmutex); + + if (num > 0 || state != TSDB_METER_STATE_READY) { + dTrace("vid:%d sid:%d id:%s, update failed, retry later, numOfQueries:%d, state:%d", + pNew->vnode, pNew->sid, pNew->meterId, num, state); + + // retry update meter in 50ms if (taosTmrStart(vnodeUpdateMeter, 50, pNew, vnodeTmrCtrl) == NULL) { - dError("vid:%d sid:%d id:%s, failed to start update timer", pNew->vnode, pNew->sid, pNew->meterId); - pObj->state = TSDB_METER_STATE_READY; + dError("vid:%d sid:%d id:%s, failed to start update timer, no retry", pNew->vnode, pNew->sid, pNew->meterId); free(pNew->schema); free(pNew); } - - dTrace("vid:%d sid:%d id:%s, there are ongoing queries, update later", pNew->vnode, pNew->sid, pNew->meterId); return; } // commit first if (!vnodeIsCacheCommitted(pObj)) { - // commit + // commit data first if (taosTmrStart(vnodeProcessUpdateSchemaTimer, 0, pObj, vnodeTmrCtrl) == NULL) { dError("vid:%d sid:%d id:%s, failed to start commit timer", pObj->vnode, pObj->sid, pObj->meterId); - pObj->state = TSDB_METER_STATE_READY; + vnodeClearMeterState(pObj, TSDB_METER_STATE_UPDATING); free(pNew->schema); free(pNew); return; @@ -691,13 +714,14 @@ void vnodeUpdateMeter(void *param, void *tmrId) { if (taosTmrStart(vnodeUpdateMeter, 50, pNew, vnodeTmrCtrl) == NULL) { dError("vid:%d sid:%d id:%s, failed to start update timer", pNew->vnode, pNew->sid, pNew->meterId); - pObj->state = TSDB_METER_STATE_READY; + vnodeClearMeterState(pObj, TSDB_METER_STATE_UPDATING); free(pNew->schema); free(pNew); } dTrace("vid:%d sid:%d meterId:%s, there are data in cache, commit first, update later", pNew->vnode, pNew->sid, pNew->meterId); + vnodeClearMeterState(pObj, TSDB_METER_STATE_UPDATING); return; } @@ -716,7 +740,7 @@ void vnodeUpdateMeter(void *param, void *tmrId) { pObj->sversion = pNew->sversion; vnodeSaveMeterObjToFile(pObj); - pObj->state = TSDB_METER_STATE_READY; + vnodeClearMeterState(pObj, TSDB_METER_STATE_UPDATING); dTrace("vid:%d sid:%d id:%s, schema is updated", pNew->vnode, pNew->sid, pNew->meterId); free(pNew); diff --git a/src/system/src/vnodeQueryImpl.c b/src/system/src/vnodeQueryImpl.c index 7b315e9753..c958754298 100644 --- a/src/system/src/vnodeQueryImpl.c +++ b/src/system/src/vnodeQueryImpl.c @@ -1730,6 +1730,17 @@ static int64_t getOldestKey(int32_t numOfFiles, int64_t fileId, SVnodeCfg *pCfg) bool isQueryKilled(SQuery *pQuery) { SQInfo *pQInfo = (SQInfo *)GET_QINFO_ADDR(pQuery); + + /* + * check if the queried meter is going to be deleted. + * if it will be deleted soon, stop current query ASAP. + */ + SMeterObj* pMeterObj = pQInfo->pObj; + if (vnodeIsMeterState(pMeterObj, TSDB_METER_STATE_DELETING)) { + pQInfo->killed = 1; + return true; + } + return (pQInfo->killed == 1); } diff --git a/src/system/src/vnodeShell.c b/src/system/src/vnodeShell.c index 3134a04b9a..03ddf522ce 100644 --- a/src/system/src/vnodeShell.c +++ b/src/system/src/vnodeShell.c @@ -15,12 +15,13 @@ #define _DEFAULT_SOURCE -#include "vnodeShell.h" #include #include #include #include #include "taosmsg.h" +#include "vnode.h" +#include "vnodeShell.h" #include "tschemautil.h" #include "textbuffer.h" @@ -28,6 +29,7 @@ #include "vnode.h" #include "vnodeRead.h" #include "vnodeUtil.h" + #pragma GCC diagnostic ignored "-Wint-conversion" void * pShellServer = NULL; @@ -87,6 +89,7 @@ void *vnodeProcessMsgFromShell(char *msg, void *ahandle, void *thandle) { dTrace("vid:%d sid:%d, msg:%s is received pConn:%p", vnode, sid, taosMsg[pMsg->msgType], thandle); + // set in query processing flag if (pMsg->msgType == TSDB_MSG_TYPE_QUERY) { vnodeProcessQueryRequest((char *)pMsg->content, pMsg->msgLen - sizeof(SIntMsg), pObj); } else if (pMsg->msgType == TSDB_MSG_TYPE_RETRIEVE) { @@ -96,7 +99,7 @@ void *vnodeProcessMsgFromShell(char *msg, void *ahandle, void *thandle) { } else { dError("%s is not processed", taosMsg[pMsg->msgType]); } - + return pObj; } @@ -157,16 +160,31 @@ int vnodeOpenShellVnode(int vnode) { return 0; } -void vnodeCloseShellVnode(int vnode) { - taosCloseRpcChann(pShellServer, vnode); +static void vnodeDelayedFreeResource(void *param, void *tmrId) { + int32_t vnode = *(int32_t*) param; + taosCloseRpcChann(pShellServer, vnode); // close connection + tfree (shellList[vnode]); //free SShellObj + tfree(param); +} + +void vnodeCloseShellVnode(int vnode) { if (shellList[vnode] == NULL) return; for (int i = 0; i < vnodeList[vnode].cfg.maxSessions; ++i) { vnodeFreeQInfo(shellList[vnode][i].qhandle, true); } - tfree(shellList[vnode]); + int32_t* v = malloc(sizeof(int32_t)); + *v = vnode; + + /* + * free the connection related resource after 5sec. + * 1. The msg, as well as SRpcConn may be in the task queue, free it immediate will cause crash + * 2. Free connection may cause *(SRpcConn*)pObj->thandle to be invalid to access. + */ + dTrace("vid:%d, delay 5sec to free resources", vnode); + taosTmrStart(vnodeDelayedFreeResource, 5000, v, vnodeTmrCtrl); } void vnodeCleanUpShell() { @@ -230,7 +248,7 @@ int vnodeProcessQueryRequest(char *pMsg, int msgLen, SShellObj *pObj) { } if (pQueryMsg->numOfSids <= 0) { - code = TSDB_CODE_APP_ERROR; + code = TSDB_CODE_INVALID_QUERY_MSG; goto _query_over; } @@ -245,7 +263,7 @@ int vnodeProcessQueryRequest(char *pMsg, int msgLen, SShellObj *pObj) { if (pVnode->cfg.maxSessions == 0) { dError("qmsg:%p,vid:%d is not activated yet", pQueryMsg, pQueryMsg->vnode); vnodeSendVpeerCfgMsg(pQueryMsg->vnode); - code = TSDB_CODE_INVALID_SESSION_ID; + code = TSDB_CODE_NOT_ACTIVE_SESSION; goto _query_over; } @@ -256,13 +274,13 @@ int vnodeProcessQueryRequest(char *pMsg, int msgLen, SShellObj *pObj) { if (pQueryMsg->pSidExtInfo == 0) { dTrace("qmsg:%p,SQueryMeterMsg wrong format", pQueryMsg); - code = TSDB_CODE_APP_ERROR; + code = TSDB_CODE_INVALID_QUERY_MSG; goto _query_over; } if (pVnode->meterList == NULL) { dError("qmsg:%p,vid:%d has been closed", pQueryMsg, pQueryMsg->vnode); - code = TSDB_CODE_INVALID_SESSION_ID; + code = TSDB_CODE_NOT_ACTIVE_SESSION; goto _query_over; } @@ -430,7 +448,7 @@ int vnodeProcessShellSubmitRequest(char *pMsg, int msgLen, SShellObj *pObj) { if (pSubmit->numOfSid <= 0) { dError("invalid num of meters:%d", pSubmit->numOfSid); - code = TSDB_CODE_APP_ERROR; + code = TSDB_CODE_INVALID_QUERY_MSG; goto _submit_over; } @@ -444,7 +462,7 @@ int vnodeProcessShellSubmitRequest(char *pMsg, int msgLen, SShellObj *pObj) { if (pVnode->cfg.maxSessions == 0 || pVnode->meterList == NULL) { dError("vid:%d is not activated for submit", pSubmit->vnode); vnodeSendVpeerCfgMsg(pSubmit->vnode); - code = TSDB_CODE_INVALID_SESSION_ID; + code = TSDB_CODE_NOT_ACTIVE_SESSION; goto _submit_over; } @@ -488,24 +506,39 @@ int vnodeProcessShellSubmitRequest(char *pMsg, int msgLen, SShellObj *pObj) { int subMsgLen = sizeof(pBlocks->numOfRows) + htons(pBlocks->numOfRows) * pMeterObj->bytesPerPoint; int sversion = htonl(pBlocks->sversion); - if (pMeterObj->state == TSDB_METER_STATE_READY) { - if (pSubmit->import) - code = vnodeImportPoints(pMeterObj, (char *)&(pBlocks->numOfRows), subMsgLen, TSDB_DATA_SOURCE_SHELL, pObj, + int32_t state = TSDB_METER_STATE_READY; + if (pSubmit->import) { + state = vnodeTransferMeterState(pMeterObj, TSDB_METER_STATE_IMPORTING); + } else { + state = vnodeTransferMeterState(pMeterObj, TSDB_METER_STATE_INSERT); + } + + if (state == TSDB_METER_STATE_READY) { + // meter status is ready for insert/import + if (pSubmit->import) { + code = vnodeImportPoints(pMeterObj, (char *) &(pBlocks->numOfRows), subMsgLen, TSDB_DATA_SOURCE_SHELL, pObj, sversion, &numOfPoints); - else - code = vnodeInsertPoints(pMeterObj, (char *)&(pBlocks->numOfRows), subMsgLen, TSDB_DATA_SOURCE_SHELL, NULL, + vnodeClearMeterState(pMeterObj, TSDB_METER_STATE_IMPORTING); + } else { + code = vnodeInsertPoints(pMeterObj, (char *) &(pBlocks->numOfRows), subMsgLen, TSDB_DATA_SOURCE_SHELL, NULL, sversion, &numOfPoints); - if (code != 0) break; - } else if (pMeterObj->state >= TSDB_METER_STATE_DELETING) { - dTrace("vid:%d sid:%d id:%s, is is removed, state:", pMeterObj->vnode, pMeterObj->sid, pMeterObj->meterId, - pMeterObj->state); - code = TSDB_CODE_NOT_ACTIVE_SESSION; - break; - } else { // importing state or others - dTrace("vid:%d sid:%d id:%s, try again since in state:%d", pMeterObj->vnode, pMeterObj->sid, pMeterObj->meterId, - pMeterObj->state); - code = TSDB_CODE_ACTION_IN_PROGRESS; - break; + vnodeClearMeterState(pMeterObj, TSDB_METER_STATE_INSERT); + } + + if (code != TSDB_CODE_SUCCESS) {break;} + } else { + if (vnodeIsMeterState(pMeterObj, TSDB_METER_STATE_DELETING)) { + dTrace("vid:%d sid:%d id:%s, it is removed, state:%d", pMeterObj->vnode, pMeterObj->sid, pMeterObj->meterId, + pMeterObj->state); + code = TSDB_CODE_NOT_ACTIVE_SESSION; + break; + } else {// waiting for 300ms by default and try again + dTrace("vid:%d sid:%d id:%s, try submit again since in state:%d", pMeterObj->vnode, pMeterObj->sid, + pMeterObj->meterId, pMeterObj->state); + + code = TSDB_CODE_ACTION_IN_PROGRESS; + break; + } } numOfTotalPoints += numOfPoints; diff --git a/src/system/src/vnodeStore.c b/src/system/src/vnodeStore.c index 29d4b46e8b..d596f50b61 100644 --- a/src/system/src/vnodeStore.c +++ b/src/system/src/vnodeStore.c @@ -85,13 +85,42 @@ int vnodeOpenVnode(int vnode) { return 0; } -void vnodeCloseVnode(int vnode) { - if (vnodeList == NULL) return; +static int32_t vnodeMarkAllMetersDropped(SVnodeObj* pVnode) { + if (pVnode->meterList == NULL) { + assert(pVnode->cfg.maxSessions == 0); + return TSDB_CODE_SUCCESS; + } + + bool ready = true; + for (int sid = 0; sid < pVnode->cfg.maxSessions; ++sid) { + if (!vnodeIsSafeToDeleteMeter(pVnode, sid)) { + ready = false; + } else { // set the meter is to be deleted + SMeterObj* pObj = pVnode->meterList[sid]; + if (pObj != NULL) { + pObj->state = TSDB_METER_STATE_DELETED; + } + } + } + + return ready? TSDB_CODE_SUCCESS:TSDB_CODE_ACTION_IN_PROGRESS; +} + +int vnodeCloseVnode(int vnode) { + if (vnodeList == NULL) return TSDB_CODE_SUCCESS; + + SVnodeObj* pVnode = &vnodeList[vnode]; pthread_mutex_lock(&dmutex); - if (vnodeList[vnode].cfg.maxSessions == 0) { + if (pVnode->cfg.maxSessions == 0) { pthread_mutex_unlock(&dmutex); - return; + return TSDB_CODE_SUCCESS; + } + + // set the meter is dropped flag + if (vnodeMarkAllMetersDropped(pVnode) != TSDB_CODE_SUCCESS) { + pthread_mutex_unlock(&dmutex); + return TSDB_CODE_ACTION_IN_PROGRESS; } vnodeCloseStream(vnodeList + vnode); @@ -111,6 +140,7 @@ void vnodeCloseVnode(int vnode) { vnodeCalcOpenVnodes(); pthread_mutex_unlock(&dmutex); + return TSDB_CODE_SUCCESS; } int vnodeCreateVnode(int vnode, SVnodeCfg *pCfg, SVPeerDesc *pDesc) { @@ -182,25 +212,23 @@ void vnodeRemoveDataFiles(int vnode) { dTrace("vnode %d is removed!", vnode); } -void vnodeRemoveVnode(int vnode) { - if (vnodeList == NULL) return; +int vnodeRemoveVnode(int vnode) { + if (vnodeList == NULL) return TSDB_CODE_SUCCESS; if (vnodeList[vnode].cfg.maxSessions > 0) { - vnodeCloseVnode(vnode); + int32_t ret = vnodeCloseVnode(vnode); + if (ret != TSDB_CODE_SUCCESS) { + return ret; + } vnodeRemoveDataFiles(vnode); - - // sprintf(cmd, "rm -rf %s/vnode%d", tsDirectory, vnode); - // if ( system(cmd) < 0 ) { - // dError("vid:%d, failed to run command %s vnode, reason:%s", vnode, cmd, strerror(errno)); - // } else { - // dTrace("vid:%d, this vnode is deleted!!!", vnode); - // } } else { dTrace("vid:%d, max sessions:%d, this vnode already dropped!!!", vnode, vnodeList[vnode].cfg.maxSessions); - vnodeList[vnode].cfg.maxSessions = 0; + vnodeList[vnode].cfg.maxSessions = 0; //reset value vnodeCalcOpenVnodes(); } + + return TSDB_CODE_SUCCESS; } int vnodeInitStore() { diff --git a/src/system/src/vnodeStream.c b/src/system/src/vnodeStream.c index f2a0eeccbb..d7b7f62bae 100644 --- a/src/system/src/vnodeStream.c +++ b/src/system/src/vnodeStream.c @@ -13,8 +13,9 @@ * along with this program. If not, see . */ +#include "taosmsg.h" #include "vnode.h" -#include +#include "vnodeUtil.h" /* static TAOS *dbConn = NULL; */ void vnodeCloseStreamCallback(void *param); @@ -51,8 +52,17 @@ void vnodeProcessStreamRes(void *param, TAOS_RES *tres, TAOS_ROW row) { } contLen += sizeof(SSubmitMsg); + int32_t numOfPoints = 0; - vnodeInsertPoints(pObj, (char *)pMsg, contLen, TSDB_DATA_SOURCE_SHELL, NULL, pObj->sversion, &numOfPoints); + + int32_t state = vnodeTransferMeterState(pObj, TSDB_METER_STATE_INSERT); + if (state == TSDB_METER_STATE_READY) { + vnodeInsertPoints(pObj, (char *)pMsg, contLen, TSDB_DATA_SOURCE_SHELL, NULL, pObj->sversion, &numOfPoints); + vnodeClearMeterState(pObj, TSDB_METER_STATE_INSERT); + } else { + dError("vid:%d sid:%d id:%s, failed to insert continuous query results, state:%d", pObj->vnode, pObj->sid, + pObj->meterId, state); + } assert(numOfPoints >= 0 && numOfPoints <= 1); tfree(pTemp); @@ -76,7 +86,7 @@ void vnodeOpenStreams(void *param, void *tmrId) { for (int sid = 0; sid < pVnode->cfg.maxSessions; ++sid) { pObj = pVnode->meterList[sid]; - if (pObj == NULL || pObj->sqlLen == 0 || pObj->status == 1 || pObj->state == TSDB_METER_STATE_DELETED) continue; + if (pObj == NULL || pObj->sqlLen == 0 || vnodeIsMeterState(pObj, TSDB_METER_STATE_DELETING)) continue; dTrace("vid:%d sid:%d id:%s, open stream:%s", pObj->vnode, sid, pObj->meterId, pObj->pSql); diff --git a/src/system/src/vnodeUtil.c b/src/system/src/vnodeUtil.c index 51fe9fc3bd..f97b7cb00b 100644 --- a/src/system/src/vnodeUtil.c +++ b/src/system/src/vnodeUtil.c @@ -361,6 +361,7 @@ void vnodeUpdateFilterColumnIndex(SQuery* pQuery) { // TODO support k<12 and k<>9 int32_t vnodeCreateFilterInfo(void* pQInfo, SQuery* pQuery) { + for (int32_t i = 0; i < pQuery->numOfCols; ++i) { if (pQuery->colList[i].data.filterOn > 0) { pQuery->numOfFilterCols++; @@ -401,8 +402,6 @@ int32_t vnodeCreateFilterInfo(void* pQInfo, SQuery* pQuery) { pFilterInfo->fp = rangeFilterArray[2]; } } else { - assert(lower == TSDB_RELATION_LARGE); - if (upper == TSDB_RELATION_LESS_EQUAL) { pFilterInfo->fp = rangeFilterArray[3]; } else { @@ -421,6 +420,7 @@ int32_t vnodeCreateFilterInfo(void* pQInfo, SQuery* pQuery) { pFilterInfo->fp = filterArray[upper]; } } + pFilterInfo->elemSize = bytes; j++; } @@ -470,6 +470,18 @@ bool vnodeIsProjectionQuery(SSqlFunctionExpr* pExpr, int32_t numOfOutput) { return true; } +/* + * the pMeter->state may be changed by vnodeIsSafeToDeleteMeter and import/update processor, the check of + * the state will not always be correct. + * + * The import/update/deleting is actually blocked by current query processing if the check of meter state is + * passed, but later queries are denied. + * + * 1. vnodeIsSafeToDelete will wait for this complete, since it also use the vmutex to check the numOfQueries + * 2. import will check the numOfQueries again after setting state to be TSDB_METER_STATE_IMPORTING, while the + * vmutex is also used. + * 3. insert has nothing to do with the query processing. + */ int32_t vnodeIncQueryRefCount(SQueryMeterMsg* pQueryMsg, SMeterSidExtInfo** pSids, SMeterObj** pMeterObjList, int32_t* numOfInc) { SVnodeObj* pVnode = &vnodeList[pQueryMsg->vnode]; @@ -477,21 +489,24 @@ int32_t vnodeIncQueryRefCount(SQueryMeterMsg* pQueryMsg, SMeterSidExtInfo** pSid int32_t num = 0; int32_t code = TSDB_CODE_SUCCESS; - // check all meter metadata to ensure all metadata are identical. for (int32_t i = 0; i < pQueryMsg->numOfSids; ++i) { SMeterObj* pMeter = pVnode->meterList[pSids[i]->sid]; - if (pMeter == NULL || pMeter->state != TSDB_METER_STATE_READY) { - if (pMeter == NULL) { + if (pMeter == NULL || (pMeter->state > TSDB_METER_STATE_INSERT)) { + if (pMeter == NULL || vnodeIsMeterState(pMeter, TSDB_METER_STATE_DELETING)) { code = TSDB_CODE_NOT_ACTIVE_SESSION; - dError("qmsg:%p, vid:%d sid:%d, not there", pQueryMsg, pQueryMsg->vnode, pSids[i]->sid); + dError("qmsg:%p, vid:%d sid:%d, not there or will be dropped", pQueryMsg, pQueryMsg->vnode, pSids[i]->sid); vnodeSendMeterCfgMsg(pQueryMsg->vnode, pSids[i]->sid); - } else { + } else {//update or import code = TSDB_CODE_ACTION_IN_PROGRESS; dTrace("qmsg:%p, vid:%d sid:%d id:%s, it is in state:%d, wait!", pQueryMsg, pQueryMsg->vnode, pSids[i]->sid, pMeter->meterId, pMeter->state); } } else { + /* + * vnodeIsSafeToDeleteMeter will wait for this function complete, and then it can + * check if the numOfQueries is 0 or not. + */ pMeterObjList[(*numOfInc)++] = pMeter; __sync_fetch_and_add(&pMeter->numOfQueries, 1); @@ -517,7 +532,6 @@ void vnodeDecQueryRefCount(SQueryMeterMsg* pQueryMsg, SMeterObj** pMeterObjList, SMeterObj* pMeter = pMeterObjList[i]; if (pMeter != NULL) { // here, do not need to lock to perform operations - assert(pMeter->state != TSDB_METER_STATE_DELETING && pMeter->state != TSDB_METER_STATE_DELETED); __sync_fetch_and_sub(&pMeter->numOfQueries, 1); if (pMeter->numOfQueries > 0) { @@ -571,3 +585,66 @@ void vnodeUpdateQueryColumnIndex(SQuery* pQuery, SMeterObj* pMeterObj) { } } } + +int32_t vnodeTransferMeterState(SMeterObj* pMeterObj, int32_t state) { + return __sync_val_compare_and_swap(&pMeterObj->state, TSDB_METER_STATE_READY, state); +} + +void vnodeClearMeterState(SMeterObj* pMeterObj, int32_t state) { + pMeterObj->state &= (~state); +} + +bool vnodeIsMeterState(SMeterObj* pMeterObj, int32_t state) { + if (state == TSDB_METER_STATE_READY) { + return pMeterObj->state == TSDB_METER_STATE_READY; + } else if (state == TSDB_METER_STATE_DELETING) { + return pMeterObj->state >= state; + } else { + return (((pMeterObj->state) & state) == state); + } +} + +void vnodeSetMeterDeleting(SMeterObj* pMeterObj) { + if (pMeterObj == NULL) { + return; + } + + pMeterObj->state |= TSDB_METER_STATE_DELETING; +} + +bool vnodeIsSafeToDeleteMeter(SVnodeObj* pVnode, int32_t sid) { + SMeterObj* pObj = pVnode->meterList[sid]; + + if (pObj == NULL || vnodeIsMeterState(pObj, TSDB_METER_STATE_DELETED)) { + return true; + } + + int32_t prev = vnodeTransferMeterState(pObj, TSDB_METER_STATE_DELETING); + + /* + * if the meter is not in ready/deleting state, it must be in insert/import/update, + * set the deleting state and wait the procedure to be completed + */ + if (prev != TSDB_METER_STATE_READY && prev < TSDB_METER_STATE_DELETING) { + vnodeSetMeterDeleting(pObj); + + dWarn("vid:%d sid:%d id:%s, can not be deleted, state:%d, wait", pObj->vnode, pObj->sid, pObj->meterId, prev); + return false; + } + + bool ready = true; + + /* + * the query will be stopped ASAP, since the state of meter is set to TSDB_METER_STATE_DELETING, + * and new query will abort since the meter is deleted. + */ + pthread_mutex_lock(&pVnode->vmutex); + if (pObj->numOfQueries > 0) { + dWarn("vid:%d sid:%d id:%s %d queries executing on it, wait query to be finished", + pObj->vnode, pObj->sid, pObj->meterId, pObj->numOfQueries); + ready = false; + } + pthread_mutex_unlock(&pVnode->vmutex); + + return ready; +} diff --git a/src/util/src/textbuffer.c b/src/util/src/textbuffer.c index 060ed82fba..7c283c0f38 100644 --- a/src/util/src/textbuffer.c +++ b/src/util/src/textbuffer.c @@ -1532,7 +1532,7 @@ void tColModelDisplayEx(tColModel *pModel, void *pData, int32_t numOfRows, int32 } //////////////////////////////////////////////////////////////////////////////////////////// -void tColModelCompress(tColModel *pModel, tFilePage *inputBuffer, int32_t maxElemsCapacity) { +void tColModelCompact(tColModel *pModel, tFilePage *inputBuffer, int32_t maxElemsCapacity) { if (inputBuffer->numOfElems == 0 || maxElemsCapacity == inputBuffer->numOfElems) { return; } diff --git a/src/util/src/tinterpolation.c b/src/util/src/tinterpolation.c index d036989410..12f3064d7a 100644 --- a/src/util/src/tinterpolation.c +++ b/src/util/src/tinterpolation.c @@ -117,7 +117,7 @@ int32_t taosGetNumOfResWithoutLimit(SInterpolationInfo* pInterpoInfo, int64_t* p } } -bool taosHasNoneInterpoPoints(SInterpolationInfo* pInterpoInfo) { return taosNumOfRemainPoints(pInterpoInfo) > 0; } +bool taosHasRemainsDataForInterpolation(SInterpolationInfo* pInterpoInfo) { return taosNumOfRemainPoints(pInterpoInfo) > 0; } int32_t taosNumOfRemainPoints(SInterpolationInfo* pInterpoInfo) { if (pInterpoInfo->rowIdx == -1 || pInterpoInfo->numOfRawDataInRows == 0) { diff --git a/src/util/src/version.c b/src/util/src/version.c index 3eeb5d52fa..1823f152d9 100755 --- a/src/util/src/version.c +++ b/src/util/src/version.c @@ -1,4 +1,4 @@ -char version[64] = "1.6.1.2"; -char compatible_version[64] = "1.6.0.0"; -char gitinfo[128] = "ddcb2519e895c2e2101089aedaf529cee5cefe04"; -char buildinfo[512] = "Built by plum at 2019-07-29 10:41"; +char version[64] = "1.6.1.4"; +char compatible_version[64] = "1.6.1.0"; +char gitinfo[128] = "36936e19eb26b5e45107bca95394133e3ac7c3a1"; +char buildinfo[512] = "Built by slguan at 2019-08-05 09:24"; diff --git a/tests/examples/JDBC/JDBCDemo/pom.xml b/tests/examples/JDBC/JDBCDemo/pom.xml index 1c62379af9..d015c62772 100644 --- a/tests/examples/JDBC/JDBCDemo/pom.xml +++ b/tests/examples/JDBC/JDBCDemo/pom.xml @@ -63,7 +63,7 @@ com.taosdata.jdbc taos-jdbcdriver - 1.0.0 + 1.0.1