Merge remote-tracking branch 'origin/develop' into feature/crash_gen

CMakeLists.txt

@@ -1,7 +1,6 @@
 CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
 PROJECT(TDengine)
 
-SET(TD_SYNC FALSE)
 SET(TD_ACCOUNT FALSE)
 SET(TD_ADMIN FALSE)
 SET(TD_GRANT FALSE)

cmake/define.inc

@@ -1,10 +1,6 @@
 CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
 PROJECT(TDengine)
 
-IF (TD_SYNC)
-  ADD_DEFINITIONS(-D_SYNC)
-ENDIF ()
-
 IF (TD_ACCOUNT)
   ADD_DEFINITIONS(-D_ACCT)
 ENDIF ()

cmake/env.inc

@@ -9,7 +9,6 @@ SET(LIBRARY_OUTPUT_PATH ${PROJECT_BINARY_DIR}/build/lib)
 SET(EXECUTABLE_OUTPUT_PATH ${PROJECT_BINARY_DIR}/build/bin)
 SET(TD_TESTS_OUTPUT_DIR ${PROJECT_BINARY_DIR}/test)
 
-MESSAGE(STATUS "Operating system dependency directory: " ${TD_OS_DIR})
 MESSAGE(STATUS "Project source directory: " ${PROJECT_SOURCE_DIR})
 MESSAGE(STATUS "Project binary files output path: " ${PROJECT_BINARY_DIR})
 MESSAGE(STATUS "Project executable files output path: " ${EXECUTABLE_OUTPUT_PATH})

cmake/platform.inc

@@ -33,20 +33,23 @@ IF (${CPUTYPE} MATCHES "aarch32")
   SET(TD_PAGMODE_LITE TRUE)
   ADD_DEFINITIONS(-D_TD_ARM_)
   ADD_DEFINITIONS(-D_TD_ARM_32_)
+  MESSAGE(STATUS "input cpuType: aarch32")
 ELSEIF (${CPUTYPE} MATCHES "aarch64")
   SET(TD_ARM TRUE)
   SET(TD_ARM_64 TRUE)
   ADD_DEFINITIONS(-D_TD_ARM_)
   ADD_DEFINITIONS(-D_TD_ARM_64_)
+   MESSAGE(STATUS "input cpuType: aarch64")
 ELSEIF (${CPUTYPE} MATCHES "mips64")
   SET(TD_MIPS TRUE)
   SET(TD_MIPS_64 TRUE)
   ADD_DEFINITIONS(-D_TD_MIPS_)
   ADD_DEFINITIONS(-D_TD_MIPS_64_)
+  MESSAGE(STATUS "input cpuType: mips64")
 ELSEIF (${CPUTYPE} MATCHES "x64")
-  MESSAGE(STATUS "input cpuType: " ${CPUTYPE})
+  MESSAGE(STATUS "input cpuType: x64")
 ELSEIF (${CPUTYPE} MATCHES "x86")
-  MESSAGE(STATUS "input cpuType: " ${CPUTYPE})
+  MESSAGE(STATUS "input cpuType: x86")
 ELSE ()
   MESSAGE(STATUS "input cpuType: " ${CPUTYPE})
 ENDIF ()
@@ -61,13 +64,13 @@ MESSAGE(STATUS "The current os is " ${TD_OS_INFO})
 IF (${CMAKE_SYSTEM_NAME} MATCHES "Linux")
   IF (${CMAKE_SIZEOF_VOID_P} MATCHES 8)
     SET(TD_LINUX_64 TRUE)
-    SET(TD_OS_DIR ${TD_COMMUNITY_DIR}/src/os/linux)
     ADD_DEFINITIONS(-D_M_X64)
+    ADD_DEFINITIONS(-D_TD_LINUX_64)
     MESSAGE(STATUS "The current platform is Linux 64-bit")
   ELSEIF (${CMAKE_SIZEOF_VOID_P} MATCHES 4)
+    SET(TD_LINUX_32 TRUE)
+    ADD_DEFINITIONS(-D_TD_LINUX_32)
     IF (TD_ARM)
-      SET(TD_LINUX_32 TRUE)
-      SET(TD_OS_DIR ${TD_COMMUNITY_DIR}/src/os/linux)
       #ADD_DEFINITIONS(-D_M_IX86)
       MESSAGE(STATUS "The current platform is Linux 32-bit")
     ELSE ()
@@ -81,7 +84,7 @@ IF (${CMAKE_SYSTEM_NAME} MATCHES "Linux")
 ELSEIF (${CMAKE_SYSTEM_NAME} MATCHES "Darwin")
   IF (${CMAKE_SIZEOF_VOID_P} MATCHES 8)
     SET(TD_DARWIN_64 TRUE)
-    SET(TD_OS_DIR ${TD_COMMUNITY_DIR}/src/os/darwin)
+    ADD_DEFINITIONS(-D_TD_DARWIN_64)
     MESSAGE(STATUS "The current platform is Darwin 64-bit")
   ELSE ()
     MESSAGE(FATAL_ERROR "The current platform is Darwin 32-bit, not supported yet")
@@ -90,10 +93,12 @@ ELSEIF (${CMAKE_SYSTEM_NAME} MATCHES "Darwin")
 ELSEIF (${CMAKE_SYSTEM_NAME} MATCHES "Windows")
   IF (${CMAKE_SIZEOF_VOID_P} MATCHES 8)
     SET(TD_WINDOWS_64 TRUE)
-    SET(TD_OS_DIR ${TD_COMMUNITY_DIR}/src/os/windows)
     ADD_DEFINITIONS(-D_M_X64)
+    ADD_DEFINITIONS(-D_TD_WINDOWS_64)
     MESSAGE(STATUS "The current platform is Windows 64-bit")
   ELSE ()
+    SET(TD_WINDOWS_32 TRUE)
+    ADD_DEFINITIONS(-D_TD_WINDOWS_32)
     MESSAGE(FATAL_ERROR "The current platform is Windows 32-bit, not supported yet")
     EXIT ()
   ENDIF ()

documentation/webdocs/markdowndocs/TAOS SQL-ch.md

@@ -556,7 +556,7 @@ TDengine支持针对数据的聚合查询。提供支持的聚合和选择函数
     功能说明：统计表中某列的值百分比分位数。  
     返回结果数据类型： 双精度浮点数Double。  
     应用字段：不能应用在timestamp、binary、nchar、bool类型字段。  
-    说明：*k*值取值范围0≤*k*≤100，为0的时候等同于MIN，为100的时候等同于MAX。
+    说明：*k*值取值范围0≤*P*≤100，为0的时候等同于MIN，为100的时候等同于MAX。
 
 - **APERCENTILE**
     ```mysql
@@ -565,7 +565,7 @@ TDengine支持针对数据的聚合查询。提供支持的聚合和选择函数
     功能说明：统计表中某列的值百分比分位数，与PERCENTILE函数相似，但是返回近似结果。  
     返回结果数据类型： 双精度浮点数Double。  
     应用字段：不能应用在timestamp、binary、nchar、bool类型字段。  
-    说明：*k*值取值范围0≤*k*≤100，为0的时候等同于MIN，为100的时候等同于MAX。推荐使用```APERCENTILE```函数，该函数性能远胜于```PERCENTILE```函数
+    说明：*k*值取值范围0≤*P*≤100，为0的时候等同于MIN，为100的时候等同于MAX。推荐使用```APERCENTILE```函数，该函数性能远胜于```PERCENTILE```函数
     
 - **LAST_ROW**
     ```mysql

documentation/webdocs/markdowndocs/TAOS SQL.md

@@ -289,10 +289,10 @@ TDengine supports aggregations over numerical values, they are listed below:
     Applied to: table/STable. 
 
 
-- **WAVG**
+- **TWA**
   
     ```mysql
-    SELECT WAVG(field_name) FROM tb_name WHERE clause
+    SELECT TWA(field_name) FROM tb_name WHERE clause
     ```
     Function: return the time-weighted average value of a specific column  
     Return Data Type: `double`  
@@ -324,7 +324,7 @@ TDengine supports aggregations over numerical values, they are listed below:
 
 - **LEASTSQUARES**
     ```mysql
-    SELECT LEASTSQUARES(field_name) FROM tb_name [WHERE clause]
+    SELECT LEASTSQUARES(field_name, start_val, step_val) FROM tb_name [WHERE clause]
     ```
     Function: performs a linear fit to the primary timestamp and the specified column. 
     Return Data Type: return a string of the coefficient and the interception of the fitted line.  
@@ -417,6 +417,15 @@ TDengine supports aggregations over numerical values, they are listed below:
     Applied to: table/STable.  
     Note: The range of `P` is `[0, 100]`. When `P=0` , `PERCENTILE` returns the equal value as `MIN`; when `P=100`, `PERCENTILE` returns the equal value as `MAX`. 
 
+- **APERCENTILE**
+    ```mysql
+    SELECT APERCENTILE(field_name, P) FROM { tb_name | stb_name } [WHERE clause]
+    ```
+    Function: the value of the specified column below which `P` percent of the data points fall, it returns approximate value of percentile.
+    Return Data Type: double.  
+    Applicable Data Types: all types except `timestamp`, `binary`, `nchar`, `bool`. 
+    Applied to: table/STable.  
+    Note: The range of `P` is `[0, 100]`. When `P=0` , `APERCENTILE` returns the equal value as `MIN`; when `P=100`, `APERCENTILE` returns the equal value as `MAX`. `APERCENTILE` has a much better performance than `PERCENTILE`.
 
 - **LAST_ROW**
     ```mysql

src/CMakeLists.txt

@@ -10,6 +10,8 @@ ADD_SUBDIRECTORY(client)
 ADD_SUBDIRECTORY(query)
 ADD_SUBDIRECTORY(kit)
 ADD_SUBDIRECTORY(plugins)
+ADD_SUBDIRECTORY(sync)
+ADD_SUBDIRECTORY(balance)
 ADD_SUBDIRECTORY(mnode)
 ADD_SUBDIRECTORY(vnode)
 ADD_SUBDIRECTORY(tsdb)

src/balance/CMakeLists.txt

@@ -0,0 +1,16 @@
+CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
+PROJECT(TDengine)
+
+IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/common/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/mnode/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/dnode/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/sdb/inc)
+  INCLUDE_DIRECTORIES(${TD_ENTERPRISE_DIR}/src/inc)
+  INCLUDE_DIRECTORIES(inc)
+  AUX_SOURCE_DIRECTORY(src SRC)
+  ADD_LIBRARY(balance ${SRC})
+ENDIF ()

src/client/CMakeLists.txt

@@ -4,11 +4,11 @@ PROJECT(TDengine)
 INCLUDE_DIRECTORIES(inc)
 INCLUDE_DIRECTORIES(jni)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/query/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/common/inc)
 INCLUDE_DIRECTORIES(${TD_ENTERPRISE_DIR}/src/inc)
-INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
 AUX_SOURCE_DIRECTORY(src SRC)
 
 IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))

src/client/inc/tsclient.h

@@ -31,8 +31,8 @@ extern "C" {
 #include "tutil.h"
 
 #include "qExecutor.h"
+#include "qSqlparser.h"
 #include "qTsbuf.h"
-#include "qsqlparser.h"
 #include "tcmdtype.h"
 
 // forward declaration

src/client/src/tscAsync.c

@@ -430,7 +430,7 @@ void tscTableMetaCallBack(void *param, TAOS_RES *res, int code) {
   pRes->code = code;
 
   if (code != TSDB_CODE_SUCCESS) {
-    tscError("%p ge tableMeta failed, code:%s", pSql, tstrerror(code));
+    tscError("%p get tableMeta failed, code:%s", pSql, tstrerror(code));
     goto _error;
   } else {
     tscDebug("%p get tableMeta successfully", pSql);

src/client/src/tscFunctionImpl.c

@@ -2131,6 +2131,11 @@ static STopBotInfo *getTopBotOutputInfo(SQLFunctionCtx *pCtx) {
 }
 
 bool topbot_datablock_filter(SQLFunctionCtx *pCtx, int32_t functionId, const char *minval, const char *maxval) {
+  SResultInfo *pResInfo = GET_RES_INFO(pCtx);
+  if (pResInfo == NULL) {
+    return true;
+  }
+
   STopBotInfo *pTopBotInfo = getTopBotOutputInfo(pCtx);
   
   // required number of results are not reached, continue load data block

src/client/src/tscLocalMerge.c

@@ -691,9 +691,15 @@ int32_t tscLocalReducerEnvCreate(SSqlObj *pSql, tExtMemBuffer ***pMemBuffer, tOr
   
   pModel = createColumnModel(pSchema, size, capacity);
 
+  int32_t pg = DEFAULT_PAGE_SIZE;
+  int32_t overhead = sizeof(tFilePage);
+  while((pg - overhead) < pModel->rowSize * 2) {
+    pg *= 2;
+  }
+
   size_t numOfSubs = pTableMetaInfo->vgroupList->numOfVgroups;
   for (int32_t i = 0; i < numOfSubs; ++i) {
-    (*pMemBuffer)[i] = createExtMemBuffer(nBufferSizes, rlen, pModel);
+    (*pMemBuffer)[i] = createExtMemBuffer(nBufferSizes, rlen, pg, pModel);
     (*pMemBuffer)[i]->flushModel = MULTIPLE_APPEND_MODEL;
   }
 

src/client/src/tscParseInsert.c

@@ -104,7 +104,7 @@ int tsParseTime(SSQLToken *pToken, int64_t *time, char **next, char *error, int1
   } else if (strncmp(pToken->z, "0", 1) == 0 && pToken->n == 1) {
     // do nothing
   } else if (pToken->type == TK_INTEGER) {
-    useconds = str2int64(pToken->z);
+    useconds = tsosStr2int64(pToken->z);
   } else {
     // strptime("2001-11-12 18:31:01", "%Y-%m-%d %H:%M:%S", &tm);
     if (taosParseTime(pToken->z, time, pToken->n, timePrec, tsDaylight) != TSDB_CODE_SUCCESS) {

src/client/src/tscSubquery.c

@@ -1505,12 +1505,11 @@ static int32_t tscReissueSubquery(SRetrieveSupport *trsupport, SSqlObj *pSql, in
 
   SSqlObj *pNew = tscCreateSqlObjForSubquery(trsupport->pParentSql, trsupport, pSql);
 
-  // todo add to async res or not??
   if (pNew == NULL) {
-    tscError("%p sub:%p failed to create new subquery due to out of memory, abort retry, vgId:%d, orderOfSub:%d",
+    tscError("%p sub:%p failed to create new subquery due to error:%s, abort retry, vgId:%d, orderOfSub:%d",
-             trsupport->pParentSql, pSql, pVgroup->vgId, trsupport->subqueryIndex);
+             trsupport->pParentSql, pSql, tstrerror(terrno), pVgroup->vgId, trsupport->subqueryIndex);
 
-    pParentSql->res.code = TSDB_CODE_TSC_OUT_OF_MEMORY;
+    pParentSql->res.code = terrno;
     trsupport->numOfRetry = MAX_NUM_OF_SUBQUERY_RETRY;
 
     return pParentSql->res.code;

src/client/src/tscSystem.c

@@ -148,7 +148,7 @@ void taos_init_imp() {
   refreshTime = refreshTime < 10 ? 10 : refreshTime;
 
   if (tscCacheHandle == NULL) {
-    tscCacheHandle = taosCacheInit(TSDB_DATA_TYPE_BINARY, refreshTime, false, NULL, "client");
+    tscCacheHandle = taosCacheInit(TSDB_DATA_TYPE_BINARY, refreshTime, false, NULL, "tableMeta");
   }
 
   tscDebug("client is initialized successfully");
@@ -220,7 +220,7 @@ static int taos_options_imp(TSDB_OPTION option, const char *pStr) {
 
         if (strlen(tsLocale) == 0) { // locale does not set yet
           char* defaultLocale = setlocale(LC_CTYPE, "");
-          tstrncpy(tsLocale, defaultLocale, sizeof(tsLocale));
+          tstrncpy(tsLocale, defaultLocale, TSDB_LOCALE_LEN);
         }
 
         // set the user specified locale
@@ -234,7 +234,7 @@ static int taos_options_imp(TSDB_OPTION option, const char *pStr) {
           tscInfo("failed to set locale:%s, current locale:%s", pStr, tsLocale);
         }
 
-        tstrncpy(tsLocale, locale, sizeof(tsLocale));
+        tstrncpy(tsLocale, locale, TSDB_LOCALE_LEN);
 
         char *charset = strrchr(tsLocale, sep);
         if (charset != NULL) {
@@ -249,7 +249,7 @@ static int taos_options_imp(TSDB_OPTION option, const char *pStr) {
               tscInfo("charset changed from %s to %s", tsCharset, charset);
             }
 
-            tstrncpy(tsCharset, charset, sizeof(tsCharset));
+            tstrncpy(tsCharset, charset, TSDB_LOCALE_LEN);
             cfg->cfgStatus = TAOS_CFG_CSTATUS_OPTION;
 
           } else {
@@ -286,7 +286,7 @@ static int taos_options_imp(TSDB_OPTION option, const char *pStr) {
             tscInfo("charset changed from %s to %s", tsCharset, pStr);
           }
 
-          tstrncpy(tsCharset, pStr, sizeof(tsCharset));
+          tstrncpy(tsCharset, pStr, TSDB_LOCALE_LEN);
           cfg->cfgStatus = TAOS_CFG_CSTATUS_OPTION;
         } else {
           tscInfo("charset:%s not valid", pStr);
@@ -304,7 +304,7 @@ static int taos_options_imp(TSDB_OPTION option, const char *pStr) {
       assert(cfg != NULL);
     
       if (cfg->cfgStatus <= TAOS_CFG_CSTATUS_OPTION) {
-        tstrncpy(tsTimezone, pStr, sizeof(tsTimezone));
+        tstrncpy(tsTimezone, pStr, TSDB_TIMEZONE_LEN);
         tsSetTimeZone();
         cfg->cfgStatus = TAOS_CFG_CSTATUS_OPTION;
         tscDebug("timezone set:%s, input:%s by taos_options", tsTimezone, pStr);

src/client/src/tscUtil.c

@@ -356,9 +356,9 @@ void tscPartiallyFreeSqlObj(SSqlObj* pSql) {
   
   // pSql->sqlstr will be used by tscBuildQueryStreamDesc
   if (pObj->signature == pObj) {
-    pthread_mutex_lock(&pObj->mutex);
+    //pthread_mutex_lock(&pObj->mutex);
     tfree(pSql->sqlstr);
-    pthread_mutex_unlock(&pObj->mutex);
+    //pthread_mutex_unlock(&pObj->mutex);
   }
   
   tscFreeSqlResult(pSql);
@@ -1675,6 +1675,7 @@ SSqlObj* createSubqueryObj(SSqlObj* pSql, int16_t tableIndex, void (*fp)(), void
   SSqlObj* pNew = (SSqlObj*)calloc(1, sizeof(SSqlObj));
   if (pNew == NULL) {
     tscError("%p new subquery failed, tableIndex:%d", pSql, tableIndex);
+    terrno = TSDB_CODE_TSC_OUT_OF_MEMORY;
     return NULL;
   }
   
@@ -1688,6 +1689,7 @@ SSqlObj* createSubqueryObj(SSqlObj* pSql, int16_t tableIndex, void (*fp)(), void
     tscError("%p new subquery failed, tableIndex:%d, vgroupIndex:%d", pSql, tableIndex, pTableMetaInfo->vgroupIndex);
 
     free(pNew);
+    terrno = TSDB_CODE_TSC_OUT_OF_MEMORY;
     return NULL;
   }
 
@@ -1706,6 +1708,7 @@ SSqlObj* createSubqueryObj(SSqlObj* pSql, int16_t tableIndex, void (*fp)(), void
 
   if (tscAddSubqueryInfo(pnCmd) != TSDB_CODE_SUCCESS) {
     tscFreeSqlObj(pNew);
+    terrno = TSDB_CODE_TSC_OUT_OF_MEMORY;
     return NULL;
   }
 
@@ -1743,6 +1746,7 @@ SSqlObj* createSubqueryObj(SSqlObj* pSql, int16_t tableIndex, void (*fp)(), void
   if (tscAllocPayload(pnCmd, TSDB_DEFAULT_PAYLOAD_SIZE) != TSDB_CODE_SUCCESS) {
     tscError("%p new subquery failed, tableIndex:%d, vgroupIndex:%d", pSql, tableIndex, pTableMetaInfo->vgroupIndex);
     tscFreeSqlObj(pNew);
+    terrno = TSDB_CODE_TSC_OUT_OF_MEMORY;
     return NULL;
   }
   
@@ -1827,8 +1831,16 @@ SSqlObj* createSubqueryObj(SSqlObj* pSql, int16_t tableIndex, void (*fp)(), void
   }
 
   if (pFinalInfo->pTableMeta == NULL) {
-    tscError("%p new subquery failed for get tableMeta is NULL from cache", pSql);
+    tscError("%p new subquery failed since no tableMeta in cache, name:%s", pSql, name);
     tscFreeSqlObj(pNew);
+
+    if (pPrevSql != NULL) {
+      assert(pPrevSql->res.code != TSDB_CODE_SUCCESS);
+      terrno = pPrevSql->res.code;
+    } else {
+      terrno = TSDB_CODE_TSC_APP_ERROR;
+    }
+
     return NULL;
   }
   

src/common/CMakeLists.txt

@@ -1,12 +1,11 @@
 CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
 PROJECT(TDengine)
 
-IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
-  INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
-  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
-  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
+INCLUDE_DIRECTORIES(inc)
-  INCLUDE_DIRECTORIES(inc)
+
-  AUX_SOURCE_DIRECTORY(src SRC)
+AUX_SOURCE_DIRECTORY(src SRC)
-  ADD_LIBRARY(common ${SRC})
+ADD_LIBRARY(common ${SRC})
-  TARGET_LINK_LIBRARIES(common tutil)
+TARGET_LINK_LIBRARIES(common tutil)
-ENDIF ()

src/common/inc/tglobal.h

@@ -43,9 +43,9 @@ extern uint32_t tsMaxTmrCtrl;
 extern float    tsNumOfThreadsPerCore;
 extern float    tsRatioOfQueryThreads;
 extern int8_t   tsDaylight;
-extern char     tsTimezone[64];
+extern char     tsTimezone[];
-extern char     tsLocale[64];
+extern char     tsLocale[];
-extern char     tsCharset[64];  // default encode string
+extern char     tsCharset[];  // default encode string
 extern int32_t  tsEnableCoreFile;
 extern int32_t  tsCompressMsgSize;
 

src/common/src/tglobal.c

@@ -50,7 +50,7 @@ int32_t tsShellActivityTimer = 3;  // second
 float   tsNumOfThreadsPerCore = 1.0;
 float   tsRatioOfQueryThreads = 0.5;
 int8_t  tsDaylight = 0;
-char    tsTimezone[64] = {0};
+char    tsTimezone[TSDB_TIMEZONE_LEN] = {0};
 char    tsLocale[TSDB_LOCALE_LEN] = {0};
 char    tsCharset[TSDB_LOCALE_LEN] = {0};  // default encode string
 int32_t tsEnableCoreFile = 0;

src/cq/CMakeLists.txt

@@ -1,8 +1,8 @@
 CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
 PROJECT(TDengine)
 
-INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/common/inc)
 INCLUDE_DIRECTORIES(inc)

src/cq/test/CMakeLists.txt

@@ -1,17 +1,12 @@
 CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
 PROJECT(TDengine)
 
-IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
-  INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
-  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
-  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/common/inc)
-  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/common/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/cq/inc)
-  INCLUDE_DIRECTORIES(../inc)
-
-  LIST(APPEND CQTEST_SRC ./cqtest.c)
-  ADD_EXECUTABLE(cqtest ${CQTEST_SRC})
-  TARGET_LINK_LIBRARIES(cqtest tcq)
-
-ENDIF ()
-
 
+LIST(APPEND CQTEST_SRC ./cqtest.c)
+ADD_EXECUTABLE(cqtest ${CQTEST_SRC})
+TARGET_LINK_LIBRARIES(cqtest tcq)

src/dnode/CMakeLists.txt

@@ -2,8 +2,8 @@ CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
 PROJECT(TDengine)
 
 IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
-  INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/query/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/mnode/inc)
@@ -16,7 +16,7 @@ IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
   AUX_SOURCE_DIRECTORY(src SRC)
 
   ADD_EXECUTABLE(taosd ${SRC})
-  TARGET_LINK_LIBRARIES(taosd mnode taos_static monitor http mqtt tsdb twal vnode cJson lz4)
+  TARGET_LINK_LIBRARIES(taosd mnode taos_static monitor http mqtt tsdb twal vnode cJson lz4 balance sync)
 
   IF (TD_ACCOUNT)
     TARGET_LINK_LIBRARIES(taosd account)
@@ -26,10 +26,6 @@ IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
     TARGET_LINK_LIBRARIES(taosd grant)
   ENDIF ()
 
-  IF (TD_SYNC)
-    TARGET_LINK_LIBRARIES(taosd balance sync)
-  ENDIF ()
-
   SET(PREPARE_ENV_CMD "prepare_env_cmd")
   SET(PREPARE_ENV_TARGET "prepare_env_target")
   ADD_CUSTOM_COMMAND(OUTPUT ${PREPARE_ENV_CMD}

src/dnode/src/dnodeSystem.c

@@ -20,7 +20,6 @@
 #include "tglobal.h"
 #include "dnodeInt.h"
 #include "dnodeMain.h"
-#include "tfile.h"
 
 static void signal_handler(int32_t signum, siginfo_t *sigInfo, void *context);
 static sem_t exitSem;
@@ -40,7 +39,7 @@ int32_t main(int32_t argc, char *argv[]) {
         exit(EXIT_FAILURE);
       }
     } else if (strcmp(argv[i], "-V") == 0) {
-#ifdef _SYNC
+#ifdef _ACCT
       char *versionStr = "enterprise";
 #else
       char *versionStr = "community";

src/dnode/src/dnodeVRead.c

@@ -49,7 +49,7 @@ static taos_qset       readQset;
 int32_t dnodeInitVnodeRead() {
   readQset = taosOpenQset();
 
-  readPool.min = 2;
+  readPool.min = tsNumOfCores;
   readPool.max = tsNumOfCores * tsNumOfThreadsPerCore;
   if (readPool.max <= readPool.min * 2) readPool.max = 2 * readPool.min;
   readPool.readWorker = (SReadWorker *)calloc(sizeof(SReadWorker), readPool.max);
@@ -206,10 +206,14 @@ static void *dnodeProcessReadQueue(void *param) {
            taosMsg[pReadMsg->rpcMsg.msgType], type);
     int32_t code = vnodeProcessRead(pVnode, pReadMsg);
 
-    if (type == TAOS_QTYPE_RPC) {
+    if (type == TAOS_QTYPE_RPC && code != TSDB_CODE_QRY_NOT_READY) {
       dnodeSendRpcReadRsp(pVnode, pReadMsg, code);
     } else {
-      dnodeDispatchNonRspMsg(pVnode, pReadMsg, code);
+      if (code == TSDB_CODE_QRY_HAS_RSP) {
+        dnodeSendRpcReadRsp(pVnode, pReadMsg, TSDB_CODE_SUCCESS);
+      } else {
+        dnodeDispatchNonRspMsg(pVnode, pReadMsg, code);
+      }
     }
 
     taosFreeQitem(pReadMsg);

src/inc/query.h

@@ -28,7 +28,7 @@ typedef void* qinfo_t;
  * @param qinfo
  * @return
  */
-int32_t qCreateQueryInfo(void* tsdb, int32_t vgId, SQueryTableMsg* pQueryTableMsg, void* param, qinfo_t* qinfo);
+int32_t qCreateQueryInfo(void* tsdb, int32_t vgId, SQueryTableMsg* pQueryTableMsg, qinfo_t* qinfo);
 
 
 /**
@@ -38,7 +38,10 @@ int32_t qCreateQueryInfo(void* tsdb, int32_t vgId, SQueryTableMsg* pQueryTableMs
  * @param qinfo
  * @return
  */
-void qTableQuery(qinfo_t qinfo);
+bool qTableQuery(qinfo_t qinfo);
+
+void* pGetRspMsg(qinfo_t qinfo);
+
 
 /**
  * Retrieve the produced results information, if current query is not paused or completed,
@@ -48,7 +51,7 @@ void qTableQuery(qinfo_t qinfo);
  * @param qinfo
  * @return
  */
-int32_t qRetrieveQueryResultInfo(qinfo_t qinfo);
+int32_t qRetrieveQueryResultInfo(qinfo_t qinfo, bool* buildRes, void* pRspContext);
 
 /**
  *
@@ -60,16 +63,9 @@ int32_t qRetrieveQueryResultInfo(qinfo_t qinfo);
  * @param contLen payload length
  * @return
  */
-int32_t qDumpRetrieveResult(qinfo_t qinfo, SRetrieveTableRsp** pRsp, int32_t* contLen);
+int32_t qDumpRetrieveResult(qinfo_t qinfo, SRetrieveTableRsp** pRsp, int32_t* contLen, bool* continueExec);
 
-/**
+void* qGetResultRetrieveMsg(qinfo_t qinfo);
- * Decide if more results will be produced or not, NOTE: this function will increase the ref count of QInfo,
- * so it can be only called once for each retrieve
- *
- * @param qinfo
- * @return
- */
-bool qHasMoreResultsToRetrieve(qinfo_t qinfo);
 
 /**
  * kill current ongoing query and free query handle automatically

src/inc/taosdef.h

@@ -250,7 +250,7 @@ void tsDataSwap(void *pLeft, void *pRight, int32_t type, int32_t size);
 #define TSDB_STATE_LEN            20
 #define TSDB_COUNTRY_LEN          20
 #define TSDB_LOCALE_LEN           64
-#define TSDB_TIMEZONE_LEN         64
+#define TSDB_TIMEZONE_LEN         96
 #define TSDB_LABEL_LEN            8 
 
 #define TSDB_FQDN_LEN             128

src/inc/taoserror.h

@@ -216,6 +216,8 @@ TAOS_DEFINE_ERROR(TSDB_CODE_QRY_OUT_OF_MEMORY,            0, 0x0703, "query out
 TAOS_DEFINE_ERROR(TSDB_CODE_QRY_APP_ERROR,                0, 0x0704, "query app error")
 TAOS_DEFINE_ERROR(TSDB_CODE_QRY_DUP_JOIN_KEY,             0, 0x0705, "query duplicated join key")
 TAOS_DEFINE_ERROR(TSDB_CODE_QRY_EXCEED_TAGS_LIMIT,        0, 0x0706, "query tag conditon too many")
+TAOS_DEFINE_ERROR(TSDB_CODE_QRY_NOT_READY,                0, 0x0707, "query not ready")
+TAOS_DEFINE_ERROR(TSDB_CODE_QRY_HAS_RSP,                  0, 0x0708, "query should response")
 
 // grant
 TAOS_DEFINE_ERROR(TSDB_CODE_GRANT_EXPIRED,                0, 0x0800, "grant expired")

src/kit/shell/CMakeLists.txt

@@ -2,11 +2,11 @@ CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
 PROJECT(TDengine)
 
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/client/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/query/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/common/inc)
-INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
 INCLUDE_DIRECTORIES(inc)
 
 IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))

src/kit/taosdemo/CMakeLists.txt

@@ -2,9 +2,9 @@ CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
 PROJECT(TDengine)
 
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/client/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
-INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
 INCLUDE_DIRECTORIES(inc)
 
 IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))

src/kit/taosdump/CMakeLists.txt

@@ -2,11 +2,11 @@ CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
 PROJECT(TDengine)
 
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/common/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/client/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/query/inc)
-INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
 INCLUDE_DIRECTORIES(inc)
 
 IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))

src/kit/taosdump/taosdump.c

@@ -545,7 +545,7 @@ int32_t taosSaveAllNormalTableToTempFile(TAOS *taosCon, char*meter, char* metric
   strcpy(tableRecord.name, meter);
   strcpy(tableRecord.metric, metric);
 
-  twrite(*fd, &tableRecord, sizeof(STableRecord));
+  taosTWrite(*fd, &tableRecord, sizeof(STableRecord));
   return 0;
 }
 
@@ -600,7 +600,7 @@ int32_t taosSaveTableOfMetricToTempFile(TAOS *taosCon, char* metric, struct argu
     strncpy(tableRecord.name, (char *)row[0], fields[0].bytes);
     strcpy(tableRecord.metric, metric);
 
-    twrite(fd, &tableRecord, sizeof(STableRecord));
+    taosTWrite(fd, &tableRecord, sizeof(STableRecord));
 
     numOfTable++;
 
@@ -1196,7 +1196,7 @@ int32_t taosDumpCreateSuperTableClause(TAOS* taosCon, char* dbName, FILE *fp)
   while ((row = taos_fetch_row(tmpResult)) != NULL) {  
     memset(&tableRecord, 0, sizeof(STableRecord));
     strncpy(tableRecord.name, (char *)row[TSDB_SHOW_TABLES_NAME_INDEX], fields[TSDB_SHOW_TABLES_NAME_INDEX].bytes);
-    twrite(fd, &tableRecord, sizeof(STableRecord));
+    taosTWrite(fd, &tableRecord, sizeof(STableRecord));
   }  
   
   taos_free_result(tmpResult);
@@ -1282,7 +1282,7 @@ int taosDumpDb(SDbInfo *dbInfo, struct arguments *arguments, FILE *fp, TAOS *tao
     strncpy(tableRecord.name, (char *)row[TSDB_SHOW_TABLES_NAME_INDEX], fields[TSDB_SHOW_TABLES_NAME_INDEX].bytes);
     strncpy(tableRecord.metric, (char *)row[TSDB_SHOW_TABLES_METRIC_INDEX], fields[TSDB_SHOW_TABLES_METRIC_INDEX].bytes);
 
-    twrite(fd, &tableRecord, sizeof(STableRecord));
+    taosTWrite(fd, &tableRecord, sizeof(STableRecord));
 
     numOfTable++;
 

src/kit/taosmigrate/CMakeLists.txt

@@ -2,8 +2,8 @@ CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
 PROJECT(TDengine)
 
 IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
-  INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/mnode/inc) 
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/vnode/inc)

src/mnode/CMakeLists.txt

@@ -2,8 +2,8 @@ CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
 PROJECT(TDengine)
 
 IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
-  INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/common/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/query/inc)

src/mnode/src/mnodeBalance.c

@@ -1,70 +0,0 @@
-/*
- * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
- *
- * This program is free software: you can use, redistribute, and/or modify
- * it under the terms of the GNU Affero General Public License, version 3
- * or later ("AGPL"), as published by the Free Software Foundation.
- *
- * This program is distributed in the hope that it will be useful, but WITHOUT
- * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
- * FITNESS FOR A PARTICULAR PURPOSE.
- *
- * You should have received a copy of the GNU Affero General Public License
- * along with this program. If not, see <http://www.gnu.org/licenses/>.
- */
-
-#define _DEFAULT_SOURCE
-#include "os.h"
-#include "tglobal.h"
-#include "mnodeDef.h"
-#include "mnodeInt.h"
-#include "mnodeDnode.h"
-#include "mnodeSdb.h"
-
-#ifndef _SYNC
-
-int32_t balanceInit() { return TSDB_CODE_SUCCESS; }
-void    balanceCleanUp() {}
-void    balanceAsyncNotify() {}
-void    balanceSyncNotify() {}
-void    balanceReset() {}
-int32_t balanceAlterDnode(struct SDnodeObj *pDnode, int32_t vnodeId, int32_t dnodeId) { return TSDB_CODE_SYN_NOT_ENABLED; }
-
-int32_t balanceAllocVnodes(SVgObj *pVgroup) {
-  void *     pIter = NULL;
-  SDnodeObj *pDnode = NULL;
-  SDnodeObj *pSelDnode = NULL;
-  float      vnodeUsage = 1000.0;
-
-  while (1) {
-    pIter = mnodeGetNextDnode(pIter, &pDnode);
-    if (pDnode == NULL) break;
-
-    if (pDnode->numOfCores > 0 && pDnode->openVnodes < TSDB_MAX_VNODES) {
-      float openVnodes = pDnode->openVnodes;
-      if (pDnode->isMgmt) openVnodes += tsMnodeEqualVnodeNum;
-
-      float usage = openVnodes / pDnode->numOfCores;
-      if (usage <= vnodeUsage) {
-        pSelDnode = pDnode;
-        vnodeUsage = usage;
-      }
-    }
-    mnodeDecDnodeRef(pDnode);
-  }
-
-  sdbFreeIter(pIter);
-
-  if (pSelDnode == NULL) {
-    mError("failed to alloc vnode to vgroup");
-    return TSDB_CODE_MND_NO_ENOUGH_DNODES;
-  }
-
-  pVgroup->vnodeGid[0].dnodeId = pSelDnode->dnodeId;
-  pVgroup->vnodeGid[0].pDnode = pSelDnode;
-
-  mDebug("dnode:%d, alloc one vnode to vgroup, openVnodes:%d", pSelDnode->dnodeId, pSelDnode->openVnodes);
-  return TSDB_CODE_SUCCESS;
-}
-
-#endif 

src/mnode/src/mnodeDb.c

@@ -301,13 +301,6 @@ static int32_t mnodeCheckDbCfg(SDbCfg *pCfg) {
     return TSDB_CODE_MND_INVALID_DB_OPTION;
   }
 
-#ifndef _SYNC
-  if (pCfg->replications != 1) {
-    mError("invalid db option replications:%d can only be 1 in this version", pCfg->replications);
-    return TSDB_CODE_MND_INVALID_DB_OPTION;
-  }
-#endif
-
   return TSDB_CODE_SUCCESS;
 }
 

src/mnode/src/mnodeDnode.c

@@ -79,9 +79,6 @@ static int32_t mnodeDnodeActionInsert(SSdbOper *pOper) {
 static int32_t mnodeDnodeActionDelete(SSdbOper *pOper) {
   SDnodeObj *pDnode = pOper->pObj;
  
-#ifndef _SYNC 
-  mnodeDropAllDnodeVgroups(pDnode);
-#endif  
   mnodeDropMnodeLocal(pDnode->dnodeId);
   balanceAsyncNotify();
 
@@ -552,12 +549,7 @@ static int32_t mnodeDropDnodeByEp(char *ep, SMnodeMsg *pMsg) {
 
   mInfo("dnode:%d, start to drop it", pDnode->dnodeId);
 
-#ifndef _SYNC
-  int32_t code = mnodeDropDnode(pDnode, pMsg);
-#else
   int32_t code = balanceDropDnode(pDnode);
-#endif
-
   mnodeDecDnodeRef(pDnode);
   return code;
 }

src/mnode/src/mnodeProfile.c

@@ -68,7 +68,7 @@ int32_t mnodeInitProfile() {
   mnodeAddWriteMsgHandle(TSDB_MSG_TYPE_CM_KILL_STREAM, mnodeProcessKillStreamMsg);
   mnodeAddWriteMsgHandle(TSDB_MSG_TYPE_CM_KILL_CONN, mnodeProcessKillConnectionMsg);
 
-  tsMnodeConnCache = taosCacheInit(TSDB_DATA_TYPE_INT, CONN_CHECK_TIME, false, mnodeFreeConn, "conn");
+  tsMnodeConnCache = taosCacheInit(TSDB_DATA_TYPE_INT, CONN_CHECK_TIME, true, mnodeFreeConn, "conn");
   return 0;
 }
 
@@ -119,7 +119,7 @@ SConnObj *mnodeAccquireConn(int32_t connId, char *user, uint32_t ip, uint16_t po
     return NULL;
   }
 
-  if (pConn->ip != ip || pConn->port != port /* || strcmp(pConn->user, user) != 0 */) {
+  if (/* pConn->ip != ip || */ pConn->port != port /* || strcmp(pConn->user, user) != 0 */) {
     mError("connId:%d, incoming conn user:%s ip:%s:%u, not match exist conn user:%s ip:%s:%u", connId, user,
            taosIpStr(ip), port, pConn->user, taosIpStr(pConn->ip), pConn->port);
     taosCacheRelease(tsMnodeConnCache, (void **)&pConn, false);

src/os/CMakeLists.txt

@@ -1,6 +1,16 @@
 CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
 PROJECT(TDengine)
 
-ADD_SUBDIRECTORY(linux)
+IF (TD_LINUX_64)
-ADD_SUBDIRECTORY(windows)
+  ADD_SUBDIRECTORY(src/linux64)
-ADD_SUBDIRECTORY(darwin)
+ELSEIF (TD_LINUX_32)
+  ADD_SUBDIRECTORY(src/linux32)
+ELSEIF (TD_DARWIN_64)
+  ADD_SUBDIRECTORY(src/darwin64)  
+ELSEIF (TD_WINDOWS_64)
+  ADD_SUBDIRECTORY(src/windows64)
+ELSEIF (TD_WINDOWS_32)
+  ADD_SUBDIRECTORY(src/windows32)
+ENDIF ()
+
+ADD_SUBDIRECTORY(src/detail)

src/os/darwin/CMakeLists.txt

@@ -1,9 +0,0 @@
-CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
-PROJECT(TDengine)
-
-IF (TD_DARWIN_64)
-  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
-  INCLUDE_DIRECTORIES(inc)
-  AUX_SOURCE_DIRECTORY(src SRC)
-  ADD_LIBRARY(os ${SRC})
-ENDIF ()

src/os/inc/os.h

@@ -0,0 +1,53 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef TDENGINE_OS_H
+#define TDENGINE_OS_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef _TD_DARWIN_64
+#include "osDarwin64.h"
+#endif
+
+#ifdef _TD_LINUX_64
+#include "osLinux64.h"
+#endif
+
+#ifdef _TD_LINUX_32
+#include "osLinux32.h"
+#endif
+
+#ifdef _TD_ALPINE
+#include "osAlpine.h"
+#endif
+
+#ifdef _TD_WINDOWS_64
+#include "osWindows64.h"
+#endif
+
+#ifdef _TD_WINDOWS_32
+#include "osWindows32.h"
+#endif
+
+#include "osSpec.h"
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

src/os/inc/osAlpine.h

@@ -0,0 +1,91 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef TDENGINE_OS_LINUX64_H
+#define TDENGINE_OS_LINUX64_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#include <argp.h>
+#include <arpa/inet.h>
+#include <assert.h>
+#include <ctype.h>
+#include <dirent.h>
+#include <endian.h>
+#include <errno.h>
+#include <float.h>
+#include <ifaddrs.h>
+#include <libgen.h>
+#include <limits.h>
+#include <locale.h>
+#include <math.h>
+#include <netdb.h>
+#include <netinet/in.h>
+#include <netinet/ip.h>
+#include <netinet/tcp.h>
+#include <netinet/udp.h>
+#include <pthread.h>
+#include <pwd.h>
+#include <regex.h>
+#include <semaphore.h>
+#include <signal.h>
+#include <stdarg.h>
+#include <stdbool.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+#include <strings.h>
+#include <sys/epoll.h>
+#include <sys/eventfd.h>
+#include <sys/file.h>
+#include <sys/ioctl.h>
+#include <sys/mman.h>
+#include <sys/sendfile.h>
+#include <sys/socket.h>
+#include <sys/stat.h>
+#include <sys/syscall.h>
+#include <sys/statvfs.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <sys/uio.h>
+#include <sys/un.h>
+#include <syslog.h>
+#include <termios.h>
+#include <unistd.h>
+#include <wchar.h>
+#include <wordexp.h>
+#include <wctype.h>
+#include <inttypes.h>
+#include <fcntl.h>
+#include <sys/utsname.h>
+#include <sys/resource.h>
+#include <linux/sysctl.h>
+
+typedef int(*__compar_fn_t)(const void *, const void *);
+void  error (int, int, const char *);
+#ifndef PTHREAD_MUTEX_RECURSIVE_NP
+  #define  PTHREAD_MUTEX_RECURSIVE_NP PTHREAD_MUTEX_RECURSIVE
+#endif
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

src/os/inc/osDarwin64.h

@@ -201,6 +201,8 @@ int tsem_destroy(dispatch_semaphore_t *sem);
 
 void osInit();
 
+ssize_t tread(int fd, void *buf, size_t count);
+
 ssize_t twrite(int fd, void *buf, size_t n);
 
 char *taosCharsetReplace(char *charsetstr);
@@ -233,6 +235,8 @@ int fsendfile(FILE* out_file, FILE* in_file, int64_t* offset, int32_t count);
 
 void taosSetCoreDump();
 
+int tSystem(const char * cmd);
+
 typedef int(*__compar_fn_t)(const void *, const void *);
 
 // for send function in tsocket.c
@@ -255,6 +259,17 @@ typedef int(*__compar_fn_t)(const void *, const void *);
 #define BUILDIN_CLZ(val)  __builtin_clz(val)
 #define BUILDIN_CTZ(val)  __builtin_ctz(val)
 
+#undef threadlocal
+#ifdef _ISOC11_SOURCE
+  #define threadlocal _Thread_local
+#elif defined(__APPLE__)
+  #define threadlocal
+#elif defined(__GNUC__) && !defined(threadlocal)
+  #define threadlocal __thread
+#else
+  #define threadlocal
+#endif
+
 #ifdef __cplusplus
 }
 #endif

src/os/inc/osLinux32.h

@@ -0,0 +1,85 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef TDENGINE_OS_LINUX64_H
+#define TDENGINE_OS_LINUX64_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#include <argp.h>
+#include <arpa/inet.h>
+#include <assert.h>
+#include <ctype.h>
+#include <dirent.h>
+#include <endian.h>
+#include <errno.h>
+#include <float.h>
+#include <ifaddrs.h>
+#include <libgen.h>
+#include <limits.h>
+#include <locale.h>
+#include <math.h>
+#include <netdb.h>
+#include <netinet/in.h>
+#include <netinet/ip.h>
+#include <netinet/tcp.h>
+#include <netinet/udp.h>
+#include <pthread.h>
+#include <pwd.h>
+#include <regex.h>
+#include <semaphore.h>
+#include <signal.h>
+#include <stdarg.h>
+#include <stdbool.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+#include <strings.h>
+#include <sys/epoll.h>
+#include <sys/eventfd.h>
+#include <sys/file.h>
+#include <sys/ioctl.h>
+#include <sys/mman.h>
+#include <sys/sendfile.h>
+#include <sys/socket.h>
+#include <sys/stat.h>
+#include <sys/syscall.h>
+#include <sys/statvfs.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <sys/uio.h>
+#include <sys/un.h>
+#include <syslog.h>
+#include <termios.h>
+#include <unistd.h>
+#include <wchar.h>
+#include <wordexp.h>
+#include <wctype.h>
+#include <inttypes.h>
+#include <fcntl.h>
+#include <sys/utsname.h>
+#include <sys/resource.h>
+#include <error.h>
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

src/os/inc/osLinux64.h

@@ -0,0 +1,85 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef TDENGINE_OS_LINUX64_H
+#define TDENGINE_OS_LINUX64_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <argp.h>
+#include <arpa/inet.h>
+#include <assert.h>
+#include <ctype.h>
+#include <dirent.h>
+#include <endian.h>
+#include <errno.h>
+#include <float.h>
+#include <ifaddrs.h>
+#include <libgen.h>
+#include <limits.h>
+#include <locale.h>
+#include <math.h>
+#include <netdb.h>
+#include <netinet/in.h>
+#include <netinet/ip.h>
+#include <netinet/tcp.h>
+#include <netinet/udp.h>
+#include <pthread.h>
+#include <pwd.h>
+#include <regex.h>
+#include <semaphore.h>
+#include <signal.h>
+#include <stdarg.h>
+#include <stdbool.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+#include <strings.h>
+#include <sys/epoll.h>
+#include <sys/eventfd.h>
+#include <sys/file.h>
+#include <sys/ioctl.h>
+#include <sys/mman.h>
+#include <sys/sendfile.h>
+#include <sys/socket.h>
+#include <sys/stat.h>
+#include <sys/syscall.h>
+#include <sys/statvfs.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <sys/uio.h>
+#include <sys/un.h>
+#include <syslog.h>
+#include <termios.h>
+#include <unistd.h>
+#include <wchar.h>
+#include <wordexp.h>
+#include <wctype.h>
+#include <inttypes.h>
+#include <fcntl.h>
+#include <sys/utsname.h>
+#include <sys/resource.h>
+#include <error.h>
+#include <linux/sysctl.h>
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

src/os/inc/osSpec.h

@@ -0,0 +1,240 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef TDENGINE_OS_SPEC_H
+#define TDENGINE_OS_SPEC_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef TAOS_OS_FUNC_MATH
+  #define SWAP(a, b, c)        \
+    do {                       \
+        typeof(a) __tmp = (a); \
+        (a) = (b);             \
+        (b) = __tmp;           \
+    } while (0)
+
+  #define MAX(a, b)              \
+    ({                           \
+        typeof(a) __a = (a);     \
+        typeof(b) __b = (b);     \
+        (__a > __b) ? __a : __b; \
+    })
+
+  #define MIN(a, b)              \
+    ({                           \
+        typeof(a) __a = (a);     \
+        typeof(b) __b = (b);     \
+        (__a < __b) ? __a : __b; \
+    })
+#endif
+
+#ifndef TAOS_OS_DEF_TIME
+  #define MILLISECOND_PER_SECOND ((int64_t)1000L)
+#endif
+
+#ifndef TAOS_OS_FUNC_SEMPHONE
+  #define tsem_t sem_t
+  #define tsem_init sem_init
+  #define tsem_wait sem_wait
+  #define tsem_post sem_post
+  #define tsem_destroy sem_destroy
+#endif
+
+#ifndef TAOS_OS_FUNC_ATOMIC
+  #define atomic_load_8(ptr) __atomic_load_n((ptr), __ATOMIC_SEQ_CST)
+  #define atomic_load_16(ptr) __atomic_load_n((ptr), __ATOMIC_SEQ_CST)
+  #define atomic_load_32(ptr) __atomic_load_n((ptr), __ATOMIC_SEQ_CST)
+  #define atomic_load_64(ptr) __atomic_load_n((ptr), __ATOMIC_SEQ_CST)
+  #define atomic_load_ptr(ptr) __atomic_load_n((ptr), __ATOMIC_SEQ_CST)
+
+  #define atomic_store_8(ptr, val) __atomic_store_n((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_store_16(ptr, val) __atomic_store_n((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_store_32(ptr, val) __atomic_store_n((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_store_64(ptr, val) __atomic_store_n((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_store_ptr(ptr, val) __atomic_store_n((ptr), (val), __ATOMIC_SEQ_CST)
+
+  #define atomic_exchange_8(ptr, val) __atomic_exchange_n((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_exchange_16(ptr, val) __atomic_exchange_n((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_exchange_32(ptr, val) __atomic_exchange_n((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_exchange_64(ptr, val) __atomic_exchange_n((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_exchange_ptr(ptr, val) __atomic_exchange_n((ptr), (val), __ATOMIC_SEQ_CST)
+
+  #define atomic_val_compare_exchange_8 __sync_val_compare_and_swap
+  #define atomic_val_compare_exchange_16 __sync_val_compare_and_swap
+  #define atomic_val_compare_exchange_32 __sync_val_compare_and_swap
+  #define atomic_val_compare_exchange_64 __sync_val_compare_and_swap
+  #define atomic_val_compare_exchange_ptr __sync_val_compare_and_swap
+
+  #define atomic_add_fetch_8(ptr, val) __atomic_add_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_add_fetch_16(ptr, val) __atomic_add_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_add_fetch_32(ptr, val) __atomic_add_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_add_fetch_64(ptr, val) __atomic_add_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_add_fetch_ptr(ptr, val) __atomic_add_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+
+  #define atomic_fetch_add_8(ptr, val) __atomic_fetch_add((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_add_16(ptr, val) __atomic_fetch_add((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_add_32(ptr, val) __atomic_fetch_add((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_add_64(ptr, val) __atomic_fetch_add((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_add_ptr(ptr, val) __atomic_fetch_add((ptr), (val), __ATOMIC_SEQ_CST)
+
+  #define atomic_sub_fetch_8(ptr, val) __atomic_sub_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_sub_fetch_16(ptr, val) __atomic_sub_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_sub_fetch_32(ptr, val) __atomic_sub_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_sub_fetch_64(ptr, val) __atomic_sub_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_sub_fetch_ptr(ptr, val) __atomic_sub_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+
+  #define atomic_fetch_sub_8(ptr, val) __atomic_fetch_sub((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_sub_16(ptr, val) __atomic_fetch_sub((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_sub_32(ptr, val) __atomic_fetch_sub((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_sub_64(ptr, val) __atomic_fetch_sub((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_sub_ptr(ptr, val) __atomic_fetch_sub((ptr), (val), __ATOMIC_SEQ_CST)
+
+  #define atomic_and_fetch_8(ptr, val) __atomic_and_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_and_fetch_16(ptr, val) __atomic_and_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_and_fetch_32(ptr, val) __atomic_and_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_and_fetch_64(ptr, val) __atomic_and_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_and_fetch_ptr(ptr, val) __atomic_and_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+
+  #define atomic_fetch_and_8(ptr, val) __atomic_fetch_and((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_and_16(ptr, val) __atomic_fetch_and((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_and_32(ptr, val) __atomic_fetch_and((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_and_64(ptr, val) __atomic_fetch_and((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_and_ptr(ptr, val) __atomic_fetch_and((ptr), (val), __ATOMIC_SEQ_CST)
+
+  #define atomic_or_fetch_8(ptr, val) __atomic_or_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_or_fetch_16(ptr, val) __atomic_or_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_or_fetch_32(ptr, val) __atomic_or_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_or_fetch_64(ptr, val) __atomic_or_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_or_fetch_ptr(ptr, val) __atomic_or_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+
+  #define atomic_fetch_or_8(ptr, val) __atomic_fetch_or((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_or_16(ptr, val) __atomic_fetch_or((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_or_32(ptr, val) __atomic_fetch_or((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_or_64(ptr, val) __atomic_fetch_or((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_or_ptr(ptr, val) __atomic_fetch_or((ptr), (val), __ATOMIC_SEQ_CST)
+
+  #define atomic_xor_fetch_8(ptr, val) __atomic_xor_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_xor_fetch_16(ptr, val) __atomic_xor_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_xor_fetch_32(ptr, val) __atomic_xor_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_xor_fetch_64(ptr, val) __atomic_xor_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_xor_fetch_ptr(ptr, val) __atomic_xor_fetch((ptr), (val), __ATOMIC_SEQ_CST)
+
+  #define atomic_fetch_xor_8(ptr, val) __atomic_fetch_xor((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_xor_16(ptr, val) __atomic_fetch_xor((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_xor_32(ptr, val) __atomic_fetch_xor((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_xor_64(ptr, val) __atomic_fetch_xor((ptr), (val), __ATOMIC_SEQ_CST)
+  #define atomic_fetch_xor_ptr(ptr, val) __atomic_fetch_xor((ptr), (val), __ATOMIC_SEQ_CST)
+#endif
+
+ssize_t taosTReadImp(int fd, void *buf, size_t count);
+ssize_t taosTWriteImp(int fd, void *buf, size_t count);
+ssize_t taosTSendFileImp(int dfd, int sfd, off_t *offset, size_t size);
+#ifndef TAOS_OS_FUNC_FILE
+  #define taosTRead(fd, buf, count) taosTReadImp(fd, buf, count)
+  #define taosTWrite(fd, buf, count) taosTWriteImp(fd, buf, count)
+  #define taosLSeek(fd, offset, whence) lseek(fd, offset, whence)
+  #define taosTSendFile(dfd, sfd, offset, size) taosTSendFileImp(dfd, sfd, offset, size)
+#endif
+
+#ifdef TAOS_RANDOM_FILE_FAIL
+  void taosSetRandomFileFailFactor(int factor);
+  void taosSetRandomFileFailOutput(const char *path);
+  ssize_t taosReadFileRandomFail(int fd, void *buf, size_t count, const char *file, uint32_t line);
+  ssize_t taosWriteFileRandomFail(int fd, void *buf, size_t count, const char *file, uint32_t line);
+  off_t taosLSeekRandomFail(int fd, off_t offset, int whence, const char *file, uint32_t line);
+  #define taosTRead(fd, buf, count) taosReadFileRandomFail(fd, buf, count, __FILE__, __LINE__)
+  #define taosTWrite(fd, buf, count) taosWriteFileRandomFail(fd, buf, count, __FILE__, __LINE__)
+  #define taosLSeek(fd, offset, whence) taosLSeekRandomFail(fd, offset, whence, __FILE__, __LINE__)
+#endif
+
+#ifndef TAOS_OS_FUNC_NETWORK
+  #define taosSend(sockfd, buf, len, flags) send(sockfd, buf, len, flags)
+  #define taosSendto(sockfd, buf, len, flags, dest_addr, addrlen) sendto(sockfd, buf, len, flags, dest_addr, addrlen)
+  #define taosReadSocket(fd, buf, len) read(fd, buf, len)
+  #define taosWriteSocket(fd, buf, len) write(fd, buf, len)
+  #define taosCloseSocket(x) \
+    {                        \
+      if (FD_VALID(x)) {     \
+        close(x);            \
+        x = FD_INITIALIZER;  \
+      }                      \
+    }
+#endif
+
+#ifdef TAOS_RANDOM_NETWORK_FAIL
+  ssize_t taosSendRandomFail(int sockfd, const void *buf, size_t len, int flags);
+  ssize_t taosSendToRandomFail(int sockfd, const void *buf, size_t len, int flags, const struct sockaddr *dest_addr, socklen_t addrlen);
+  ssize_t taosReadSocketRandomFail(int fd, void *buf, size_t count);
+  ssize_t taosWriteSocketRandomFail(int fd, const void *buf, size_t count);
+  #define taosSend(sockfd, buf, len, flags) taosSendRandomFail(sockfd, buf, len, flags)
+  #define taosSendto(sockfd, buf, len, flags, dest_addr, addrlen) taosSendToRandomFail(sockfd, buf, len, flags, dest_addr, addrlen)
+  #define taosReadSocket(fd, buf, len) taosReadSocketRandomFail(fd, buf, len)
+  #define taosWriteSocket(fd, buf, len) taosWriteSocketRandomFail(fd, buf, len)
+#endif
+
+#ifndef TAOS_OS_FUNC_LZ4
+  #define BUILDIN_CLZL(val) __builtin_clzl(val)
+  #define BUILDIN_CTZL(val) __builtin_ctzl(val)
+  #define BUILDIN_CLZ(val) __builtin_clz(val)
+  #define BUILDIN_CTZ(val) __builtin_ctz(val)
+#endif
+
+#undef threadlocal
+#ifdef _ISOC11_SOURCE
+  #define threadlocal _Thread_local
+#elif defined(__APPLE__)
+  #define threadlocal
+#elif defined(__GNUC__) && !defined(threadlocal)
+  #define threadlocal __thread
+#else
+  #define threadlocal
+#endif
+
+void osInit();
+
+// TAOS_OS_FUNC_PTHREAD
+bool taosCheckPthreadValid(pthread_t thread);
+int64_t taosGetPthreadId();
+
+// TAOS_OS_FUNC_SOCKET
+int taosSetNonblocking(int sock, int on);
+int taosSetSockOpt(int socketfd, int level, int optname, void *optval, int optlen);
+void taosBlockSIGPIPE();
+
+// TAOS_OS_FUNC_SYSINFO
+void taosGetSystemInfo();
+void taosPrintOsInfo();
+void taosKillSystem();
+int tSystem(const char * cmd) ;
+
+// TAOS_OS_FUNC_CORE
+void taosSetCoreDump();
+
+// TAOS_OS_FUNC_UTIL
+int64_t tsosStr2int64(char *str);
+
+// TAOS_OS_FUNC_TIMER
+void taosMsleep(int mseconds);
+int taosInitTimer(void (*callback)(int), int ms);
+void taosUninitTimer();
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

src/os/inc/osWindows32.h

@@ -0,0 +1,54 @@
+/*
+* Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+*
+* This program is free software: you can use, redistribute, and/or modify
+* it under the terms of the GNU Affero General Public License, version 3
+* or later ("AGPL"), as published by the Free Software Foundation.
+*
+* This program is distributed in the hope that it will be useful, but WITHOUT
+* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+* FITNESS FOR A PARTICULAR PURPOSE.
+*
+* You should have received a copy of the GNU Affero General Public License
+* along with this program. If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef TDENGINE_PLATFORM_WINDOWS32_H
+#define TDENGINE_PLATFORM_WINDOWS32_H
+
+#include <assert.h>
+#include <ctype.h>
+#include <direct.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <float.h>
+#include <locale.h>
+#include <intrin.h>
+#include <io.h>
+#include <math.h>
+#include <pthread.h>
+#include <semaphore.h>
+#include <stdarg.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <signal.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <time.h>
+#include <inttypes.h>
+#include "winsock2.h"
+#include <WS2tcpip.h>
+#include <winbase.h>
+#include <Winsock2.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef __cplusplus
+}
+#endif
+#endif

src/os/inc/osWindows64.h

@@ -78,10 +78,10 @@ extern "C" {
 #define wcsncasecmp _wcsnicmp
 #define strtok_r strtok_s
 #ifdef _TD_GO_DLL_
-  int64_t str2int64(char *str);
+  int64_t tsosStr2int64(char *str);
   uint64_t htonll(uint64_t val);
 #else
-  #define str2int64 _atoi64
+  #define tsosStr2int64 _atoi64
 #endif
 
 #define snprintf _snprintf

src/os/linux/inc/os.h

@@ -1,294 +0,0 @@
-/*
- * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
- *
- * This program is free software: you can use, redistribute, and/or modify
- * it under the terms of the GNU Affero General Public License, version 3
- * or later ("AGPL"), as published by the Free Software Foundation.
- *
- * This program is distributed in the hope that it will be useful, but WITHOUT
- * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
- * FITNESS FOR A PARTICULAR PURPOSE.
- *
- * You should have received a copy of the GNU Affero General Public License
- * along with this program. If not, see <http://www.gnu.org/licenses/>.
- */
-
-#ifndef TDENGINE_PLATFORM_LINUX_H
-#define TDENGINE_PLATFORM_LINUX_H
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-#include <stdio.h>
-#include <stdlib.h>
-
-#include <argp.h>
-#include <arpa/inet.h>
-#include <assert.h>
-#include <ctype.h>
-#include <dirent.h>
-#include <endian.h>
-#include <errno.h>
-#include <float.h>
-#include <ifaddrs.h>
-#include <libgen.h>
-#include <limits.h>
-#include <locale.h>
-#include <math.h>
-#include <netdb.h>
-#include <netinet/in.h>
-#include <netinet/ip.h>
-#include <netinet/tcp.h>
-#include <netinet/udp.h>
-#include <pthread.h>
-#include <pwd.h>
-#include <regex.h>
-#include <semaphore.h>
-#include <signal.h>
-#include <stdarg.h>
-#include <stdbool.h>
-#include <stddef.h>
-#include <stdint.h>
-#include <string.h>
-#include <strings.h>
-#include <sys/epoll.h>
-#include <sys/eventfd.h>
-#include <sys/file.h>
-#include <sys/ioctl.h>
-#include <sys/mman.h>
-#include <sys/sendfile.h>
-#include <sys/socket.h>
-#include <sys/stat.h>
-#include <sys/syscall.h>
-#include <sys/statvfs.h>
-#include <sys/time.h>
-#include <sys/types.h>
-#include <sys/uio.h>
-#include <sys/un.h>
-#include <syslog.h>
-#include <termios.h>
-#include <unistd.h>
-#include <wchar.h>
-#include <wordexp.h>
-#include <wctype.h>
-#include <inttypes.h>
-#include <fcntl.h>
-#include <sys/utsname.h>
-#include <sys/resource.h>
-#include <error.h>
-
-#define taosCloseSocket(x) \
-  {                        \
-    if (FD_VALID(x)) {     \
-      close(x);            \
-      x = FD_INITIALIZER;  \
-    }                      \
-  }
-  
-#ifdef TAOS_RANDOM_NETWORK_FAIL
-
-ssize_t taos_send_random_fail(int sockfd, const void *buf, size_t len, int flags);
-
-ssize_t taos_sendto_random_fail(int sockfd, const void *buf, size_t len, int flags,
-                      const struct sockaddr *dest_addr, socklen_t addrlen);
-ssize_t taos_read_random_fail(int fd, void *buf, size_t count);
-ssize_t taos_write_random_fail(int fd, const void *buf, size_t count);
-
-#define send(sockfd, buf, len, flags) taos_send_random_fail(sockfd, buf, len, flags)
-#define sendto(sockfd, buf, len, flags, dest_addr, addrlen) \
-          taos_sendto_random_fail(sockfd, buf, len, flags, dest_addr, addrlen)
-#define taosWriteSocket(fd, buf, len) taos_write_random_fail(fd, buf, len)
-#define taosReadSocket(fd, buf, len) taos_read_random_fail(fd, buf, len)
-
-#else
-
-#define taosWriteSocket(fd, buf, len) write(fd, buf, len)
-#define taosReadSocket(fd, buf, len) read(fd, buf, len)
-
-#endif  /* TAOS_RANDOM_NETWORK_FAIL */
-
-#define atomic_load_8(ptr) __atomic_load_n((ptr), __ATOMIC_SEQ_CST)
-#define atomic_load_16(ptr) __atomic_load_n((ptr), __ATOMIC_SEQ_CST)
-#define atomic_load_32(ptr) __atomic_load_n((ptr), __ATOMIC_SEQ_CST)
-#define atomic_load_64(ptr) __atomic_load_n((ptr), __ATOMIC_SEQ_CST)
-#define atomic_load_ptr(ptr) __atomic_load_n((ptr), __ATOMIC_SEQ_CST)
-
-#define atomic_store_8(ptr, val) __atomic_store_n((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_store_16(ptr, val) __atomic_store_n((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_store_32(ptr, val) __atomic_store_n((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_store_64(ptr, val) __atomic_store_n((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_store_ptr(ptr, val) __atomic_store_n((ptr), (val), __ATOMIC_SEQ_CST)
-
-#define atomic_exchange_8(ptr, val) __atomic_exchange_n((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_exchange_16(ptr, val) __atomic_exchange_n((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_exchange_32(ptr, val) __atomic_exchange_n((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_exchange_64(ptr, val) __atomic_exchange_n((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_exchange_ptr(ptr, val) __atomic_exchange_n((ptr), (val), __ATOMIC_SEQ_CST)
-
-#define atomic_val_compare_exchange_8 __sync_val_compare_and_swap
-#define atomic_val_compare_exchange_16 __sync_val_compare_and_swap
-#define atomic_val_compare_exchange_32 __sync_val_compare_and_swap
-#define atomic_val_compare_exchange_64 __sync_val_compare_and_swap
-#define atomic_val_compare_exchange_ptr __sync_val_compare_and_swap
-
-#define atomic_add_fetch_8(ptr, val) __atomic_add_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_add_fetch_16(ptr, val) __atomic_add_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_add_fetch_32(ptr, val) __atomic_add_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_add_fetch_64(ptr, val) __atomic_add_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_add_fetch_ptr(ptr, val) __atomic_add_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-
-#define atomic_fetch_add_8(ptr, val) __atomic_fetch_add((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_add_16(ptr, val) __atomic_fetch_add((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_add_32(ptr, val) __atomic_fetch_add((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_add_64(ptr, val) __atomic_fetch_add((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_add_ptr(ptr, val) __atomic_fetch_add((ptr), (val), __ATOMIC_SEQ_CST)
-
-#define atomic_sub_fetch_8(ptr, val) __atomic_sub_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_sub_fetch_16(ptr, val) __atomic_sub_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_sub_fetch_32(ptr, val) __atomic_sub_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_sub_fetch_64(ptr, val) __atomic_sub_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_sub_fetch_ptr(ptr, val) __atomic_sub_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-
-#define atomic_fetch_sub_8(ptr, val) __atomic_fetch_sub((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_sub_16(ptr, val) __atomic_fetch_sub((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_sub_32(ptr, val) __atomic_fetch_sub((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_sub_64(ptr, val) __atomic_fetch_sub((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_sub_ptr(ptr, val) __atomic_fetch_sub((ptr), (val), __ATOMIC_SEQ_CST)
-
-#define atomic_and_fetch_8(ptr, val) __atomic_and_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_and_fetch_16(ptr, val) __atomic_and_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_and_fetch_32(ptr, val) __atomic_and_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_and_fetch_64(ptr, val) __atomic_and_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_and_fetch_ptr(ptr, val) __atomic_and_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-
-#define atomic_fetch_and_8(ptr, val) __atomic_fetch_and((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_and_16(ptr, val) __atomic_fetch_and((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_and_32(ptr, val) __atomic_fetch_and((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_and_64(ptr, val) __atomic_fetch_and((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_and_ptr(ptr, val) __atomic_fetch_and((ptr), (val), __ATOMIC_SEQ_CST)
-
-#define atomic_or_fetch_8(ptr, val) __atomic_or_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_or_fetch_16(ptr, val) __atomic_or_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_or_fetch_32(ptr, val) __atomic_or_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_or_fetch_64(ptr, val) __atomic_or_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_or_fetch_ptr(ptr, val) __atomic_or_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-
-#define atomic_fetch_or_8(ptr, val) __atomic_fetch_or((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_or_16(ptr, val) __atomic_fetch_or((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_or_32(ptr, val) __atomic_fetch_or((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_or_64(ptr, val) __atomic_fetch_or((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_or_ptr(ptr, val) __atomic_fetch_or((ptr), (val), __ATOMIC_SEQ_CST)
-
-#define atomic_xor_fetch_8(ptr, val) __atomic_xor_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_xor_fetch_16(ptr, val) __atomic_xor_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_xor_fetch_32(ptr, val) __atomic_xor_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_xor_fetch_64(ptr, val) __atomic_xor_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_xor_fetch_ptr(ptr, val) __atomic_xor_fetch((ptr), (val), __ATOMIC_SEQ_CST)
-
-#define atomic_fetch_xor_8(ptr, val) __atomic_fetch_xor((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_xor_16(ptr, val) __atomic_fetch_xor((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_xor_32(ptr, val) __atomic_fetch_xor((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_xor_64(ptr, val) __atomic_fetch_xor((ptr), (val), __ATOMIC_SEQ_CST)
-#define atomic_fetch_xor_ptr(ptr, val) __atomic_fetch_xor((ptr), (val), __ATOMIC_SEQ_CST)
-
-#define SWAP(a, b, c)      \
-  do {                     \
-    typeof(a) __tmp = (a); \
-    (a) = (b);             \
-    (b) = __tmp;           \
-  } while (0)
-
-#define MAX(a, b)            \
-  ({                         \
-    typeof(a) __a = (a);     \
-    typeof(b) __b = (b);     \
-    (__a > __b) ? __a : __b; \
-  })
-
-#define MIN(a, b)            \
-  ({                         \
-    typeof(a) __a = (a);     \
-    typeof(b) __b = (b);     \
-    (__a < __b) ? __a : __b; \
-  })
-
-#define MILLISECOND_PER_SECOND ((int64_t)1000L)
-
-#define tsem_t sem_t
-#define tsem_init sem_init
-#define tsem_wait sem_wait
-#define tsem_post sem_post
-#define tsem_destroy sem_destroy
-
-void osInit();
-
-ssize_t tsendfile(int dfd, int sfd, off_t *offset, size_t size);
-
-ssize_t twrite(int fd, void *buf, size_t n);
-
-ssize_t tread(int fd, void *buf, size_t count);
-
-bool taosCheckPthreadValid(pthread_t thread);
-
-void taosResetPthread(pthread_t *thread);
-
-int64_t taosGetPthreadId();
-
-int taosSetNonblocking(int sock, int on);
-
-int taosSetSockOpt(int socketfd, int level, int optname, void *optval, int optlen);
-
-void taosPrintOsInfo();
-
-char *taosCharsetReplace(char *charsetstr);
-
-void taosGetSystemInfo();
-
-void taosKillSystem();
-
-bool taosSkipSocketCheck();
-
-int64_t str2int64(char *str);
-
-void taosSetCoreDump();
-
-void taosBlockSIGPIPE();
-
-int tSystem(const char * cmd) ;
-
-
-#ifdef _ALPINE
-  typedef int(*__compar_fn_t)(const void *, const void *);
-  void  error (int, int, const char *);
-  #ifndef PTHREAD_MUTEX_RECURSIVE_NP
-    #define  PTHREAD_MUTEX_RECURSIVE_NP PTHREAD_MUTEX_RECURSIVE
-  #endif
-#endif
-
-#ifndef _TD_ARM_32_
-#define BUILDIN_CLZL(val) __builtin_clzl(val)
-#define BUILDIN_CTZL(val) __builtin_ctzl(val)
-#else
-#define BUILDIN_CLZL(val) __builtin_clzll(val)
-#define BUILDIN_CTZL(val) __builtin_ctzll(val)
-#endif
-#define BUILDIN_CLZ(val) __builtin_clz(val)
-#define BUILDIN_CTZ(val) __builtin_ctz(val)
-
-#undef threadlocal
-#ifdef _ISOC11_SOURCE
-  #define threadlocal _Thread_local
-#elif defined(__APPLE__)
-  #define threadlocal
-#elif defined(__GNUC__) && !defined(threadlocal)
-  #define threadlocal __thread
-#else
-  #define threadlocal
-#endif
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif

src/os/linux/src/linuxPlatform.c

@@ -1,318 +0,0 @@
-/*
- * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
- *
- * This program is free software: you can use, redistribute, and/or modify
- * it under the terms of the GNU Affero General Public License, version 3
- * or later ("AGPL"), as published by the Free Software Foundation.
- *
- * This program is distributed in the hope that it will be useful, but WITHOUT
- * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
- * FITNESS FOR A PARTICULAR PURPOSE.
- *
- * You should have received a copy of the GNU Affero General Public License
- * along with this program. If not, see <http://www.gnu.org/licenses/>.
- */
-
-#define _DEFAULT_SOURCE
-#include "os.h"
-#include "taosdef.h"
-#include "tglobal.h"
-#include "ttimer.h"
-#include "tulog.h"
-#include "tutil.h"
-
-int64_t str2int64(char *str) {
-  char *endptr = NULL;
-  return strtoll(str, &endptr, 10);
-}
-
-/*
-  to make taosMsleep work,
-   signal SIGALRM shall be blocked in the calling thread,
-
-  sigset_t set;
-  sigemptyset(&set);
-  sigaddset(&set, SIGALRM);
-  pthread_sigmask(SIG_BLOCK, &set, NULL);
-*/
-void taosMsleep(int mseconds) {
-  struct timeval timeout;
-  int            seconds, useconds;
-
-  seconds = mseconds / 1000;
-  useconds = (mseconds % 1000) * 1000;
-  timeout.tv_sec = seconds;
-  timeout.tv_usec = useconds;
-
-  /* sigset_t set; */
-  /* sigemptyset(&set); */
-  /* sigaddset(&set, SIGALRM); */
-  /* pthread_sigmask(SIG_BLOCK, &set, NULL); */
-
-  select(0, NULL, NULL, NULL, &timeout);
-
-  /* pthread_sigmask(SIG_UNBLOCK, &set, NULL); */
-}
-
-bool taosCheckPthreadValid(pthread_t thread) { return thread != 0; }
-
-int64_t taosGetPthreadId() { return (int64_t)pthread_self(); }
-
-int taosSetNonblocking(int sock, int on) {
-  int flags = 0;
-  if ((flags = fcntl(sock, F_GETFL, 0)) < 0) {
-    uError("fcntl(F_GETFL) error: %d (%s)\n", errno, strerror(errno));
-    return 1;
-  }
-
-  if (on)
-    flags |= O_NONBLOCK;
-  else
-    flags &= ~O_NONBLOCK;
-
-  if ((flags = fcntl(sock, F_SETFL, flags)) < 0) {
-    uError("fcntl(F_SETFL) error: %d (%s)\n", errno, strerror(errno));
-    return 1;
-  }
-
-  return 0;
-}
-
-int taosSetSockOpt(int socketfd, int level, int optname, void *optval, int optlen) {
-  return setsockopt(socketfd, level, optname, optval, (socklen_t)optlen);
-}
-static void taosDeleteTimer(void *tharg) {
-  timer_t *pTimer = tharg;
-  timer_delete(*pTimer);
-}
-
-static pthread_t timerThread;
-static timer_t         timerId;
-static volatile bool stopTimer = false;
-
-void *taosProcessAlarmSignal(void *tharg) {
-  // Block the signal
-  sigset_t sigset;
-  sigemptyset(&sigset);
-  sigaddset(&sigset, SIGALRM);
-  sigprocmask(SIG_BLOCK, &sigset, NULL);
-  void (*callback)(int) = tharg;
-
-  struct sigevent sevent = {{0}};
-
-  #ifdef _ALPINE
-    sevent.sigev_notify = SIGEV_THREAD;
-    sevent.sigev_value.sival_int = syscall(__NR_gettid);
-  #else
-    sevent.sigev_notify = SIGEV_THREAD_ID;
-    sevent._sigev_un._tid = syscall(__NR_gettid);
-  #endif
-  
-  sevent.sigev_signo = SIGALRM;
-
-  if (timer_create(CLOCK_REALTIME, &sevent, &timerId) == -1) {
-    uError("Failed to create timer");
-  }
-
-  pthread_cleanup_push(taosDeleteTimer, &timerId);
-
-  struct itimerspec ts;
-  ts.it_value.tv_sec = 0;
-  ts.it_value.tv_nsec = 1000000 * MSECONDS_PER_TICK;
-  ts.it_interval.tv_sec = 0;
-  ts.it_interval.tv_nsec = 1000000 * MSECONDS_PER_TICK;
-
-  if (timer_settime(timerId, 0, &ts, NULL)) {
-    uError("Failed to init timer");
-    return NULL;
-  }
-
-  int signo;
-  while (!stopTimer) {
-    if (sigwait(&sigset, &signo)) {
-      uError("Failed to wait signal: number %d", signo);
-      continue;
-    }
-    /* printf("Signal handling: number %d ......\n", signo); */
-
-    callback(0);
-  }
-  
-  pthread_cleanup_pop(1);
-
-  return NULL;
-}
-
-
-int taosInitTimer(void (*callback)(int), int ms) {
-  pthread_attr_t tattr;
-  pthread_attr_init(&tattr);
-  int code = pthread_create(&timerThread, &tattr, taosProcessAlarmSignal, callback);
-  pthread_attr_destroy(&tattr);
-  if (code != 0) {
-    uError("failed to create timer thread");
-    return -1;
-  }
-  return 0;
-}
-
-void taosUninitTimer() {
-  stopTimer = true;
-  pthread_join(timerThread, NULL);
-}
-
-ssize_t tread(int fd, void *buf, size_t count) {
-  size_t  leftbytes = count;
-  ssize_t readbytes;
-  char *  tbuf = (char *)buf;
-
-  while (leftbytes > 0) {
-    readbytes = read(fd, (void *)tbuf, leftbytes);
-    if (readbytes < 0) {
-      if (errno == EINTR) {
-        continue;
-      } else {
-        return -1;
-      }
-    } else if (readbytes == 0) {
-      return (ssize_t)(count - leftbytes);
-    }
-
-    leftbytes -= readbytes;
-    tbuf += readbytes;
-  }
-
-  return (ssize_t)count;
-}
-
-ssize_t tsendfile(int dfd, int sfd, off_t *offset, size_t size) {
-  size_t  leftbytes = size;
-  ssize_t sentbytes;
-
-  while (leftbytes > 0) {
-    /*
-     * TODO : Think to check if file is larger than 1GB
-     */
-    //if (leftbytes > 1000000000) leftbytes = 1000000000;
-    sentbytes = sendfile(dfd, sfd, offset, leftbytes);
-    if (sentbytes == -1) {
-      if (errno == EINTR) {
-        continue;
-      }
-      else {
-        return -1;
-      }
-    } else if (sentbytes == 0) {
-      return (ssize_t)(size - leftbytes);
-    }
-
-    leftbytes -= sentbytes;
-  }
-
-  return size;
-}
-
-ssize_t twrite(int fd, void *buf, size_t n) {
-  size_t nleft = n; 
-  ssize_t nwritten = 0;
-  char *tbuf = (char *)buf;
-
-  while (nleft > 0) {
-    nwritten = write(fd, (void *)tbuf, nleft);
-    if (nwritten < 0) {
-      if (errno == EINTR) {
-        continue;
-      }
-      return -1;
-    }
-    nleft -= nwritten;
-    tbuf += nwritten;
-  }
-
-  return n;
-}
-
-void taosBlockSIGPIPE() {
-  sigset_t signal_mask;
-  sigemptyset(&signal_mask);
-  sigaddset(&signal_mask, SIGPIPE);
-  int rc = pthread_sigmask(SIG_BLOCK, &signal_mask, NULL);
-  if (rc != 0) {
-    uError("failed to block SIGPIPE");
-  }
-}
-
-int tSystem(const char * cmd) 
-{ 
-  FILE * fp; 
-  int res; 
-  char buf[1024]; 
-  if (cmd == NULL) { 
-    uError("tSystem cmd is NULL!\n");
-    return -1;
-  } 
-  
-  if ((fp = popen(cmd, "r") ) == NULL) { 
-    uError("popen cmd:%s error: %s/n", cmd, strerror(errno)); 
-    return -1; 
-  } else {
-    while(fgets(buf, sizeof(buf), fp))  { 
-      uDebug("popen result:%s", buf); 
-    } 
-
-    if ((res = pclose(fp)) == -1) { 
-      uError("close popen file pointer fp error!\n");
-    } else { 
-      uDebug("popen res is :%d\n", res);
-    } 
-
-    return res;
-  }
-}
-
-#ifdef TAOS_RANDOM_NETWORK_FAIL
-
-#define RANDOM_NETWORK_FAIL_FACTOR  20
-
-ssize_t taos_send_random_fail(int sockfd, const void *buf, size_t len, int flags)
-{
-  if (rand() % RANDOM_NETWORK_FAIL_FACTOR == 0) {
-    errno = ECONNRESET;
-    return -1;
-  }
-
-  return send(sockfd, buf, len, flags);
-}
-
-ssize_t taos_sendto_random_fail(int sockfd, const void *buf, size_t len, int flags,
-                      const struct sockaddr *dest_addr, socklen_t addrlen)
-{
-  if (rand() % RANDOM_NETWORK_FAIL_FACTOR == 0) {
-    errno = ECONNRESET;
-    return -1;
-  }
-
-  return sendto(sockfd, buf, len, flags, dest_addr, addrlen);
-}
-
-ssize_t taos_read_random_fail(int fd, void *buf, size_t count)
-{
-  if (rand() % RANDOM_NETWORK_FAIL_FACTOR == 0) {
-    errno = ECONNRESET;
-    return -1;
-  }
-
-  return read(fd, buf, count);
-}
-
-ssize_t taos_write_random_fail(int fd, const void *buf, size_t count)
-{
-  if (rand() % RANDOM_NETWORK_FAIL_FACTOR == 0) {
-    errno = EINTR;
-    return -1;
-  }
-
-  return write(fd, buf, count);
-}
-
-#endif /* TAOS_RANDOM_NETWORK_FAIL */

src/os/src/alpine/CMakeLists.txt

@@ -0,0 +1,12 @@
+CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
+PROJECT(TDengine)
+
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/common/inc)
+AUX_SOURCE_DIRECTORY(. SRC)
+
+ADD_LIBRARY(os ${SRC})
+TARGET_LINK_LIBRARIES(os m rt)
+

src/os/src/darwin64/CMakeLists.txt

@@ -1,16 +1,13 @@
 CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
 PROJECT(TDengine)
 
-IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
+IF (TD_DARWIN_64)
-  INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/common/inc)
   INCLUDE_DIRECTORIES(inc)
   AUX_SOURCE_DIRECTORY(src SRC)
-
   ADD_LIBRARY(os ${SRC})
-  TARGET_LINK_LIBRARIES(os m rt)
 ENDIF ()
 
-SET_SOURCE_FILES_PROPERTIES(src/linuxSysPara.c PROPERTIES COMPILE_FLAGS -w)

src/os/src/darwin64/darwinPlatform.c

@@ -15,12 +15,15 @@
 
 #define _DEFAULT_SOURCE
 #include "os.h"
-#include "tglobal.h"
+#include "os.h"
 #include "taosdef.h"
-#include "tutil.h"
+#include "tglobal.h"
+#include "tconfig.h"
+#include "ttimer.h"
 #include "tulog.h"
+#include "tutil.h"
 
-int64_t str2int64(char *str) {
+int64_t tsosStr2int64(char *str) {
   char *endptr = NULL;
   return strtoll(str, &endptr, 10);
 }
@@ -159,12 +162,14 @@ int taosInitTimer(void (*callback)(int), int ms) {
   tv.it_interval.tv_usec = 1000 * ms;  // resolution is in msecond
   tv.it_value = tv.it_interval;
 
-  return setitimer(ITIMER_REAL, &tv, NULL);
+  setitimer(ITIMER_REAL, &tv, NULL);
+
+  return 0;
 }
 
 void taosUninitTimer() {
   struct itimerval tv = { 0 };
-  return setitimer(ITIMER_REAL, &tv, NULL);
+  setitimer(ITIMER_REAL, &tv, NULL);
 }
 
 void taosGetSystemTimezone() {
@@ -207,7 +212,7 @@ void taosGetSystemLocale() {
 void taosPrintOsInfo() {}
 
 void taosKillSystem() {
-  tError("function taosKillSystem, exit!");
+  uError("function taosKillSystem, exit!");
   exit(0);
 }
 
@@ -221,95 +226,75 @@ void taosGetSystemInfo() {
 }
 
 void *taosInitTcpClient(char *ip, uint16_t port, char *flabel, int num, void *fp, void *shandle) {
-  tError("function taosInitTcpClient is not implemented in darwin system, exit!");
+  uError("function taosInitTcpClient is not implemented in darwin system, exit!");
   exit(0);
 }
 
 void taosCloseTcpClientConnection(void *chandle) {
-  tError("function taosCloseTcpClientConnection is not implemented in darwin system, exit!");
+  uError("function taosCloseTcpClientConnection is not implemented in darwin system, exit!");
   exit(0);
 }
 
 void *taosOpenTcpClientConnection(void *shandle, void *thandle, char *ip, uint16_t port) {
-  tError("function taosOpenTcpClientConnection is not implemented in darwin system, exit!");
+  uError("function taosOpenTcpClientConnection is not implemented in darwin system, exit!");
   exit(0);
 }
 
 int taosSendTcpClientData(unsigned int ip, uint16_t port, char *data, int len, void *chandle) {
-  tError("function taosSendTcpClientData is not implemented in darwin system, exit!");
+  uError("function taosSendTcpClientData is not implemented in darwin system, exit!");
   exit(0);
 }
 
 void taosCleanUpTcpClient(void *chandle) {
-  tError("function taosCleanUpTcpClient is not implemented in darwin system, exit!");
+  uError("function taosCleanUpTcpClient is not implemented in darwin system, exit!");
   exit(0);
 }
 
 void taosCloseTcpServerConnection(void *chandle) {
-  tError("function taosCloseTcpServerConnection is not implemented in darwin system, exit!");
+  uError("function taosCloseTcpServerConnection is not implemented in darwin system, exit!");
   exit(0);
 }
 
 void taosCleanUpTcpServer(void *handle) {
-  tError("function taosCleanUpTcpServer is not implemented in darwin system, exit!");
+  uError("function taosCleanUpTcpServer is not implemented in darwin system, exit!");
   exit(0);
 }
 
 void *taosInitTcpServer(char *ip, uint16_t port, char *label, int numOfThreads, void *fp, void *shandle) {
-  tError("function taosInitTcpServer is not implemented in darwin system, exit!");
+  uError("function taosInitTcpServer is not implemented in darwin system, exit!");
   exit(0);
 }
 
 int taosSendTcpServerData(unsigned int ip, uint16_t port, char *data, int len, void *chandle) {
-  tError("function taosSendTcpServerData is not implemented in darwin system, exit!");
+  uError("function taosSendTcpServerData is not implemented in darwin system, exit!");
   exit(0);
 }
 
 void taosFreeMsgHdr(void *hdr) {
-  tError("function taosFreeMsgHdr is not implemented in darwin system, exit!");
+  uError("function taosFreeMsgHdr is not implemented in darwin system, exit!");
   exit(0);
 }
 
 int taosMsgHdrSize(void *hdr) {
-  tError("function taosMsgHdrSize is not implemented in darwin system, exit!");
+  uError("function taosMsgHdrSize is not implemented in darwin system, exit!");
   exit(0);
 }
 
 void taosSendMsgHdr(void *hdr, int fd) {
-  tError("function taosSendMsgHdr is not implemented in darwin system, exit!");
+  uError("function taosSendMsgHdr is not implemented in darwin system, exit!");
   exit(0);
 }
 
 void taosInitMsgHdr(void **hdr, void *dest, int maxPkts) {
-  tError("function taosInitMsgHdr is not implemented in darwin system, exit!");
+  uError("function taosInitMsgHdr is not implemented in darwin system, exit!");
   exit(0);
 }
 
 void taosSetMsgHdrData(void *hdr, char *data, int dataLen) {
-  tError("function taosSetMsgHdrData is not implemented in darwin system, exit!");
+  uError("function taosSetMsgHdrData is not implemented in darwin system, exit!");
   exit(0);
 }
 
-ssize_t twrite(int fd, void *buf, size_t n) {
-  size_t nleft = n; 
-  ssize_t nwritten = 0;
-  char *tbuf = (char *)buf;
-
-  while (nleft > 0) {
-    nwritten = write(fd, (void *)tbuf, nleft);
-    if (nwritten < 0) {
-      if (errno == EINTR) {
-        continue;
-      }
-      return -1;
-    }
-    nleft -= nwritten;
-    tbuf += nwritten;
-  }
-
-  return n;
-}
-
 bool taosSkipSocketCheck() {
   return true;
 }

src/os/src/detail/CMakeLists.txt

@@ -0,0 +1,14 @@
+CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
+PROJECT(TDengine)
+
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/common/inc)
+INCLUDE_DIRECTORIES(.)
+AUX_SOURCE_DIRECTORY(. SRC)
+SET_SOURCE_FILES_PROPERTIES(osSysinfo.c PROPERTIES COMPILE_FLAGS -w)
+SET_SOURCE_FILES_PROPERTIES(osCoredump.c PROPERTIES COMPILE_FLAGS -w)
+
+ADD_LIBRARY(osdetail ${SRC})
+TARGET_LINK_LIBRARIES(osdetail os)

src/os/src/detail/osCoredump.c

@@ -0,0 +1,98 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#define _DEFAULT_SOURCE
+#include "os.h"
+#include "tconfig.h"
+#include "tglobal.h"
+#include "tulog.h"
+#include "tutil.h"
+#include "tsystem.h"
+
+#ifndef TAOS_OS_FUNC_CORE
+
+int _sysctl(struct __sysctl_args *args );
+
+void taosSetCoreDump() {
+  if (0 == tsEnableCoreFile) {
+    return;
+  }
+  
+  // 1. set ulimit -c unlimited
+  struct rlimit rlim;
+  struct rlimit rlim_new;
+  if (getrlimit(RLIMIT_CORE, &rlim) == 0) {
+    uInfo("the old unlimited para: rlim_cur=%" PRIu64 ", rlim_max=%" PRIu64, rlim.rlim_cur, rlim.rlim_max);
+    rlim_new.rlim_cur = RLIM_INFINITY;
+    rlim_new.rlim_max = RLIM_INFINITY;
+    if (setrlimit(RLIMIT_CORE, &rlim_new) != 0) {
+      uInfo("set unlimited fail, error: %s", strerror(errno));
+      rlim_new.rlim_cur = rlim.rlim_max;
+      rlim_new.rlim_max = rlim.rlim_max;
+      (void)setrlimit(RLIMIT_CORE, &rlim_new);
+    }
+  }
+
+  if (getrlimit(RLIMIT_CORE, &rlim) == 0) {
+    uInfo("the new unlimited para: rlim_cur=%" PRIu64 ", rlim_max=%" PRIu64, rlim.rlim_cur, rlim.rlim_max);
+  }
+
+#ifndef _TD_ARM_
+  // 2. set the path for saving core file
+  struct __sysctl_args args;
+  int     old_usespid = 0;
+  size_t  old_len     = 0;
+  int     new_usespid = 1;
+  size_t  new_len     = sizeof(new_usespid);
+  
+  int name[] = {CTL_KERN, KERN_CORE_USES_PID};
+  
+  memset(&args, 0, sizeof(struct __sysctl_args));
+  args.name    = name;
+  args.nlen    = sizeof(name)/sizeof(name[0]);
+  args.oldval  = &old_usespid;
+  args.oldlenp = &old_len;
+  args.newval  = &new_usespid;
+  args.newlen  = new_len;
+  
+  old_len = sizeof(old_usespid);
+  
+  if (syscall(SYS__sysctl, &args) == -1) {
+      uInfo("_sysctl(kern_core_uses_pid) set fail: %s", strerror(errno));
+  }
+  
+  uInfo("The old core_uses_pid[%" PRIu64 "]: %d", old_len, old_usespid);
+
+
+  old_usespid = 0;
+  old_len     = 0;
+  memset(&args, 0, sizeof(struct __sysctl_args));
+  args.name    = name;
+  args.nlen    = sizeof(name)/sizeof(name[0]);
+  args.oldval  = &old_usespid;
+  args.oldlenp = &old_len;
+  
+  old_len = sizeof(old_usespid);
+  
+  if (syscall(SYS__sysctl, &args) == -1) {
+      uInfo("_sysctl(kern_core_uses_pid) get fail: %s", strerror(errno));
+  }
+  
+  uInfo("The new core_uses_pid[%" PRIu64 "]: %d", old_len, old_usespid);
+#endif
+
+}
+
+#endif

src/os/src/detail/osFail.c

@@ -13,31 +13,62 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
 
-#include <stdio.h>
+#define _DEFAULT_SOURCE
-#include <stdlib.h>
-#include <error.h>
-#include <errno.h>
-#include <stdarg.h>
-#include <unistd.h>
-
-#include <sys/types.h>
-#include <sys/stat.h>
-#include <fcntl.h>
-
 #include "os.h"
 
+#define RANDOM_NETWORK_FAIL_FACTOR  20
+
+#ifdef TAOS_RANDOM_NETWORK_FAIL
+
+ssize_t taosSendRandomFail(int sockfd, const void *buf, size_t len, int flags) {
+  if (rand() % RANDOM_NETWORK_FAIL_FACTOR == 0) {
+    errno = ECONNRESET;
+    return -1;
+  }
+
+  return send(sockfd, buf, len, flags);
+}
+
+ssize_t taosSendToRandomFail(int sockfd, const void *buf, size_t len, int flags, const struct sockaddr *dest_addr,
+                           socklen_t addrlen) {
+  if (rand() % RANDOM_NETWORK_FAIL_FACTOR == 0) {
+    errno = ECONNRESET;
+    return -1;
+  }
+
+  return sendto(sockfd, buf, len, flags, dest_addr, addrlen);
+}
+
+ssize_t taosReadSocketRandomFail(int fd, void *buf, size_t count) {
+  if (rand() % RANDOM_NETWORK_FAIL_FACTOR == 0) {
+    errno = ECONNRESET;
+    return -1;
+  }
+
+  return read(fd, buf, count);
+}
+
+ssize_t taosWriteSocketRandomFail(int fd, const void *buf, size_t count) {
+  if (rand() % RANDOM_NETWORK_FAIL_FACTOR == 0) {
+    errno = EINTR;
+    return -1;
+  }
+
+  return write(fd, buf, count);
+}
+
+#endif //TAOS_RANDOM_NETWORK_FAIL
+
 #ifdef TAOS_RANDOM_FILE_FAIL
 
 static int random_file_fail_factor = 20;
 static FILE *fpRandomFileFailOutput = NULL;
 
-void taosSetRandomFileFailFactor(int factor)
+void taosSetRandomFileFailFactor(int factor) { 
-{
   random_file_fail_factor = factor;
 }
 
-static void close_random_file_fail_output()
+static void close_random_file_fail_output() {
-{
   if (fpRandomFileFailOutput != NULL) {
     if (fpRandomFileFailOutput != stdout) {
       fclose(fpRandomFileFailOutput);
@@ -46,8 +77,7 @@ static void close_random_file_fail_output()
   }
 }
 
-static void random_file_fail_output_sig(int sig)
+static void random_file_fail_output_sig(int sig) {
-{
   fprintf(fpRandomFileFailOutput, "signal %d received.\n", sig);
 
   struct sigaction act = {0};
@@ -58,8 +88,7 @@ static void random_file_fail_output_sig(int sig)
   exit(EXIT_FAILURE);
 }
 
-void taosSetRandomFileFailOutput(const char *path)
+void taosSetRandomFileFailOutput(const char *path) {
-{
   if (path == NULL) {
     fpRandomFileFailOutput = stdout;
   } else if ((fpRandomFileFailOutput = fopen(path, "w")) != NULL) {
@@ -75,43 +104,38 @@ void taosSetRandomFileFailOutput(const char *path)
   sigaction(SIGSEGV, &act, NULL);
   sigaction(SIGILL, &act, NULL);
 }
-#endif
 
-ssize_t taos_tread(int fd, void *buf, size_t count, const char *file, uint32_t line)
+ssize_t taosReadFileRandomFail(int fd, void *buf, size_t count, const char *file, uint32_t line) {
-{
-#ifdef TAOS_RANDOM_FILE_FAIL
   if (random_file_fail_factor > 0) {
     if (rand() % random_file_fail_factor == 0) {
       errno = EIO;
       return -1;
     }
   }
-#endif
+
-  return tread(fd, buf, count);
+  return taosTReadImp(fd, buf, count);
 }
 
-ssize_t taos_twrite(int fd, void *buf, size_t count, const char *file, uint32_t line)
+ssize_t taosWriteFileRandomFail(int fd, void *buf, size_t count, const char *file, uint32_t line) {
-{
-#ifdef TAOS_RANDOM_FILE_FAIL
   if (random_file_fail_factor > 0) {
     if (rand() % random_file_fail_factor == 0) {
       errno = EIO;
       return -1;
     }
   }
-#endif
+
-  return twrite(fd, buf, count);
+  return taosTWriteImp(fd, buf, count);
 }
 
-off_t taos_lseek(int fd, off_t offset, int whence, const char *file, uint32_t line)
+off_t taosLSeekRandomFail(int fd, off_t offset, int whence, const char *file, uint32_t line) {
-{
-#ifdef TAOS_RANDOM_FILE_FAIL
   if (random_file_fail_factor > 0) {
     if (rand() % random_file_fail_factor == 0) {
       errno = EIO;
       return -1;
     }
   }
-#endif
+
   return lseek(fd, offset, whence);
 }
+
+#endif //TAOS_RANDOM_FILE_FAIL

src/os/src/detail/osFile.c

@@ -0,0 +1,88 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#define _DEFAULT_SOURCE
+#include "os.h"
+
+ssize_t taosTReadImp(int fd, void *buf, size_t count) {
+  size_t  leftbytes = count;
+  ssize_t readbytes;
+  char *  tbuf = (char *)buf;
+
+  while (leftbytes > 0) {
+    readbytes = read(fd, (void *)tbuf, leftbytes);
+    if (readbytes < 0) {
+      if (errno == EINTR) {
+        continue;
+      } else {
+        return -1;
+      }
+    } else if (readbytes == 0) {
+      return (ssize_t)(count - leftbytes);
+    }
+
+    leftbytes -= readbytes;
+    tbuf += readbytes;
+  }
+
+  return (ssize_t)count;
+}
+
+ssize_t taosTSendFileImp(int dfd, int sfd, off_t *offset, size_t size) {
+  size_t  leftbytes = size;
+  ssize_t sentbytes;
+
+  while (leftbytes > 0) {
+    /*
+     * TODO : Think to check if file is larger than 1GB
+     */
+    //if (leftbytes > 1000000000) leftbytes = 1000000000;
+    sentbytes = sendfile(dfd, sfd, offset, leftbytes);
+    if (sentbytes == -1) {
+      if (errno == EINTR) {
+        continue;
+      }
+      else {
+        return -1;
+      }
+    } else if (sentbytes == 0) {
+      return (ssize_t)(size - leftbytes);
+    }
+
+    leftbytes -= sentbytes;
+  }
+
+  return size;
+}
+
+ssize_t taosTWriteImp(int fd, void *buf, size_t n) {
+  size_t nleft = n; 
+  ssize_t nwritten = 0;
+  char *tbuf = (char *)buf;
+
+  while (nleft > 0) {
+    nwritten = write(fd, (void *)tbuf, nleft);
+    if (nwritten < 0) {
+      if (errno == EINTR) {
+        continue;
+      }
+      return -1;
+    }
+    nleft -= nwritten;
+    tbuf += nwritten;
+  }
+
+  return n;
+}

src/os/src/detail/osPThread.c

@@ -0,0 +1,24 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#define _DEFAULT_SOURCE
+#include "os.h"
+
+#ifndef TAOS_OS_FUNC_PTHREAD
+
+bool    taosCheckPthreadValid(pthread_t thread) { return thread != 0; }
+int64_t taosGetPthreadId() { return (int64_t)pthread_self(); }
+
+#endif

src/os/src/detail/osSocket.c

@@ -0,0 +1,56 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#define _DEFAULT_SOURCE
+#include "os.h"
+#include "tulog.h"
+
+#ifndef TAOS_OS_FUNC_SOCKET
+
+int taosSetNonblocking(int sock, int on) {
+  int flags = 0;
+  if ((flags = fcntl(sock, F_GETFL, 0)) < 0) {
+    uError("fcntl(F_GETFL) error: %d (%s)\n", errno, strerror(errno));
+    return 1;
+  }
+
+  if (on)
+    flags |= O_NONBLOCK;
+  else
+    flags &= ~O_NONBLOCK;
+
+  if ((flags = fcntl(sock, F_SETFL, flags)) < 0) {
+    uError("fcntl(F_SETFL) error: %d (%s)\n", errno, strerror(errno));
+    return 1;
+  }
+
+  return 0;
+}
+
+int taosSetSockOpt(int socketfd, int level, int optname, void *optval, int optlen) {
+  return setsockopt(socketfd, level, optname, optval, (socklen_t)optlen);
+}
+
+void taosBlockSIGPIPE() {
+  sigset_t signal_mask;
+  sigemptyset(&signal_mask);
+  sigaddset(&signal_mask, SIGPIPE);
+  int rc = pthread_sigmask(SIG_BLOCK, &signal_mask, NULL);
+  if (rc != 0) {
+    uError("failed to block SIGPIPE");
+  }
+}
+
+#endif

src/os/src/detail/osSysinfo.c

@@ -15,13 +15,14 @@
 
 #define _DEFAULT_SOURCE
 #include "os.h"
-#include "tsysctl.h"
 #include "tconfig.h"
 #include "tglobal.h"
 #include "tulog.h"
 #include "tutil.h"
 #include "tsystem.h"
 
+#ifndef TAOS_OS_FUNC_SYSINFO
+
 #define PROCESS_ITEM 12
 
 typedef struct {
@@ -200,7 +201,7 @@ static void taosGetSystemTimezone() {
    * Asia/Shanghai   (CST, +0800)
    * Europe/London   (BST, +0100)
    */
-  sprintf(tsTimezone, "%s (%s, %s%02d00)", buf, tzname[daylight], tz >= 0 ? "+" : "-", abs(tz));
+  snprintf(tsTimezone, TSDB_TIMEZONE_LEN, "%s (%s, %s%02d00)", buf, tzname[daylight], tz >= 0 ? "+" : "-", abs(tz));
 
   // cfg_timezone->cfgStatus = TAOS_CFG_CSTATUS_DEFAULT;
   uInfo("timezone not configured, set to system default:%s", tsTimezone);
@@ -234,7 +235,7 @@ static void taosGetSystemLocale() {  // get and set default locale
       uError("can't get locale from system, set it to en_US.UTF-8");
       strcpy(tsLocale, "en_US.UTF-8");
     } else {
-      tstrncpy(tsLocale, locale, sizeof(tsLocale));
+      tstrncpy(tsLocale, locale, TSDB_LOCALE_LEN);
       uError("locale not configured, set to system default:%s", tsLocale);
     }
   }
@@ -247,7 +248,7 @@ static void taosGetSystemLocale() {  // get and set default locale
       str++;
 
       char *revisedCharset = taosCharsetReplace(str);
-      tstrncpy(tsCharset, revisedCharset, sizeof(tsCharset));
+      tstrncpy(tsCharset, revisedCharset, TSDB_LOCALE_LEN);
 
       free(revisedCharset);
       uWarn("charset not configured, set to system default:%s", tsCharset);
@@ -542,126 +543,31 @@ void taosKillSystem() {
   kill(tsProcId, 2);
 }
 
-int _sysctl(struct __sysctl_args *args );
+int tSystem(const char *cmd) {
-void taosSetCoreDump() {
+  FILE *fp;
-  if (0 == tsEnableCoreFile) {
+  int   res;
-    return;
+  char  buf[1024];
+  if (cmd == NULL) {
+    uError("tSystem cmd is NULL!\n");
+    return -1;
   }
-  
+
-  // 1. set ulimit -c unlimited
+  if ((fp = popen(cmd, "r")) == NULL) {
-  struct rlimit rlim;
+    uError("popen cmd:%s error: %s/n", cmd, strerror(errno));
-  struct rlimit rlim_new;
+    return -1;
-  if (getrlimit(RLIMIT_CORE, &rlim) == 0) {
+  } else {
-    uInfo("the old unlimited para: rlim_cur=%" PRIu64 ", rlim_max=%" PRIu64, rlim.rlim_cur, rlim.rlim_max);
+    while (fgets(buf, sizeof(buf), fp)) {
-    rlim_new.rlim_cur = RLIM_INFINITY;
+      uDebug("popen result:%s", buf);
-    rlim_new.rlim_max = RLIM_INFINITY;
-    if (setrlimit(RLIMIT_CORE, &rlim_new) != 0) {
-      uInfo("set unlimited fail, error: %s", strerror(errno));
-      rlim_new.rlim_cur = rlim.rlim_max;
-      rlim_new.rlim_max = rlim.rlim_max;
-      (void)setrlimit(RLIMIT_CORE, &rlim_new);
     }
-  }
 
-  if (getrlimit(RLIMIT_CORE, &rlim) == 0) {
+    if ((res = pclose(fp)) == -1) {
-    uInfo("the new unlimited para: rlim_cur=%" PRIu64 ", rlim_max=%" PRIu64, rlim.rlim_cur, rlim.rlim_max);
+      uError("close popen file pointer fp error!\n");
-  }
+    } else {
-
+      uDebug("popen res is :%d\n", res);
-#ifndef _TD_ARM_
-  // 2. set the path for saving core file
-  struct __sysctl_args args;
-  int     old_usespid = 0;
-  size_t  old_len     = 0;
-  int     new_usespid = 1;
-  size_t  new_len     = sizeof(new_usespid);
-  
-  int name[] = {CTL_KERN, KERN_CORE_USES_PID};
-  
-  memset(&args, 0, sizeof(struct __sysctl_args));
-  args.name    = name;
-  args.nlen    = sizeof(name)/sizeof(name[0]);
-  args.oldval  = &old_usespid;
-  args.oldlenp = &old_len;
-  args.newval  = &new_usespid;
-  args.newlen  = new_len;
-  
-  old_len = sizeof(old_usespid);
-  
-  if (syscall(SYS__sysctl, &args) == -1) {
-      uInfo("_sysctl(kern_core_uses_pid) set fail: %s", strerror(errno));
-  }
-  
-  uInfo("The old core_uses_pid[%" PRIu64 "]: %d", old_len, old_usespid);
-
-
-  old_usespid = 0;
-  old_len     = 0;
-  memset(&args, 0, sizeof(struct __sysctl_args));
-  args.name    = name;
-  args.nlen    = sizeof(name)/sizeof(name[0]);
-  args.oldval  = &old_usespid;
-  args.oldlenp = &old_len;
-  
-  old_len = sizeof(old_usespid);
-  
-  if (syscall(SYS__sysctl, &args) == -1) {
-      uInfo("_sysctl(kern_core_uses_pid) get fail: %s", strerror(errno));
-  }
-  
-  uInfo("The new core_uses_pid[%" PRIu64 "]: %d", old_len, old_usespid);
-#endif
-  
-#if 0
-  // 3. create the path for saving core file
-  int status; 
-  char coredump_dir[32] = "/var/log/taosdump";
-  if (opendir(coredump_dir) == NULL) {
-    status = mkdir(coredump_dir, S_IRWXU | S_IRWXG | S_IRWXO); 
-    if (status) {
-      uInfo("mkdir fail, error: %s\n", strerror(errno));
     }
+
+    return res;
   }
-
-  // 4. set kernel.core_pattern
-   struct __sysctl_args args;
-   char    old_corefile[128];
-   size_t  old_len;
-   char    new_corefile[128] = "/var/log/taosdump/core-%e-%p";
-   size_t  new_len = sizeof(new_corefile);
-   
-   int name[] = {CTL_KERN, KERN_CORE_PATTERN};
-
-   memset(&args, 0, sizeof(struct __sysctl_args));
-   args.name    = name;
-   args.nlen    = sizeof(name)/sizeof(name[0]);
-   args.oldval  = old_corefile;
-   args.oldlenp = &old_len;
-   args.newval  = new_corefile;
-   args.newlen  = new_len;
-
-   old_len = sizeof(old_corefile);
-
-   if (syscall(SYS__sysctl, &args) == -1) {
-       uInfo("_sysctl(kern_core_pattern) set fail: %s", strerror(errno));
-   }
-   
-   uInfo("The old kern_core_pattern: %*s\n", old_len, old_corefile);
-
-
-   memset(&args, 0, sizeof(struct __sysctl_args));
-   args.name    = name;
-   args.nlen    = sizeof(name)/sizeof(name[0]);
-   args.oldval  = old_corefile;
-   args.oldlenp = &old_len;
-   
-   old_len = sizeof(old_corefile);
-
-   if (syscall(SYS__sysctl, &args) == -1) {
-       uInfo("_sysctl(kern_core_pattern) get fail: %s", strerror(errno));
-   }
-   
-   uInfo("The new kern_core_pattern: %*s\n", old_len, old_corefile);
-#endif
-  
 }
 
+#endif

src/os/src/detail/osTimer.c

@@ -0,0 +1,133 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "os.h"
+#include "taosdef.h"
+#include "tglobal.h"
+#include "ttimer.h"
+#include "tulog.h"
+#include "tutil.h"
+#include <signal.h>
+
+#ifndef TAOS_OS_FUNC_TIMER
+
+/*
+  to make taosMsleep work,
+   signal SIGALRM shall be blocked in the calling thread,
+
+  sigset_t set;
+  sigemptyset(&set);
+  sigaddset(&set, SIGALRM);
+  pthread_sigmask(SIG_BLOCK, &set, NULL);
+*/
+void taosMsleep(int mseconds) {
+  struct timeval timeout;
+  int            seconds, useconds;
+
+  seconds = mseconds / 1000;
+  useconds = (mseconds % 1000) * 1000;
+  timeout.tv_sec = seconds;
+  timeout.tv_usec = useconds;
+
+  /* sigset_t set; */
+  /* sigemptyset(&set); */
+  /* sigaddset(&set, SIGALRM); */
+  /* pthread_sigmask(SIG_BLOCK, &set, NULL); */
+
+  select(0, NULL, NULL, NULL, &timeout);
+
+  /* pthread_sigmask(SIG_UNBLOCK, &set, NULL); */
+}
+
+
+static void taosDeleteTimer(void *tharg) {
+  timer_t *pTimer = tharg;
+  timer_delete(*pTimer);
+}
+
+static pthread_t timerThread;
+static timer_t         timerId;
+static volatile bool stopTimer = false;
+static void *taosProcessAlarmSignal(void *tharg) {
+  // Block the signal
+  sigset_t sigset;
+  sigemptyset(&sigset);
+  sigaddset(&sigset, SIGALRM);
+  sigprocmask(SIG_BLOCK, &sigset, NULL);
+  void (*callback)(int) = tharg;
+
+  struct sigevent sevent = {{0}};
+
+  #ifdef _ALPINE
+    sevent.sigev_notify = SIGEV_THREAD;
+    sevent.sigev_value.sival_int = syscall(__NR_gettid);
+  #else
+    sevent.sigev_notify = SIGEV_THREAD_ID;
+    sevent._sigev_un._tid = syscall(__NR_gettid);
+  #endif
+  
+  sevent.sigev_signo = SIGALRM;
+
+  if (timer_create(CLOCK_REALTIME, &sevent, &timerId) == -1) {
+    uError("Failed to create timer");
+  }
+
+  pthread_cleanup_push(taosDeleteTimer, &timerId);
+
+  struct itimerspec ts;
+  ts.it_value.tv_sec = 0;
+  ts.it_value.tv_nsec = 1000000 * MSECONDS_PER_TICK;
+  ts.it_interval.tv_sec = 0;
+  ts.it_interval.tv_nsec = 1000000 * MSECONDS_PER_TICK;
+
+  if (timer_settime(timerId, 0, &ts, NULL)) {
+    uError("Failed to init timer");
+    return NULL;
+  }
+
+  int signo;
+  while (!stopTimer) {
+    if (sigwait(&sigset, &signo)) {
+      uError("Failed to wait signal: number %d", signo);
+      continue;
+    }
+    /* printf("Signal handling: number %d ......\n", signo); */
+
+    callback(0);
+  }
+  
+  pthread_cleanup_pop(1);
+
+  return NULL;
+}
+
+int taosInitTimer(void (*callback)(int), int ms) {
+  pthread_attr_t tattr;
+  pthread_attr_init(&tattr);
+  int code = pthread_create(&timerThread, &tattr, taosProcessAlarmSignal, callback);
+  pthread_attr_destroy(&tattr);
+  if (code != 0) {
+    uError("failed to create timer thread");
+    return -1;
+  }
+  return 0;
+}
+
+void taosUninitTimer() {
+  stopTimer = true;
+  pthread_join(timerThread, NULL);
+}
+
+#endif

src/os/src/detail/osUtil.c

@@ -0,0 +1,26 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#define _DEFAULT_SOURCE
+#include "os.h"
+
+#ifndef TAOS_OS_FUNC_UTIL
+
+int64_t tsosStr2int64(char *str) {
+  char *endptr = NULL;
+  return strtoll(str, &endptr, 10);
+}
+
+#endif

src/os/src/linux64/CMakeLists.txt

@@ -0,0 +1,12 @@
+CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
+PROJECT(TDengine)
+
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/common/inc)
+AUX_SOURCE_DIRECTORY(. SRC)
+
+ADD_LIBRARY(os ${SRC})
+TARGET_LINK_LIBRARIES(os m rt)
+

src/os/src/linux64/linuxEnv.c

@@ -0,0 +1,30 @@
+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#define _DEFAULT_SOURCE
+#include "os.h"
+#include "tglobal.h"
+#include "tulog.h"
+
+void osInit() {
+  strcpy(configDir, "/etc/taos");
+  strcpy(tsVnodeDir, "");
+  strcpy(tsDnodeDir, "");
+  strcpy(tsMnodeDir, "");
+  strcpy(tsDataDir, "/var/lib/taos");
+  strcpy(tsLogDir, "/var/log/taos");
+  strcpy(tsScriptDir, "/etc/taos");
+  strcpy(tsOsName, "Linux");
+}

src/os/src/windows64/twindows.c

@@ -398,7 +398,7 @@ char *strndup(const char *s, size_t n) {
 void taosSetCoreDump() {}
 
 #ifdef _TD_GO_DLL_
-int64_t str2int64(char *str) {
+int64_t tsosStr2int64(char *str) {
   char *endptr = NULL;
   return strtoll(str, &endptr, 10);
 }

src/plugins/http/CMakeLists.txt

@@ -3,6 +3,7 @@ PROJECT(TDengine)
 
 IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/client/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/query/inc)
@@ -10,7 +11,6 @@ IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/deps/zlib-1.2.11/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/deps/cJson/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/deps/lz4/inc)
-  INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
   INCLUDE_DIRECTORIES(inc)
   AUX_SOURCE_DIRECTORY(src SRC)
   ADD_LIBRARY(http ${SRC})

src/plugins/http/src/httpContext.c

@@ -58,7 +58,7 @@ static void httpDestroyContext(void *data) {
 }
 
 bool httpInitContexts() {
-  tsHttpServer.contextCache = taosCacheInit(TSDB_DATA_TYPE_BIGINT, 2, false, httpDestroyContext, "restc");
+  tsHttpServer.contextCache = taosCacheInit(TSDB_DATA_TYPE_BIGINT, 2, true, httpDestroyContext, "restc");
   if (tsHttpServer.contextCache == NULL) {
     httpError("failed to init context cache");
     return false;

src/plugins/http/src/httpJson.c

@@ -45,7 +45,7 @@ int httpWriteBufByFd(struct HttpContext* pContext, const char* buf, int sz) {
 
   do {
     if (pContext->fd > 2){
-      len = (int)send(pContext->fd, buf + writeLen, (size_t)(sz - writeLen), MSG_NOSIGNAL);
+      len = (int)taosSend(pContext->fd, buf + writeLen, (size_t)(sz - writeLen), MSG_NOSIGNAL);
     }
     else {
       return sz;

src/plugins/monitor/CMakeLists.txt

@@ -3,11 +3,11 @@ PROJECT(TDengine)
 
 IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/client/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/query/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/common/inc)
-  INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
   INCLUDE_DIRECTORIES(inc)
   AUX_SOURCE_DIRECTORY(./src SRC)
   ADD_LIBRARY(monitor ${SRC})

src/plugins/mqtt/CMakeLists.txt

@@ -3,6 +3,7 @@ PROJECT(TDengine)
 
 IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/client/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/query/inc)
@@ -10,7 +11,6 @@ IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/deps/cJson/inc)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/deps/MQTT-C/include)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/deps/MQTT-C/examples/templates)
-  INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
   INCLUDE_DIRECTORIES(inc)
   AUX_SOURCE_DIRECTORY(src SRC)
   ADD_LIBRARY(mqtt ${SRC})

src/query/CMakeLists.txt

@@ -1,9 +1,9 @@
 CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
 PROJECT(TDengine)
 
-INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
-INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/common/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/tsdb/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/client/inc)

src/query/inc/qExecutor.h

@@ -20,8 +20,8 @@
 #include "hash.h"
 #include "qFill.h"
 #include "qResultbuf.h"
+#include "qSqlparser.h"
 #include "qTsbuf.h"
-#include "qsqlparser.h"
 #include "query.h"
 #include "taosdef.h"
 #include "tarray.h"
@@ -43,7 +43,7 @@ typedef struct SSqlGroupbyExpr {
 
 typedef struct SPosInfo {
   int32_t pageId;
-  int16_t rowId;
+  int32_t rowId;
 } SPosInfo;
 
 typedef struct SWindowStatus {
@@ -177,13 +177,18 @@ typedef struct SQueryRuntimeEnv {
   SDiskbasedResultBuf* pResultBuf;       // query result buffer based on blocked-wised disk file
 } SQueryRuntimeEnv;
 
+enum {
+  QUERY_RESULT_NOT_READY = 1,
+  QUERY_RESULT_READY     = 2,
+};
+
 typedef struct SQInfo {
   void*            signature;
   int32_t          pointsInterpo;
   int32_t          code;  // error code to returned to client
-  sem_t            dataReady;
+//  sem_t            dataReady;
+
   void*            tsdb;
-  void*            param;
   int32_t          vgId;
   STableGroupInfo  tableGroupInfo;       // table id list < only includes the STable list>
   STableGroupInfo  tableqinfoGroupInfo;  // this is a group array list, including SArray<STableQueryInfo*> structure
@@ -200,8 +205,11 @@ typedef struct SQInfo {
    */
   int32_t          tableIndex;
   int32_t          numOfGroupResultPages;
-  void*            pBuf; // allocated buffer for STableQueryInfo, sizeof(STableQueryInfo)*numOfTables;
+  void*            pBuf;        // allocated buffer for STableQueryInfo, sizeof(STableQueryInfo)*numOfTables;
 
+  pthread_mutex_t  lock;        // used to synchronize the rsp/query threads
+  int32_t          dataReady;   // denote if query result is ready or not
+  void*            rspContext;  // response context
 } SQInfo;
 
 #endif  // TDENGINE_QUERYEXECUTOR_H

src/query/inc/qExtbuffer.h

@@ -19,7 +19,6 @@
 extern "C" {
 #endif
 
-
 #include "os.h"
 #include "taosmsg.h"
 
@@ -28,9 +27,9 @@ extern "C" {
 #include "tdataformat.h"
 #include "talgo.h"
 
-#define DEFAULT_PAGE_SIZE (1024L*4)  // 16k larger than the SHistoInfo
+#define MAX_TMPFILE_PATH_LENGTH        PATH_MAX
-#define MAX_TMPFILE_PATH_LENGTH PATH_MAX
 #define INITIAL_ALLOCATION_BUFFER_SIZE 64
+#define DEFAULT_PAGE_SIZE              (4096L)  // 16k larger than the SHistoInfo
 
 typedef enum EXT_BUFFER_FLUSH_MODEL {
   /*
@@ -126,7 +125,7 @@ typedef struct tExtMemBuffer {
  * @param pModel
  * @return
  */
-tExtMemBuffer *createExtMemBuffer(int32_t inMemSize, int32_t elemSize, SColumnModel *pModel);
+tExtMemBuffer *createExtMemBuffer(int32_t inMemSize, int32_t elemSize, int32_t pagesize, SColumnModel *pModel);
 
 /**
  *

src/query/inc/qResultbuf.h

@@ -13,50 +13,85 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
 
-#ifndef TDENGINE_VNODEQUERYUTIL_H
+#ifndef TDENGINE_QRESULTBUF_H
-#define TDENGINE_VNODEQUERYUTIL_H
+#define TDENGINE_QRESULTBUF_H
 
 #ifdef __cplusplus
 extern "C" {
 #endif
 
+#include <tlist.h>
 #include "hash.h"
 #include "os.h"
 #include "qExtbuffer.h"
+#include "tlockfree.h"
 
 typedef struct SArray* SIDList;
 
+typedef struct SPageDiskInfo {
+  int32_t offset;
+  int32_t length;
+} SPageDiskInfo;
+
+typedef struct SPageInfo {
+  SListNode*    pn;       // point to list node
+  int32_t       pageId;
+  SPageDiskInfo info;
+  void*         pData;
+  bool          used;     // set current page is in used
+} SPageInfo;
+
+typedef struct SFreeListItem {
+  int32_t offset;
+  int32_t len;
+} SFreeListItem;
+
+typedef struct SResultBufStatis {
+  int32_t flushBytes;
+  int32_t loadBytes;
+  int32_t getPages;
+  int32_t releasePages;
+  int32_t flushPages;
+} SResultBufStatis;
+
 typedef struct SDiskbasedResultBuf {
   int32_t   numOfRowsPerPage;
   int32_t   numOfPages;
   int64_t   totalBufSize;
-  int32_t   fd;                  // data file fd
+  int64_t   fileSize;            // disk file size
+  FILE*     file;
   int32_t   allocateId;          // allocated page id
-  int32_t   incStep;             // minimum allocated pages
-  void*     pBuf;                // mmap buffer pointer
   char*     path;                // file path
   int32_t   pageSize;            // current used page size
   int32_t   inMemPages;          // numOfPages that are allocated in memory
-  SHashObj* idsTable;            // id hash table
+  SHashObj* groupSet;            // id hash table
-  SIDList   list;                // for each id, there is a page id list
+  SHashObj* all;
-
+  SList*    lruList;
-  void*     iBuf;                // inmemory buf
-  void*     handle;              // for debug purpose
   void*     emptyDummyIdList;    // dummy id list
+  void*     assistBuf;           // assistant buffer for compress/decompress data
+  SArray*   pFree;               // free area in file
+  bool      comp;                // compressed before flushed to disk
+  int32_t   nextPos;             // next page flush position
+
+  const void*      handle;        // for debug purpose
+  SResultBufStatis statis;
 } SDiskbasedResultBuf;
 
-#define DEFAULT_INTERN_BUF_PAGE_SIZE (1024L)
+#define DEFAULT_INTERN_BUF_PAGE_SIZE  (4096L)
 #define DEFAULT_INMEM_BUF_PAGES       10
+#define PAGE_INFO_INITIALIZER         (SPageDiskInfo){-1, -1}
 
 /**
  * create disk-based result buffer
  * @param pResultBuf
- * @param size
  * @param rowSize
+ * @param pagesize
+ * @param inMemPages
+ * @param handle
  * @return
  */
-int32_t createDiskbasedResultBuffer(SDiskbasedResultBuf** pResultBuf, int32_t numOfPages, int32_t rowSize, int32_t pagesize,
+int32_t createDiskbasedResultBuffer(SDiskbasedResultBuf** pResultBuf, int32_t rowSize, int32_t pagesize,
-    int32_t inMemPages, void* handle);
+                                    int32_t inMemBufSize, const void* handle);
 
 /**
  *
@@ -72,7 +107,7 @@ tFilePage* getNewDataBuf(SDiskbasedResultBuf* pResultBuf, int32_t groupId, int32
  * @param pResultBuf
  * @return
  */
-int32_t getNumOfRowsPerPage(SDiskbasedResultBuf* pResultBuf);
+size_t getNumOfRowsPerPage(const SDiskbasedResultBuf* pResultBuf);
 
 /**
  *
@@ -88,42 +123,52 @@ SIDList getDataBufPagesIdList(SDiskbasedResultBuf* pResultBuf, int32_t groupId);
  * @param id
  * @return
  */
-static FORCE_INLINE tFilePage* getResBufPage(SDiskbasedResultBuf* pResultBuf, int32_t id) {
+tFilePage* getResBufPage(SDiskbasedResultBuf* pResultBuf, int32_t id);
-  if (id < pResultBuf->inMemPages) {
+
-    return (tFilePage*) ((char*) pResultBuf->iBuf + id * pResultBuf->pageSize);
+/**
-  } else {
+ * release the referenced buf pages
-    return (tFilePage*) ((char*) pResultBuf->pBuf + (id - pResultBuf->inMemPages) * pResultBuf->pageSize);
+ * @param pResultBuf
-  }
+ * @param page
-}
+ */
+void releaseResBufPage(SDiskbasedResultBuf* pResultBuf, void* page);
+
+/**
+ *
+ * @param pResultBuf
+ * @param pi
+ */
+void releaseResBufPageInfo(SDiskbasedResultBuf* pResultBuf, SPageInfo* pi);
+
+
 /**
  * get the total buffer size in the format of disk file
  * @param pResultBuf
  * @return
  */
-int32_t getResBufSize(SDiskbasedResultBuf* pResultBuf);
+size_t getResBufSize(const SDiskbasedResultBuf* pResultBuf);
 
 /**
  * get the number of groups in the result buffer
  * @param pResultBuf
  * @return
  */
-int32_t getNumOfResultBufGroupId(SDiskbasedResultBuf* pResultBuf);
+size_t getNumOfResultBufGroupId(const SDiskbasedResultBuf* pResultBuf);
 
 /**
  * destroy result buffer
  * @param pResultBuf
  */
-void destroyResultBuf(SDiskbasedResultBuf* pResultBuf, void* handle);
+void destroyResultBuf(SDiskbasedResultBuf* pResultBuf);
 
 /**
  *
  * @param pList
  * @return
  */
-int32_t getLastPageId(SIDList pList);
+SPageInfo* getLastPageInfo(SIDList pList);
 
 #ifdef __cplusplus
 }
 #endif
 
-#endif  // TDENGINE_VNODEQUERYUTIL_H
+#endif  // TDENGINE_QRESULTBUF_H

src/query/inc/qUtil.h

@@ -45,13 +45,14 @@ bool isWindowResClosed(SWindowResInfo *pWindowResInfo, int32_t slot);
 
 int32_t createQueryResultInfo(SQuery *pQuery, SWindowResult *pResultRow, bool isSTableQuery, size_t interBufSize);
 
-static FORCE_INLINE char *getPosInResultPage(SQueryRuntimeEnv *pRuntimeEnv, int32_t columnIndex, SWindowResult *pResult) {
+static FORCE_INLINE char *getPosInResultPage(SQueryRuntimeEnv *pRuntimeEnv, int32_t columnIndex, SWindowResult *pResult,
+    tFilePage* page) {
   assert(pResult != NULL && pRuntimeEnv != NULL);
 
   SQuery    *pQuery = pRuntimeEnv->pQuery;
-  tFilePage *page = getResBufPage(pRuntimeEnv->pResultBuf, pResult->pos.pageId);
+//  tFilePage *page = getResBufPage(pRuntimeEnv->pResultBuf, pResult->pos.pageId);
-  int32_t realRowId = pResult->pos.rowId * GET_ROW_PARAM_FOR_MULTIOUTPUT(pQuery, pRuntimeEnv->topBotQuery, pRuntimeEnv->stableQuery);
 
+  int32_t realRowId = pResult->pos.rowId * GET_ROW_PARAM_FOR_MULTIOUTPUT(pQuery, pRuntimeEnv->topBotQuery, pRuntimeEnv->stableQuery);
   return ((char *)page->data) + pRuntimeEnv->offset[columnIndex] * pRuntimeEnv->numOfRowsPerPage +
       pQuery->pSelectExpr[columnIndex].bytes * realRowId;
 }

src/query/src/qAst.c

@@ -18,8 +18,8 @@
 
 #include "exception.h"
 #include "qAst.h"
+#include "qSqlparser.h"
 #include "qSyntaxtreefunction.h"
-#include "qsqlparser.h"
 #include "taosdef.h"
 #include "taosmsg.h"
 #include "tarray.h"

src/query/src/qExecutor.c

@@ -221,7 +221,7 @@ void updateNumOfResult(SQueryRuntimeEnv *pRuntimeEnv, int32_t numOfRes) {
 }
 
 static int32_t getGroupResultId(int32_t groupIndex) {
-  int32_t base = 200000;
+  int32_t base = 20000000;
   return base + (groupIndex * 10000);
 }
 
@@ -478,10 +478,14 @@ static int32_t addNewWindowResultBuf(SWindowResult *pWindowRes, SDiskbasedResult
   if (taosArrayGetSize(list) == 0) {
     pData = getNewDataBuf(pResultBuf, sid, &pageId);
   } else {
-    pageId = getLastPageId(list);
+    SPageInfo* pi = getLastPageInfo(list);
-    pData = getResBufPage(pResultBuf, pageId);
+    pData = getResBufPage(pResultBuf, pi->pageId);
+    pageId = pi->pageId;
 
     if (pData->num >= numOfRowsPerPage) {
+      // release current page first, and prepare the next one
+      releaseResBufPageInfo(pResultBuf, pi);
+
       pData = getNewDataBuf(pResultBuf, sid, &pageId);
       if (pData != NULL) {
         assert(pData->num == 0);  // number of elements must be 0 for new allocated buffer
@@ -497,6 +501,8 @@ static int32_t addNewWindowResultBuf(SWindowResult *pWindowRes, SDiskbasedResult
   if (pWindowRes->pos.pageId == -1) {  // not allocated yet, allocate new buffer
     pWindowRes->pos.pageId = pageId;
     pWindowRes->pos.rowId = pData->num++;
+
+    assert(pWindowRes->pos.pageId >= 0);
   }
 
   return 0;
@@ -1490,8 +1496,6 @@ static int32_t setupQueryRuntimeEnv(SQueryRuntimeEnv *pRuntimeEnv, int16_t order
     goto _clean;
   }
 
-  qDebug("QInfo:%p setup runtime env1", GET_QINFO_ADDR(pRuntimeEnv));
-
   pRuntimeEnv->offset[0] = 0;
   for (int32_t i = 0; i < pQuery->numOfOutput; ++i) {
     SSqlFuncMsg *pSqlFuncMsg = &pQuery->pSelectExpr[i].base;
@@ -1536,8 +1540,6 @@ static int32_t setupQueryRuntimeEnv(SQueryRuntimeEnv *pRuntimeEnv, int16_t order
       }
     }
 
-    qDebug("QInfo:%p setup runtime env2", GET_QINFO_ADDR(pRuntimeEnv));
-
     // set the order information for top/bottom query
     int32_t functionId = pCtx->functionId;
 
@@ -1558,25 +1560,19 @@ static int32_t setupQueryRuntimeEnv(SQueryRuntimeEnv *pRuntimeEnv, int16_t order
     }
   }
 
-  qDebug("QInfo:%p setup runtime env3", GET_QINFO_ADDR(pRuntimeEnv));
-
   char* buf = (char*) pRuntimeEnv->resultInfo + sizeof(SResultInfo) * pQuery->numOfOutput;
 
   // set the intermediate result output buffer
   setWindowResultInfo(pRuntimeEnv->resultInfo, pQuery, pRuntimeEnv->stableQuery, buf);
 
-  qDebug("QInfo:%p setup runtime env4", GET_QINFO_ADDR(pRuntimeEnv));
-
   // if it is group by normal column, do not set output buffer, the output buffer is pResult
   if (!pRuntimeEnv->groupbyNormalCol && !pRuntimeEnv->stableQuery) {
     resetCtxOutputBuf(pRuntimeEnv);
   }
 
-  qDebug("QInfo:%p setup runtime env5", GET_QINFO_ADDR(pRuntimeEnv));
-
   setCtxTagColumnInfo(pRuntimeEnv, pRuntimeEnv->pCtx);
 
-  qDebug("QInfo:%p init completed", GET_QINFO_ADDR(pRuntimeEnv));
+  qDebug("QInfo:%p init runtime completed", GET_QINFO_ADDR(pRuntimeEnv));
   return TSDB_CODE_SUCCESS;
 
 _clean:
@@ -1615,7 +1611,7 @@ static void teardownQueryRuntimeEnv(SQueryRuntimeEnv *pRuntimeEnv) {
 
   pRuntimeEnv->pFillInfo = taosDestoryFillInfo(pRuntimeEnv->pFillInfo);
 
-  destroyResultBuf(pRuntimeEnv->pResultBuf, pQInfo);
+  destroyResultBuf(pRuntimeEnv->pResultBuf);
   tsdbCleanupQueryHandle(pRuntimeEnv->pQueryHandle);
   tsdbCleanupQueryHandle(pRuntimeEnv->pSecQueryHandle);
 
@@ -2111,9 +2107,6 @@ int32_t loadDataBlockOnDemand(SQueryRuntimeEnv *pRuntimeEnv, void* pQueryHandle,
     }
     
     if (!needToLoadDataBlock(pRuntimeEnv, *pStatis, pRuntimeEnv->pCtx, pBlockInfo->rows)) {
-#if defined(_DEBUG_VIEW)
-      qDebug("QInfo:%p block discarded by per-filter", GET_QINFO_ADDR(pRuntimeEnv));
-#endif
       // current block has been discard due to filter applied
       pRuntimeEnv->summary.discardBlocks += 1;
       qDebug("QInfo:%p data block discard, brange:%"PRId64 "-%"PRId64", rows:%d", GET_QINFO_ADDR(pRuntimeEnv),
@@ -2446,6 +2439,8 @@ static void doMerge(SQueryRuntimeEnv *pRuntimeEnv, int64_t timestamp, SWindowRes
   SQuery *        pQuery = pRuntimeEnv->pQuery;
   SQLFunctionCtx *pCtx = pRuntimeEnv->pCtx;
 
+  tFilePage *page = getResBufPage(pRuntimeEnv->pResultBuf, pWindowRes->pos.pageId);
+
   for (int32_t i = 0; i < pQuery->numOfOutput; ++i) {
     int32_t functionId = pQuery->pSelectExpr[i].base.functionId;
     if (!mergeFlag) {
@@ -2458,7 +2453,7 @@ static void doMerge(SQueryRuntimeEnv *pRuntimeEnv, int64_t timestamp, SWindowRes
 
     pCtx[i].hasNull = true;
     pCtx[i].nStartQueryTimestamp = timestamp;
-    pCtx[i].aInputElemBuf = getPosInResultPage(pRuntimeEnv, i, pWindowRes);
+    pCtx[i].aInputElemBuf = getPosInResultPage(pRuntimeEnv, i, pWindowRes, page);
 
     // in case of tag column, the tag information should be extracted from input buffer
     if (functionId == TSDB_FUNC_TAG_DUMMY || functionId == TSDB_FUNC_TAG) {
@@ -2615,14 +2610,16 @@ int32_t tableResultComparFn(const void *pLeft, const void *pRight, void *param)
 
   SWindowResInfo *pWindowResInfo1 = &supporter->pTableQueryInfo[left]->windowResInfo;
   SWindowResult * pWindowRes1 = getWindowResult(pWindowResInfo1, leftPos);
+  tFilePage *page1 = getResBufPage(pRuntimeEnv->pResultBuf, pWindowRes1->pos.pageId);
 
-  char *b1 = getPosInResultPage(pRuntimeEnv, PRIMARYKEY_TIMESTAMP_COL_INDEX, pWindowRes1);
+  char *b1 = getPosInResultPage(pRuntimeEnv, PRIMARYKEY_TIMESTAMP_COL_INDEX, pWindowRes1, page1);
   TSKEY leftTimestamp = GET_INT64_VAL(b1);
 
   SWindowResInfo *pWindowResInfo2 = &supporter->pTableQueryInfo[right]->windowResInfo;
   SWindowResult * pWindowRes2 = getWindowResult(pWindowResInfo2, rightPos);
+  tFilePage *page2 = getResBufPage(pRuntimeEnv->pResultBuf, pWindowRes2->pos.pageId);
 
-  char *b2 = getPosInResultPage(pRuntimeEnv, PRIMARYKEY_TIMESTAMP_COL_INDEX, pWindowRes2);
+  char *b2 = getPosInResultPage(pRuntimeEnv, PRIMARYKEY_TIMESTAMP_COL_INDEX, pWindowRes2, page2);
   TSKEY rightTimestamp = GET_INT64_VAL(b2);
 
   if (leftTimestamp == rightTimestamp) {
@@ -2685,35 +2682,26 @@ void copyResToQueryResultBuf(SQInfo *pQInfo, SQuery *pQuery) {
   int32_t id = getGroupResultId(pQInfo->groupIndex - 1);
   SIDList list = getDataBufPagesIdList(pResultBuf, pQInfo->offset + id);
 
-  int32_t total = 0;
   int32_t size = taosArrayGetSize(list);
-  for (int32_t i = 0; i < size; ++i) {
-    int32_t* pgId = taosArrayGet(list, i);
-    tFilePage *pData = getResBufPage(pResultBuf, *pgId);
-    total += pData->num;
-  }
-
-  int32_t rows = total;
 
   int32_t offset = 0;
   for (int32_t j = 0; j < size; ++j) {
-    int32_t* pgId = taosArrayGet(list, j);
+    SPageInfo* pi = *(SPageInfo**) taosArrayGet(list, j);
-    tFilePage *pData = getResBufPage(pResultBuf, *pgId);
+    tFilePage *pData = getResBufPage(pResultBuf, pi->pageId);
 
     for (int32_t i = 0; i < pQuery->numOfOutput; ++i) {
       int32_t bytes = pRuntimeEnv->pCtx[i].outputBytes;
       char *  pDest = pQuery->sdata[i]->data;
-
+      memcpy(pDest + offset * bytes, pData->data + pRuntimeEnv->offset[i] * pData->num, bytes * pData->num);
-      memcpy(pDest + offset * bytes, pData->data + pRuntimeEnv->offset[i] * pData->num,
-             bytes * pData->num);
     }
 
+//    rows += pData->num;
     offset += pData->num;
   }
 
   assert(pQuery->rec.rows == 0);
 
-  pQuery->rec.rows += rows;
+  pQuery->rec.rows += offset;
   pQInfo->offset += 1;
 }
 
@@ -2777,7 +2765,6 @@ int32_t mergeIntoGroupResultImpl(SQInfo *pQInfo, SArray *pGroup) {
     assert(pQInfo->numOfGroupResultPages == 0);
     return 0;
   } else if (numOfTables == 1) { // no need to merge results since only one table in each group
-
   }
 
   SCompSupporter cs = {pTableList, posList, pQInfo};
@@ -2802,8 +2789,9 @@ int32_t mergeIntoGroupResultImpl(SQInfo *pQInfo, SArray *pGroup) {
 
     SWindowResInfo *pWindowResInfo = &pTableList[pos]->windowResInfo;
     SWindowResult * pWindowRes = getWindowResult(pWindowResInfo, cs.position[pos]);
+    tFilePage *page = getResBufPage(pRuntimeEnv->pResultBuf, pWindowRes->pos.pageId);
 
-    char *b = getPosInResultPage(pRuntimeEnv, PRIMARYKEY_TIMESTAMP_COL_INDEX, pWindowRes);
+    char *b = getPosInResultPage(pRuntimeEnv, PRIMARYKEY_TIMESTAMP_COL_INDEX, pWindowRes, page);
     TSKEY ts = GET_INT64_VAL(b);
 
     assert(ts == pWindowRes->window.skey);
@@ -3517,9 +3505,11 @@ void setWindowResOutputBuf(SQueryRuntimeEnv *pRuntimeEnv, SWindowResult *pResult
   SQuery *pQuery = pRuntimeEnv->pQuery;
 
   // Note: pResult->pos[i]->num == 0, there is only fixed number of results for each group
+  tFilePage *page = getResBufPage(pRuntimeEnv->pResultBuf, pResult->pos.pageId);
+
   for (int32_t i = 0; i < pQuery->numOfOutput; ++i) {
     SQLFunctionCtx *pCtx = &pRuntimeEnv->pCtx[i];
-    pCtx->aOutputBuf = getPosInResultPage(pRuntimeEnv, i, pResult);
+    pCtx->aOutputBuf = getPosInResultPage(pRuntimeEnv, i, pResult, page);
 
     int32_t functionId = pQuery->pSelectExpr[i].base.functionId;
     if (functionId == TSDB_FUNC_TOP || functionId == TSDB_FUNC_BOTTOM || functionId == TSDB_FUNC_DIFF) {
@@ -3542,6 +3532,8 @@ void setWindowResOutputBufInitCtx(SQueryRuntimeEnv *pRuntimeEnv, SWindowResult *
   SQuery *pQuery = pRuntimeEnv->pQuery;
 
   // Note: pResult->pos[i]->num == 0, there is only fixed number of results for each group
+  tFilePage* bufPage = getResBufPage(pRuntimeEnv->pResultBuf, pResult->pos.pageId);
+
   for (int32_t i = 0; i < pQuery->numOfOutput; ++i) {
     SQLFunctionCtx *pCtx = &pRuntimeEnv->pCtx[i];
 
@@ -3550,7 +3542,7 @@ void setWindowResOutputBufInitCtx(SQueryRuntimeEnv *pRuntimeEnv, SWindowResult *
       continue;
     }
 
-    pCtx->aOutputBuf = getPosInResultPage(pRuntimeEnv, i, pResult);
+    pCtx->aOutputBuf = getPosInResultPage(pRuntimeEnv, i, pResult, bufPage);
     pCtx->currentStage = 0;
 
     int32_t functionId = pCtx->functionId;
@@ -3713,11 +3705,13 @@ static int32_t doCopyToSData(SQInfo *pQInfo, SWindowResInfo *pResultInfo, int32_
       pQInfo->groupIndex += 1;
     }
 
+    tFilePage *page = getResBufPage(pRuntimeEnv->pResultBuf, result[i].pos.pageId);
+
     for (int32_t j = 0; j < pQuery->numOfOutput; ++j) {
       int32_t size = pRuntimeEnv->pCtx[j].outputBytes;
 
       char *out = pQuery->sdata[j]->data + numOfResult * size;
-      char *in = getPosInResultPage(pRuntimeEnv, j, &result[i]);
+      char *in = getPosInResultPage(pRuntimeEnv, j, &result[i], page);
       memcpy(out, in + oldOffset * size, size * numOfRowsToCopy);
     }
 
@@ -4238,10 +4232,10 @@ int32_t doInitQInfo(SQInfo *pQInfo, STSBuf *pTsBuf, void *tsdb, int32_t vgId, bo
   int32_t ps = DEFAULT_PAGE_SIZE;
   int32_t rowsize = 0;
   getIntermediateBufInfo(pRuntimeEnv, &ps, &rowsize);
+  int32_t TWOMB = 1024*1024*2;
 
   if (isSTableQuery && !onlyQueryTags(pRuntimeEnv->pQuery)) {
-    int32_t numOfPages = getInitialPageNum(pQInfo);
+    code = createDiskbasedResultBuffer(&pRuntimeEnv->pResultBuf, rowsize, ps, TWOMB, pQInfo);
-    code = createDiskbasedResultBuffer(&pRuntimeEnv->pResultBuf, numOfPages, rowsize, ps, numOfPages, pQInfo);
     if (code != TSDB_CODE_SUCCESS) {
       return code;
     }
@@ -4269,8 +4263,7 @@ int32_t doInitQInfo(SQInfo *pQInfo, STSBuf *pTsBuf, void *tsdb, int32_t vgId, bo
   } else if (pRuntimeEnv->groupbyNormalCol || QUERY_IS_INTERVAL_QUERY(pQuery)) {
     int32_t numOfResultRows = getInitialPageNum(pQInfo);
     getIntermediateBufInfo(pRuntimeEnv, &ps, &rowsize);
-
+    code = createDiskbasedResultBuffer(&pRuntimeEnv->pResultBuf, rowsize, ps, TWOMB, pQInfo);
-    code = createDiskbasedResultBuffer(&pRuntimeEnv->pResultBuf, numOfResultRows, rowsize, ps, numOfResultRows, pQInfo);
     if (code != TSDB_CODE_SUCCESS) {
       return code;
     }
@@ -5894,16 +5887,11 @@ static SQInfo *createQInfoImpl(SQueryTableMsg *pQueryMsg, SArray* pTableIdList,
   }
 
   pQInfo->arrTableIdInfo = taosArrayInit(tableIndex, sizeof(STableIdInfo));
+  pQInfo->dataReady = QUERY_RESULT_NOT_READY;
+  pthread_mutex_init(&pQInfo->lock, NULL);
 
   pQuery->pos = -1;
   pQuery->window = pQueryMsg->window;
-
-  if (sem_init(&pQInfo->dataReady, 0, 0) != 0) {
-    int32_t code = TAOS_SYSTEM_ERROR(errno);
-    qError("QInfo:%p init dataReady sem failed, reason:%s", pQInfo, tstrerror(code));
-    goto _cleanup;
-  }
-
   colIdCheck(pQuery);
 
   qDebug("qmsg:%p QInfo:%p created", pQueryMsg, pQInfo);
@@ -5945,7 +5933,7 @@ static bool isValidQInfo(void *param) {
   return (sig == (uint64_t)pQInfo);
 }
 
-static int32_t initQInfo(SQueryTableMsg *pQueryMsg, void *tsdb, int32_t vgId, SQInfo *pQInfo, bool isSTable, void* param) {
+static int32_t initQInfo(SQueryTableMsg *pQueryMsg, void *tsdb, int32_t vgId, SQInfo *pQInfo, bool isSTable) {
   int32_t code = TSDB_CODE_SUCCESS;
   SQuery *pQuery = pQInfo->runtimeEnv.pQuery;
 
@@ -5965,18 +5953,12 @@ static int32_t initQInfo(SQueryTableMsg *pQueryMsg, void *tsdb, int32_t vgId, SQ
            pQuery->window.ekey, pQuery->order.order);
     setQueryStatus(pQuery, QUERY_COMPLETED);
     pQInfo->tableqinfoGroupInfo.numOfTables = 0;
-
-    sem_post(&pQInfo->dataReady);
     return TSDB_CODE_SUCCESS;
   }
 
-  pQInfo->param = param;
-
   if (pQInfo->tableqinfoGroupInfo.numOfTables == 0) {
     qDebug("QInfo:%p no table qualified for tag filter, abort query", pQInfo);
     setQueryStatus(pQuery, QUERY_COMPLETED);
-  
-    sem_post(&pQInfo->dataReady);
     return TSDB_CODE_SUCCESS;
   }
 
@@ -6018,7 +6000,6 @@ static void freeQInfo(SQInfo *pQInfo) {
     tfree(pQuery->sdata[col]);
   }
 
-  sem_destroy(&(pQInfo->dataReady));
   teardownQueryRuntimeEnv(&pQInfo->runtimeEnv);
 
   for (int32_t i = 0; i < pQuery->numOfFilterCols; ++i) {
@@ -6170,7 +6151,7 @@ typedef struct SQueryMgmt {
   pthread_mutex_t lock;
 } SQueryMgmt;
 
-int32_t qCreateQueryInfo(void* tsdb, int32_t vgId, SQueryTableMsg* pQueryMsg, void* param, qinfo_t* pQInfo) {
+int32_t qCreateQueryInfo(void* tsdb, int32_t vgId, SQueryTableMsg* pQueryMsg, qinfo_t* pQInfo) {
   assert(pQueryMsg != NULL && tsdb != NULL);
 
   int32_t code = TSDB_CODE_SUCCESS;
@@ -6266,7 +6247,7 @@ int32_t qCreateQueryInfo(void* tsdb, int32_t vgId, SQueryTableMsg* pQueryMsg, vo
     goto _over;
   }
 
-  code = initQInfo(pQueryMsg, tsdb, vgId, *pQInfo, isSTableQuery, param);
+  code = initQInfo(pQueryMsg, tsdb, vgId, *pQInfo, isSTableQuery);
 
 _over:
   free(tagCond);
@@ -6306,24 +6287,32 @@ void qDestroyQueryInfo(qinfo_t qHandle) {
   freeQInfo(pQInfo);
 }
 
-void qTableQuery(qinfo_t qinfo) {
+static void setQueryResultReady(SQInfo* pQInfo) {
+  pthread_mutex_lock(&pQInfo->lock);
+  pQInfo->dataReady = QUERY_RESULT_READY;
+  pthread_mutex_unlock(&pQInfo->lock);
+}
+
+bool qTableQuery(qinfo_t qinfo) {
   SQInfo *pQInfo = (SQInfo *)qinfo;
 
   if (pQInfo == NULL || pQInfo->signature != pQInfo) {
     qDebug("QInfo:%p has been freed, no need to execute", pQInfo);
-    return;
+    return false;
   }
 
   if (IS_QUERY_KILLED(pQInfo)) {
     qDebug("QInfo:%p it is already killed, abort", pQInfo);
-    sem_post(&pQInfo->dataReady);
+    setQueryResultReady(pQInfo);
-    return;
+    return false;
   }
 
   if (pQInfo->tableqinfoGroupInfo.numOfTables == 0) {
+    setQueryStatus(pQInfo->runtimeEnv.pQuery, QUERY_COMPLETED);
+    setQueryResultReady(pQInfo);
+
     qDebug("QInfo:%p no table exists for query, abort", pQInfo);
-    sem_post(&pQInfo->dataReady);
+    return false;
-    return;
   }
 
   // error occurs, record the error code and return to client
@@ -6331,8 +6320,9 @@ void qTableQuery(qinfo_t qinfo) {
   if (ret != TSDB_CODE_SUCCESS) {
     pQInfo->code = ret;
     qDebug("QInfo:%p query abort due to error/cancel occurs, code:%s", pQInfo, tstrerror(pQInfo->code));
-    sem_post(&pQInfo->dataReady);
+
-    return;
+    setQueryResultReady(pQInfo);
+    return false;
   }
 
   qDebug("QInfo:%p query task is launched", pQInfo);
@@ -6357,10 +6347,20 @@ void qTableQuery(qinfo_t qinfo) {
            pQInfo, pQuery->rec.rows, pQuery->rec.total + pQuery->rec.rows);
   }
 
-  sem_post(&pQInfo->dataReady);
+  bool buildRes = false;
+  pthread_mutex_lock(&pQInfo->lock);
+  pQInfo->dataReady = QUERY_RESULT_READY;
+
+  if (pQInfo->rspContext != NULL) {
+    buildRes = true;
+  }
+
+
+  pthread_mutex_unlock(&pQInfo->lock);
+  return buildRes;
 }
 
-int32_t qRetrieveQueryResultInfo(qinfo_t qinfo) {
+int32_t qRetrieveQueryResultInfo(qinfo_t qinfo, bool* buildRes, void* pRspContext) {
   SQInfo *pQInfo = (SQInfo *)qinfo;
 
   if (pQInfo == NULL || !isValidQInfo(pQInfo)) {
@@ -6373,11 +6373,21 @@ int32_t qRetrieveQueryResultInfo(qinfo_t qinfo) {
     return pQInfo->code;
   }
 
-  sem_wait(&pQInfo->dataReady);
+  int32_t code = TSDB_CODE_SUCCESS;
-  qDebug("QInfo:%p retrieve result info, rowsize:%d, rows:%"PRId64", code:%d", pQInfo, pQuery->rowSize, pQuery->rec.rows,
+  pthread_mutex_lock(&pQInfo->lock);
-         pQInfo->code);
+  if (pQInfo->dataReady == QUERY_RESULT_READY) {
+    *buildRes = true;
+    qDebug("QInfo:%p retrieve result info, rowsize:%d, rows:%"PRId64", code:%d", pQInfo, pQuery->rowSize, pQuery->rec.rows,
+           pQInfo->code);
+  } else {
+    *buildRes = false;
+    qDebug("QInfo:%p retrieve req set query return result after paused", pQInfo);
+    pQInfo->rspContext = pRspContext;
+  }
 
-  return pQInfo->code;
+  code = pQInfo->code;
+  pthread_mutex_unlock(&pQInfo->lock);
+  return code;
 }
 
 bool qHasMoreResultsToRetrieve(qinfo_t qinfo) {
@@ -6389,6 +6399,7 @@ bool qHasMoreResultsToRetrieve(qinfo_t qinfo) {
   }
 
   SQuery *pQuery = pQInfo->runtimeEnv.pQuery;
+
   bool ret = false;
   if (Q_STATUS_EQUAL(pQuery->status, QUERY_OVER)) {
     ret = false;
@@ -6407,7 +6418,7 @@ bool qHasMoreResultsToRetrieve(qinfo_t qinfo) {
   return ret;
 }
 
-int32_t qDumpRetrieveResult(qinfo_t qinfo, SRetrieveTableRsp **pRsp, int32_t *contLen) {
+int32_t qDumpRetrieveResult(qinfo_t qinfo, SRetrieveTableRsp **pRsp, int32_t *contLen, bool* continueExec) {
   SQInfo *pQInfo = (SQInfo *)qinfo;
 
   if (pQInfo == NULL || !isValidQInfo(pQInfo)) {
@@ -6417,8 +6428,10 @@ int32_t qDumpRetrieveResult(qinfo_t qinfo, SRetrieveTableRsp **pRsp, int32_t *co
   SQueryRuntimeEnv* pRuntimeEnv = &pQInfo->runtimeEnv;
   SQuery *pQuery = pQInfo->runtimeEnv.pQuery;
   size_t  size = getResultSize(pQInfo, &pQuery->rec.rows);
+
   size += sizeof(int32_t);
   size += sizeof(STableIdInfo) * taosArrayGetSize(pQInfo->arrTableIdInfo);
+
   *contLen = size + sizeof(SRetrieveTableRsp);
 
   // todo proper handle failed to allocate memory,
@@ -6427,6 +6440,7 @@ int32_t qDumpRetrieveResult(qinfo_t qinfo, SRetrieveTableRsp **pRsp, int32_t *co
   if (*pRsp == NULL) {
     return TSDB_CODE_QRY_OUT_OF_MEMORY;
   }
+
   (*pRsp)->numOfRows = htonl(pQuery->rec.rows);
 
   int32_t code = pQInfo->code;
@@ -6434,8 +6448,8 @@ int32_t qDumpRetrieveResult(qinfo_t qinfo, SRetrieveTableRsp **pRsp, int32_t *co
     (*pRsp)->offset = htobe64(pQuery->limit.offset);
     (*pRsp)->useconds = htobe64(pRuntimeEnv->summary.elapsedTime);
   } else {
-    (*pRsp)->offset = 0;
     (*pRsp)->useconds = 0;
+    (*pRsp)->offset = 0;
   }
   
   (*pRsp)->precision = htons(pQuery->precision);
@@ -6446,10 +6460,20 @@ int32_t qDumpRetrieveResult(qinfo_t qinfo, SRetrieveTableRsp **pRsp, int32_t *co
     code = pQInfo->code;
   }
 
+  pQInfo->rspContext = NULL;
+  pQInfo->dataReady = QUERY_RESULT_NOT_READY;
+
   if (IS_QUERY_KILLED(pQInfo) || Q_STATUS_EQUAL(pQuery->status, QUERY_OVER)) {
     (*pRsp)->completed = 1;  // notify no more result to client
   }
 
+  if (qHasMoreResultsToRetrieve(pQInfo)) {
+    *continueExec = true;
+  } else { // failed to dump result, free qhandle immediately
+    *continueExec = false;
+    qKillQuery(pQInfo);
+  }
+
   return code;
 }
 
@@ -6460,7 +6484,7 @@ int32_t qKillQuery(qinfo_t qinfo) {
     return TSDB_CODE_QRY_INVALID_QHANDLE;
   }
 
-  sem_post(&pQInfo->dataReady);
+//  sem_post(&pQInfo->dataReady);
   setQueryKilled(pQInfo);
   return TSDB_CODE_SUCCESS;
 }
@@ -6607,6 +6631,13 @@ static void buildTagQueryResult(SQInfo* pQInfo) {
   setQueryStatus(pQuery, QUERY_COMPLETED);
 }
 
+void* qGetResultRetrieveMsg(qinfo_t qinfo) {
+  SQInfo* pQInfo = (SQInfo*) qinfo;
+  assert(pQInfo != NULL);
+
+  return pQInfo->rspContext;
+}
+
 void freeqinfoFn(void *qhandle) {
   void** handle = qhandle;
   if (handle == NULL || *handle == NULL) {
@@ -6618,19 +6649,21 @@ void freeqinfoFn(void *qhandle) {
 }
 
 void* qOpenQueryMgmt(int32_t vgId) {
-  const int32_t REFRESH_HANDLE_INTERVAL = 2; // every 2 seconds, refresh handle pool
+  const int32_t REFRESH_HANDLE_INTERVAL = 30; // every 30 seconds, refresh handle pool
 
   char cacheName[128] = {0};
   sprintf(cacheName, "qhandle_%d", vgId);
 
-  SQueryMgmt* pQueryHandle = calloc(1, sizeof(SQueryMgmt));
+  SQueryMgmt* pQueryMgmt = calloc(1, sizeof(SQueryMgmt));
 
-  pQueryHandle->qinfoPool = taosCacheInit(TSDB_DATA_TYPE_BIGINT, REFRESH_HANDLE_INTERVAL, true, freeqinfoFn, cacheName);
+  pQueryMgmt->qinfoPool = taosCacheInit(TSDB_DATA_TYPE_BIGINT, REFRESH_HANDLE_INTERVAL, true, freeqinfoFn, cacheName);
-  pQueryHandle->closed    = false;
+  pQueryMgmt->closed    = false;
-  pthread_mutex_init(&pQueryHandle->lock, NULL);
+  pQueryMgmt->vgId      = vgId;
+
+  pthread_mutex_init(&pQueryMgmt->lock, NULL);
 
   qDebug("vgId:%d, open querymgmt success", vgId);
-  return pQueryHandle;
+  return pQueryMgmt;
 }
 
 static void queryMgmtKillQueryFn(void* handle) {
@@ -6670,7 +6703,7 @@ void qCleanupQueryMgmt(void* pQMgmt) {
   pthread_mutex_destroy(&pQueryMgmt->lock);
   tfree(pQueryMgmt);
 
-  qDebug("vgId:%d querymgmt cleanup completed", vgId);
+  qDebug("vgId:%d queryMgmt cleanup completed", vgId);
 }
 
 void** qRegisterQInfo(void* pMgmt, uint64_t qInfo) {
@@ -6727,3 +6760,4 @@ void** qReleaseQInfo(void* pMgmt, void* pQInfo, bool needFree) {
   return 0;
 }
 
+

src/query/src/qExtbuffer.c

@@ -28,10 +28,10 @@
 /*
  * SColumnModel is deeply copy
  */
-tExtMemBuffer* createExtMemBuffer(int32_t inMemSize, int32_t elemSize, SColumnModel *pModel) {
+tExtMemBuffer* createExtMemBuffer(int32_t inMemSize, int32_t elemSize, int32_t pagesize, SColumnModel *pModel) {
   tExtMemBuffer* pMemBuffer = (tExtMemBuffer *)calloc(1, sizeof(tExtMemBuffer));
 
-  pMemBuffer->pageSize = DEFAULT_PAGE_SIZE;
+  pMemBuffer->pageSize = pagesize;
   pMemBuffer->inMemCapacity = ALIGN8(inMemSize) / pMemBuffer->pageSize;
   pMemBuffer->nElemSize = elemSize;
 

src/query/src/qParserImpl.c

@@ -14,7 +14,7 @@
  */
 
 #include "os.h"
-#include "qsqlparser.h"
+#include "qSqlparser.h"
 #include "queryLog.h"
 #include "taosdef.h"
 #include "taosmsg.h"

src/query/src/qPercentile.c

@@ -535,7 +535,7 @@ void tMemBucketPut(tMemBucket *pBucket, void *data, int32_t numOfRows) {
 
     if (pSeg->pBuffer[slotIdx] == NULL) {
       pSeg->pBuffer[slotIdx] = createExtMemBuffer(pBucket->numOfTotalPages * pBucket->pageSize, pBucket->nElemSize,
-                                                  pBucket->pOrderDesc->pColumnModel);
+                                                  pBucket->pageSize, pBucket->pOrderDesc->pColumnModel);
       pSeg->pBuffer[slotIdx]->flushModel = SINGLE_APPEND_MODEL;
       pBucket->pOrderDesc->pColumnModel->capacity = pSeg->pBuffer[slotIdx]->numOfElemsPerPage;
     }

src/query/src/qResultbuf.c

@@ -1,221 +1,439 @@
 #include "qResultbuf.h"
+#include "stddef.h"
+#include "tscompression.h"
 #include "hash.h"
 #include "qExtbuffer.h"
 #include "queryLog.h"
 #include "taoserror.h"
 
-int32_t createDiskbasedResultBuffer(SDiskbasedResultBuf** pResultBuf, int32_t numOfPages, int32_t rowSize,
+#define GET_DATA_PAYLOAD(_p) ((_p)->pData + POINTER_BYTES)
-    int32_t pagesize, int32_t inMemPages, void* handle) {
 
+int32_t createDiskbasedResultBuffer(SDiskbasedResultBuf** pResultBuf, int32_t rowSize, int32_t pagesize,
+                                    int32_t inMemBufSize, const void* handle) {
   *pResultBuf = calloc(1, sizeof(SDiskbasedResultBuf));
+
   SDiskbasedResultBuf* pResBuf = *pResultBuf;
   if (pResBuf == NULL) {
     return TSDB_CODE_COM_OUT_OF_MEMORY;  
   }
 
-  pResBuf->pageSize   = pagesize;
+  pResBuf->pageSize     = pagesize;
-  pResBuf->numOfPages = inMemPages;  // all pages are in buffer in the first place
+  pResBuf->numOfPages   = 0;                        // all pages are in buffer in the first place
-  pResBuf->inMemPages = inMemPages;
+  pResBuf->totalBufSize = 0;
-  assert(inMemPages <= numOfPages);
+  pResBuf->inMemPages   = inMemBufSize/pagesize;    // maximum allowed pages, it is a soft limit.
+  pResBuf->allocateId   = -1;
+  pResBuf->comp         = true;
+  pResBuf->file         = NULL;
+  pResBuf->handle       = handle;
+  pResBuf->fileSize = 0;
+
+  // at least more than 2 pages must be in memory
+  assert(inMemBufSize >= pagesize * 2);
 
   pResBuf->numOfRowsPerPage = (pagesize - sizeof(tFilePage)) / rowSize;
-
+  pResBuf->lruList = tdListNew(POINTER_BYTES);
-  pResBuf->totalBufSize = pResBuf->numOfPages * pagesize;
-  pResBuf->incStep = 4;
-  pResBuf->allocateId = -1;
-
-  pResBuf->iBuf = calloc(pResBuf->inMemPages, pResBuf->pageSize);
 
   // init id hash table
-  pResBuf->idsTable = taosHashInit(numOfPages, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), false);
+  pResBuf->groupSet  = taosHashInit(10, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), false);
-  pResBuf->list = taosArrayInit(numOfPages, POINTER_BYTES);
+  pResBuf->assistBuf = malloc(pResBuf->pageSize + 2); // EXTRA BYTES
+  pResBuf->all = taosHashInit(10, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), false);
 
   char path[PATH_MAX] = {0};
-  getTmpfilePath("tsdb_qbuf", path);
+  getTmpfilePath("qbuf", path);
   pResBuf->path = strdup(path);
 
-  pResBuf->fd = FD_INITIALIZER;
-  pResBuf->pBuf = NULL;
   pResBuf->emptyDummyIdList = taosArrayInit(1, sizeof(int32_t));
 
-  qDebug("QInfo:%p create resBuf for output, page size:%d, initial pages:%d, %" PRId64 "bytes", handle,
+  qDebug("QInfo:%p create resBuf for output, page size:%d, inmem buf pages:%d, file:%s", handle, pResBuf->pageSize,
-      pResBuf->pageSize, pResBuf->numOfPages, pResBuf->totalBufSize);
+         pResBuf->inMemPages, pResBuf->path);
-  
+
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t getNumOfResultBufGroupId(SDiskbasedResultBuf* pResultBuf) { return taosHashGetSize(pResultBuf->idsTable); }
+static int32_t createDiskFile(SDiskbasedResultBuf* pResultBuf) {
-
+  pResultBuf->file = fopen(pResultBuf->path, "wb+");
-int32_t getResBufSize(SDiskbasedResultBuf* pResultBuf) { return pResultBuf->totalBufSize; }
+  if (pResultBuf->file == NULL) {
-
-#define NUM_OF_PAGES_ON_DISK(_r) ((_r)->numOfPages - (_r)->inMemPages)
-#define FILE_SIZE_ON_DISK(_r) (NUM_OF_PAGES_ON_DISK(_r) * (_r)->pageSize)
-
-static int32_t createDiskResidesBuf(SDiskbasedResultBuf* pResultBuf) {
-  pResultBuf->fd = open(pResultBuf->path, O_CREAT | O_RDWR | O_TRUNC, 0666);
-  if (!FD_VALID(pResultBuf->fd)) {
     qError("failed to create tmp file: %s on disk. %s", pResultBuf->path, strerror(errno));
     return TAOS_SYSTEM_ERROR(errno);
   }
 
-  assert(pResultBuf->numOfPages == pResultBuf->inMemPages);
-  pResultBuf->numOfPages += pResultBuf->incStep;
-
-  int32_t ret = ftruncate(pResultBuf->fd, NUM_OF_PAGES_ON_DISK(pResultBuf) * pResultBuf->pageSize);
-  if (ret != TSDB_CODE_SUCCESS) {
-    qError("failed to create tmp file: %s on disk. %s", pResultBuf->path, strerror(errno));
-    return TAOS_SYSTEM_ERROR(errno);
-  }
-
-  pResultBuf->pBuf = mmap(NULL, FILE_SIZE_ON_DISK(pResultBuf), PROT_READ | PROT_WRITE, MAP_SHARED, pResultBuf->fd, 0);
-  if (pResultBuf->pBuf == MAP_FAILED) {
-    qError("QInfo:%p failed to map temp file: %s. %s", pResultBuf->handle, pResultBuf->path, strerror(errno));
-    return TAOS_SYSTEM_ERROR(errno);
-  }
-
-  pResultBuf->totalBufSize = pResultBuf->numOfPages * pResultBuf->pageSize;
   return TSDB_CODE_SUCCESS;
 }
 
-static int32_t extendDiskFileSize(SDiskbasedResultBuf* pResultBuf, int32_t incNumOfPages) {
+static char* doCompressData(void* data, int32_t srcSize, int32_t *dst, SDiskbasedResultBuf* pResultBuf) { // do nothing
-  assert(pResultBuf->numOfPages * pResultBuf->pageSize == pResultBuf->totalBufSize);
+  if (!pResultBuf->comp) {
-  int32_t ret = TSDB_CODE_SUCCESS;
+    *dst = srcSize;
+    return data;
+  }
 
-  if (pResultBuf->pBuf == NULL) {
+  *dst = tsCompressString(data, srcSize, 1, pResultBuf->assistBuf, srcSize, ONE_STAGE_COMP, NULL, 0);
-    assert(pResultBuf->fd == FD_INITIALIZER);
 
-    if ((ret = createDiskResidesBuf(pResultBuf)) != TSDB_CODE_SUCCESS) {
+  memcpy(data, pResultBuf->assistBuf, *dst);
-      return ret;
+  return data;
+}
+
+static char* doDecompressData(void* data, int32_t srcSize, int32_t *dst, SDiskbasedResultBuf* pResultBuf) { // do nothing
+  if (!pResultBuf->comp) {
+    *dst = srcSize;
+    return data;
+  }
+
+  *dst = tsDecompressString(data, srcSize, 1, pResultBuf->assistBuf, pResultBuf->pageSize, ONE_STAGE_COMP, NULL, 0);
+
+  memcpy(data, pResultBuf->assistBuf, *dst);
+  return data;
+}
+
+static int32_t allocatePositionInFile(SDiskbasedResultBuf* pResultBuf, size_t size) {
+  if (pResultBuf->pFree == NULL) {
+    return pResultBuf->nextPos;
+  } else {
+    int32_t offset = -1;
+
+    size_t num = taosArrayGetSize(pResultBuf->pFree);
+    for(int32_t i = 0; i < num; ++i) {
+      SFreeListItem* pi = taosArrayGet(pResultBuf->pFree, i);
+      if (pi->len >= size) {
+        offset = pi->offset;
+        pi->offset += size;
+        pi->len -= size;
+
+        return offset;
+      }
+    }
+
+    // no available recycle space, allocate new area in file
+    return pResultBuf->nextPos;
+  }
+}
+
+static char* doFlushPageToDisk(SDiskbasedResultBuf* pResultBuf, SPageInfo* pg) {
+  assert(!pg->used && pg->pData != NULL);
+
+  int32_t size = -1;
+  char* t = doCompressData(GET_DATA_PAYLOAD(pg), pResultBuf->pageSize, &size, pResultBuf);
+
+  // this page is flushed to disk for the first time
+  if (pg->info.offset == -1) {
+    pg->info.offset = allocatePositionInFile(pResultBuf, size);
+    pResultBuf->nextPos += size;
+
+    fseek(pResultBuf->file, pg->info.offset, SEEK_SET);
+    /*int32_t ret =*/ fwrite(t, 1, size, pResultBuf->file);
+
+    if (pResultBuf->fileSize < pg->info.offset + pg->info.length) {
+      pResultBuf->fileSize = pg->info.offset + pg->info.length;
     }
   } else {
-    ret = munmap(pResultBuf->pBuf, FILE_SIZE_ON_DISK(pResultBuf));
+    // length becomes greater, current space is not enough, allocate new place, otherwise, do nothing
-    pResultBuf->numOfPages += incNumOfPages;
+    if (pg->info.length < size) {
+      // 1. add current space to free list
+      taosArrayPush(pResultBuf->pFree, &pg->info);
 
-    /*
+      // 2. allocate new position, and update the info
-     * disk-based output buffer is exhausted, try to extend the disk-based buffer, the available disk space may
+      pg->info.offset = allocatePositionInFile(pResultBuf, size);
-     * be insufficient
+      pResultBuf->nextPos += size;
-     */
-    ret = ftruncate(pResultBuf->fd, NUM_OF_PAGES_ON_DISK(pResultBuf) * pResultBuf->pageSize);
-    if (ret != TSDB_CODE_SUCCESS) {
-      //    dError("QInfo:%p failed to create intermediate result output file:%s. %s", pQInfo, pSupporter->extBufFile,
-      //           strerror(errno));
-      return TSDB_CODE_QRY_NO_DISKSPACE;
     }
 
-    pResultBuf->totalBufSize = pResultBuf->numOfPages * pResultBuf->pageSize;
+    //3. write to disk.
-    pResultBuf->pBuf = mmap(NULL, FILE_SIZE_ON_DISK(pResultBuf), PROT_READ | PROT_WRITE, MAP_SHARED, pResultBuf->fd, 0);
+    fseek(pResultBuf->file, pg->info.offset, SEEK_SET);
+    fwrite(t, size, 1, pResultBuf->file);
 
-    if (pResultBuf->pBuf == MAP_FAILED) {
+    if (pResultBuf->fileSize < pg->info.offset + pg->info.length) {
-      //    dError("QInfo:%p failed to map temp file: %s. %s", pQInfo, pSupporter->extBufFile, strerror(errno));
+      pResultBuf->fileSize = pg->info.offset + pg->info.length;
-      return TSDB_CODE_QRY_OUT_OF_MEMORY;
     }
   }
 
-  return TSDB_CODE_SUCCESS;
+  char* ret = pg->pData;
+  memset(ret, 0, pResultBuf->pageSize);
+
+  pg->pData = NULL;
+  pg->info.length = size;
+
+  pResultBuf->statis.flushBytes += pg->info.length;
+
+  return ret;
 }
 
-#define NO_AVAILABLE_PAGES(_b) ((_b)->allocateId == (_b)->numOfPages - 1)
+static char* flushPageToDisk(SDiskbasedResultBuf* pResultBuf, SPageInfo* pg) {
+  int32_t ret = TSDB_CODE_SUCCESS;
+  assert(pResultBuf->numOfPages * pResultBuf->pageSize == pResultBuf->totalBufSize && pResultBuf->numOfPages >= pResultBuf->inMemPages);
 
-static FORCE_INLINE int32_t getGroupIndex(SDiskbasedResultBuf* pResultBuf, int32_t groupId) {
+  if (pResultBuf->file == NULL) {
-  assert(pResultBuf != NULL);
+    if ((ret = createDiskFile(pResultBuf)) != TSDB_CODE_SUCCESS) {
-
+      terrno = ret;
-  char* p = taosHashGet(pResultBuf->idsTable, (const char*)&groupId, sizeof(int32_t));
-  if (p == NULL) {  // it is a new group id
-    return -1;
-  }
-
-  int32_t slot = GET_INT32_VAL(p);
-  assert(slot >= 0 && slot < taosHashGetSize(pResultBuf->idsTable));
-
-  return slot;
-}
-
-static int32_t addNewGroupId(SDiskbasedResultBuf* pResultBuf, int32_t groupId) {
-  int32_t num = getNumOfResultBufGroupId(pResultBuf); // the num is the newest allocated group id slot
-  taosHashPut(pResultBuf->idsTable, (const char*)&groupId, sizeof(int32_t), &num, sizeof(int32_t));
-
-  SArray* pa = taosArrayInit(1, sizeof(int32_t));
-  taosArrayPush(pResultBuf->list, &pa);
-
-  assert(taosArrayGetSize(pResultBuf->list) == taosHashGetSize(pResultBuf->idsTable));
-  return num;
-}
-
-static void registerPageId(SDiskbasedResultBuf* pResultBuf, int32_t groupId, int32_t pageId) {
-  int32_t slot = getGroupIndex(pResultBuf, groupId);
-  if (slot < 0) {
-    slot = addNewGroupId(pResultBuf, groupId);
-  }
-
-  SIDList pList = taosArrayGetP(pResultBuf->list, slot);
-  taosArrayPush(pList, &pageId);
-}
-
-tFilePage* getNewDataBuf(SDiskbasedResultBuf* pResultBuf, int32_t groupId, int32_t* pageId) {
-  if (NO_AVAILABLE_PAGES(pResultBuf)) {
-    if (extendDiskFileSize(pResultBuf, pResultBuf->incStep) != TSDB_CODE_SUCCESS) {
       return NULL;
     }
   }
 
-  // register new id in this group
+  return doFlushPageToDisk(pResultBuf, pg);
-  *pageId = (++pResultBuf->allocateId);
-  registerPageId(pResultBuf, groupId, *pageId);
-
-  // clear memory for the new page
-  tFilePage* page = getResBufPage(pResultBuf, *pageId);
-  memset(page, 0, pResultBuf->pageSize);
-  
-  return page;
 }
 
-int32_t getNumOfRowsPerPage(SDiskbasedResultBuf* pResultBuf) { return pResultBuf->numOfRowsPerPage; }
+// load file block data in disk
+static char* loadPageFromDisk(SDiskbasedResultBuf* pResultBuf, SPageInfo* pg) {
+  int32_t ret = fseek(pResultBuf->file, pg->info.offset, SEEK_SET);
+  ret = fread(GET_DATA_PAYLOAD(pg), 1, pg->info.length, pResultBuf->file);
+  if (ret != pg->info.length) {
+    terrno = errno;
+    return NULL;
+  }
 
-SIDList getDataBufPagesIdList(SDiskbasedResultBuf* pResultBuf, int32_t groupId) {
+  pResultBuf->statis.loadBytes += pg->info.length;
-  int32_t slot = getGroupIndex(pResultBuf, groupId);
+
-  if (slot < 0) {
+  int32_t fullSize = 0;
-    return pResultBuf->emptyDummyIdList;
+  doDecompressData(GET_DATA_PAYLOAD(pg), pg->info.length, &fullSize, pResultBuf);
+
+  return GET_DATA_PAYLOAD(pg);
+}
+
+#define NO_AVAILABLE_PAGES(_b) ((_b)->numOfPages >= (_b)->inMemPages)
+
+static SIDList addNewGroup(SDiskbasedResultBuf* pResultBuf, int32_t groupId) {
+  assert(taosHashGet(pResultBuf->groupSet, (const char*) &groupId, sizeof(int32_t)) == NULL);
+
+  SArray* pa = taosArrayInit(1, POINTER_BYTES);
+  int32_t ret = taosHashPut(pResultBuf->groupSet, (const char*)&groupId, sizeof(int32_t), &pa, POINTER_BYTES);
+  assert(ret == 0);
+
+  return pa;
+}
+
+static SPageInfo* registerPage(SDiskbasedResultBuf* pResultBuf, int32_t groupId, int32_t pageId) {
+  SIDList list = NULL;
+
+  char** p = taosHashGet(pResultBuf->groupSet, (const char*)&groupId, sizeof(int32_t));
+  if (p == NULL) {  // it is a new group id
+    list = addNewGroup(pResultBuf, groupId);
   } else {
-    return taosArrayGetP(pResultBuf->list, slot);
+    list = (SIDList) (*p);
+  }
+
+  pResultBuf->numOfPages += 1;
+
+  SPageInfo* ppi = malloc(sizeof(SPageInfo));//{ .info = PAGE_INFO_INITIALIZER, .pageId = pageId, .pn = NULL};
+  ppi->info     = PAGE_INFO_INITIALIZER;
+  ppi->pageId   = pageId;
+  ppi->pData    = NULL;
+  ppi->pn       = NULL;
+  ppi->used     = true;
+
+  return *(SPageInfo**) taosArrayPush(list, &ppi);
+}
+
+static SListNode* getEldestUnrefedPage(SDiskbasedResultBuf* pResultBuf) {
+  SListIter iter = {0};
+  tdListInitIter(pResultBuf->lruList, &iter, TD_LIST_BACKWARD);
+
+  SListNode* pn = NULL;
+  while((pn = tdListNext(&iter)) != NULL) {
+    assert(pn != NULL);
+
+    SPageInfo* pageInfo = *(SPageInfo**) pn->data;
+    assert(pageInfo->pageId >= 0 && pageInfo->pn == pn);
+
+    if (!pageInfo->used) {
+      break;
+    }
+  }
+
+  return pn;
+}
+
+static char* evicOneDataPage(SDiskbasedResultBuf* pResultBuf) {
+  char* bufPage = NULL;
+  SListNode* pn = getEldestUnrefedPage(pResultBuf);
+
+  // all pages are referenced by user, try to allocate new space
+  if (pn == NULL) {
+    int32_t prev = pResultBuf->inMemPages;
+    pResultBuf->inMemPages = pResultBuf->inMemPages * 1.5;
+
+    qWarn("%p in memory buf page not sufficient, expand from %d to %d, page size:%d", pResultBuf, prev,
+          pResultBuf->inMemPages, pResultBuf->pageSize);
+  } else {
+    pResultBuf->statis.flushPages += 1;
+    tdListPopNode(pResultBuf->lruList, pn);
+
+    SPageInfo* d = *(SPageInfo**) pn->data;
+    assert(d->pn == pn);
+
+    d->pn = NULL;
+    tfree(pn);
+
+    bufPage = flushPageToDisk(pResultBuf, d);
+  }
+
+  return bufPage;
+}
+
+static void lruListPushFront(SList *pList, SPageInfo* pi) {
+  tdListPrepend(pList, &pi);
+  SListNode* front = tdListGetHead(pList);
+  pi->pn = front;
+}
+
+static void lruListMoveToFront(SList *pList, SPageInfo* pi) {
+  tdListPopNode(pList, pi->pn);
+  tdListPrependNode(pList, pi->pn);
+}
+
+tFilePage* getNewDataBuf(SDiskbasedResultBuf* pResultBuf, int32_t groupId, int32_t* pageId) {
+  pResultBuf->statis.getPages += 1;
+
+  char* availablePage = NULL;
+  if (NO_AVAILABLE_PAGES(pResultBuf)) {
+    availablePage = evicOneDataPage(pResultBuf);
+  }
+
+  // register new id in this group
+  *pageId = (++pResultBuf->allocateId);
+
+  // register page id info
+  SPageInfo* pi = registerPage(pResultBuf, groupId, *pageId);
+
+  // add to LRU list
+  assert(listNEles(pResultBuf->lruList) < pResultBuf->inMemPages && pResultBuf->inMemPages > 0);
+
+  lruListPushFront(pResultBuf->lruList, pi);
+
+  // add to hash map
+  taosHashPut(pResultBuf->all, pageId, sizeof(int32_t), &pi, POINTER_BYTES);
+
+  // allocate buf
+  if (availablePage == NULL) {
+    pi->pData = calloc(1, pResultBuf->pageSize + POINTER_BYTES);
+  } else {
+    pi->pData = availablePage;
+  }
+
+  pResultBuf->totalBufSize += pResultBuf->pageSize;
+
+  ((void**)pi->pData)[0] = pi;
+  pi->used = true;
+
+  return GET_DATA_PAYLOAD(pi);
+}
+
+tFilePage* getResBufPage(SDiskbasedResultBuf* pResultBuf, int32_t id) {
+  assert(pResultBuf != NULL && id >= 0);
+  pResultBuf->statis.getPages += 1;
+
+  SPageInfo** pi = taosHashGet(pResultBuf->all, &id, sizeof(int32_t));
+  assert(pi != NULL && *pi != NULL);
+
+  if ((*pi)->pData != NULL) { // it is in memory
+    // no need to update the LRU list if only one page exists
+    if (pResultBuf->numOfPages == 1) {
+      (*pi)->used = true;
+      return GET_DATA_PAYLOAD(*pi);
+    }
+
+    SPageInfo** pInfo = (SPageInfo**) ((*pi)->pn->data);
+    assert(*pInfo == *pi);
+
+    lruListMoveToFront(pResultBuf->lruList, (*pi));
+    (*pi)->used = true;
+
+    return GET_DATA_PAYLOAD(*pi);
+
+  } else { // not in memory
+    assert((*pi)->pData == NULL && (*pi)->pn == NULL && (*pi)->info.length >= 0 && (*pi)->info.offset >= 0);
+
+    char* availablePage = NULL;
+    if (NO_AVAILABLE_PAGES(pResultBuf)) {
+      availablePage = evicOneDataPage(pResultBuf);
+    }
+
+    if (availablePage == NULL) {
+      (*pi)->pData = calloc(1, pResultBuf->pageSize + POINTER_BYTES);
+    } else {
+      (*pi)->pData = availablePage;
+    }
+
+    ((void**)((*pi)->pData))[0] = (*pi);
+
+    lruListPushFront(pResultBuf->lruList, *pi);
+    loadPageFromDisk(pResultBuf, *pi);
+    return GET_DATA_PAYLOAD(*pi);
   }
 }
 
-void destroyResultBuf(SDiskbasedResultBuf* pResultBuf, void* handle) {
+void releaseResBufPage(SDiskbasedResultBuf* pResultBuf, void* page) {
+  assert(pResultBuf != NULL && page != NULL);
+  char* p = (char*) page - POINTER_BYTES;
+
+  SPageInfo* ppi = ((SPageInfo**) p)[0];
+  releaseResBufPageInfo(pResultBuf, ppi);
+}
+
+void releaseResBufPageInfo(SDiskbasedResultBuf* pResultBuf, SPageInfo* pi) {
+  assert(pi->pData != NULL && pi->used);
+
+  pi->used = false;
+  pResultBuf->statis.releasePages += 1;
+}
+
+size_t getNumOfRowsPerPage(const SDiskbasedResultBuf* pResultBuf) { return pResultBuf->numOfRowsPerPage; }
+
+size_t getNumOfResultBufGroupId(const SDiskbasedResultBuf* pResultBuf) { return taosHashGetSize(pResultBuf->groupSet); }
+
+size_t getResBufSize(const SDiskbasedResultBuf* pResultBuf) { return pResultBuf->totalBufSize; }
+
+SIDList getDataBufPagesIdList(SDiskbasedResultBuf* pResultBuf, int32_t groupId) {
+  assert(pResultBuf != NULL);
+
+  char** p = taosHashGet(pResultBuf->groupSet, (const char*)&groupId, sizeof(int32_t));
+  if (p == NULL) {  // it is a new group id
+    return pResultBuf->emptyDummyIdList;
+  } else {
+    return (SArray*) (*p);
+  }
+}
+
+void destroyResultBuf(SDiskbasedResultBuf* pResultBuf) {
   if (pResultBuf == NULL) {
     return;
   }
 
-  if (FD_VALID(pResultBuf->fd)) {
+  if (pResultBuf->file != NULL) {
-    qDebug("QInfo:%p disk-based output buffer closed, total:%" PRId64 " bytes, file created:%s, file size:%d", handle,
+    qDebug("QInfo:%p disk-based output buffer closed, total:%" PRId64 " bytes, file size:%"PRId64" bytes",
-        pResultBuf->totalBufSize, pResultBuf->path, FILE_SIZE_ON_DISK(pResultBuf));
+        pResultBuf->handle, pResultBuf->totalBufSize, pResultBuf->fileSize);
 
-    close(pResultBuf->fd);
+    fclose(pResultBuf->file);
-    munmap(pResultBuf->pBuf, FILE_SIZE_ON_DISK(pResultBuf));
-    pResultBuf->pBuf = NULL;
   } else {
-    qDebug("QInfo:%p disk-based output buffer closed, total:%" PRId64 " bytes, no file created", handle,
+    qDebug("QInfo:%p disk-based output buffer closed, total:%" PRId64 " bytes, no file created", pResultBuf->handle,
            pResultBuf->totalBufSize);
   }
 
   unlink(pResultBuf->path);
   tfree(pResultBuf->path);
 
-  size_t size = taosArrayGetSize(pResultBuf->list);
+  SHashMutableIterator* iter = taosHashCreateIter(pResultBuf->groupSet);
-  for (int32_t i = 0; i < size; ++i) {
+  while(taosHashIterNext(iter)) {
-    SArray* pa = taosArrayGetP(pResultBuf->list, i);
+    SArray** p = (SArray**) taosHashIterGet(iter);
-    taosArrayDestroy(pa);
+    size_t n = taosArrayGetSize(*p);
+    for(int32_t i = 0; i < n; ++i) {
+      SPageInfo* pi = taosArrayGetP(*p, i);
+      tfree(pi->pData);
+      tfree(pi);
+    }
+
+    taosArrayDestroy(*p);
   }
 
-  taosArrayDestroy(pResultBuf->list);
+  taosHashDestroyIter(iter);
-  taosArrayDestroy(pResultBuf->emptyDummyIdList);
-  taosHashCleanup(pResultBuf->idsTable);
 
-  tfree(pResultBuf->iBuf);
+  tdListFree(pResultBuf->lruList);
+  taosArrayDestroy(pResultBuf->emptyDummyIdList);
+  taosHashCleanup(pResultBuf->groupSet);
+  taosHashCleanup(pResultBuf->all);
+
+  tfree(pResultBuf->assistBuf);
   tfree(pResultBuf);
 }
 
-int32_t getLastPageId(SIDList pList) {
+SPageInfo* getLastPageInfo(SIDList pList) {
   size_t size = taosArrayGetSize(pList);
-  return *(int32_t*) taosArrayGet(pList, size - 1);
+  return (SPageInfo*) taosArrayGetP(pList, size - 1);
 }
 

src/query/src/qTsbuf.c

@@ -740,7 +740,7 @@ int32_t tsBufMerge(STSBuf* pDestBuf, const STSBuf* pSrcBuf, int32_t vnodeId) {
   int32_t size = pSrcBuf->fileSize - offset;
 
 #ifdef LINUX
-  ssize_t rc = tsendfile(fileno(pDestBuf->f), fileno(pSrcBuf->f), &offset, size);
+  ssize_t rc = taosTSendFile(fileno(pDestBuf->f), fileno(pSrcBuf->f), &offset, size);
 #else
   ssize_t rc = fsendfile(pDestBuf->f, pSrcBuf->f, &offset, size);
 #endif

src/query/src/qUtil.c

@@ -236,11 +236,13 @@ void clearTimeWindowResBuf(SQueryRuntimeEnv *pRuntimeEnv, SWindowResult *pWindow
   if (pWindowRes == NULL) {
     return;
   }
-  
+
+  tFilePage *page = getResBufPage(pRuntimeEnv->pResultBuf, pWindowRes->pos.pageId);
+
   for (int32_t i = 0; i < pRuntimeEnv->pQuery->numOfOutput; ++i) {
     SResultInfo *pResultInfo = &pWindowRes->resultInfo[i];
     
-    char * s = getPosInResultPage(pRuntimeEnv, i, pWindowRes);
+    char * s = getPosInResultPage(pRuntimeEnv, i, pWindowRes, page);
     size_t size = pRuntimeEnv->pQuery->pSelectExpr[i].bytes;
     memset(s, 0, size);
     
@@ -277,8 +279,11 @@ void copyTimeWindowResBuf(SQueryRuntimeEnv *pRuntimeEnv, SWindowResult *dst, con
     memcpy(pDst->interResultBuf, pSrc->interResultBuf, pDst->bufLen);
     
     // copy the output buffer data from src to dst, the position info keep unchanged
-    char * dstBuf = getPosInResultPage(pRuntimeEnv, i, dst);
+    tFilePage *dstpage = getResBufPage(pRuntimeEnv->pResultBuf, dst->pos.pageId);
-    char * srcBuf = getPosInResultPage(pRuntimeEnv, i, (SWindowResult *)src);
+    char * dstBuf = getPosInResultPage(pRuntimeEnv, i, dst, dstpage);
+
+    tFilePage *srcpage = getResBufPage(pRuntimeEnv->pResultBuf, src->pos.pageId);
+    char * srcBuf = getPosInResultPage(pRuntimeEnv, i, (SWindowResult *)src, srcpage);
     size_t s = pRuntimeEnv->pQuery->pSelectExpr[i].bytes;
     
     memcpy(dstBuf, srcBuf, s);

src/query/src/sql.c

@@ -30,7 +30,7 @@
 #include <string.h>
 #include <assert.h>
 #include <stdbool.h>
-#include "qsqlparser.h"
+#include "qSqlparser.h"
 #include "tcmdtype.h"
 #include "tstoken.h"
 #include "ttokendef.h"

src/query/tests/resultBufferTest.cpp

@@ -18,17 +18,144 @@ void simpleTest() {
   tFilePage* pBufPage = getNewDataBuf(pResultBuf, groupId, &pageId);
   ASSERT_TRUE(pBufPage != NULL);
   
-  ASSERT_EQ(getNumOfRowsPerPage(pResultBuf), (16384L - sizeof(int64_t))/64);
+  ASSERT_EQ(getResBufSize(pResultBuf), 1024);
-  ASSERT_EQ(getResBufSize(pResultBuf), 1000*16384L);
   
   SIDList list = getDataBufPagesIdList(pResultBuf, groupId);
   ASSERT_EQ(taosArrayGetSize(list), 1);
   ASSERT_EQ(getNumOfResultBufGroupId(pResultBuf), 1);
-  
+
-  destroyResultBuf(pResultBuf, NULL);
+  releaseResBufPage(pResultBuf, pBufPage);
+
+  tFilePage* pBufPage1 = getNewDataBuf(pResultBuf, groupId, &pageId);
+
+  tFilePage* t = getResBufPage(pResultBuf, pageId);
+  ASSERT_TRUE(t == pBufPage1);
+
+  tFilePage* pBufPage2 = getNewDataBuf(pResultBuf, groupId, &pageId);
+  tFilePage* t1 = getResBufPage(pResultBuf, pageId);
+  ASSERT_TRUE(t1 == pBufPage2);
+
+  tFilePage* pBufPage3 = getNewDataBuf(pResultBuf, groupId, &pageId);
+  tFilePage* t2 = getResBufPage(pResultBuf, pageId);
+  ASSERT_TRUE(t2 == pBufPage3);
+
+  tFilePage* pBufPage4 = getNewDataBuf(pResultBuf, groupId, &pageId);
+  tFilePage* t3 = getResBufPage(pResultBuf, pageId);
+  ASSERT_TRUE(t3 == pBufPage4);
+
+  tFilePage* pBufPage5 = getNewDataBuf(pResultBuf, groupId, &pageId);
+  tFilePage* t4 = getResBufPage(pResultBuf, pageId);
+  ASSERT_TRUE(t4 == pBufPage5);
+
+  destroyResultBuf(pResultBuf);
+}
+
+void writeDownTest() {
+  SDiskbasedResultBuf* pResultBuf = NULL;
+  int32_t ret = createDiskbasedResultBuffer(&pResultBuf, 1000, 64, 1024, 4, NULL);
+
+  int32_t pageId = 0;
+  int32_t writePageId = 0;
+  int32_t groupId = 0;
+  int32_t nx = 12345;
+
+  tFilePage* pBufPage = getNewDataBuf(pResultBuf, groupId, &pageId);
+  ASSERT_TRUE(pBufPage != NULL);
+
+  *(int32_t*)(pBufPage->data) = nx;
+  writePageId = pageId;
+  releaseResBufPage(pResultBuf, pBufPage);
+
+  tFilePage* pBufPage1 = getNewDataBuf(pResultBuf, groupId, &pageId);
+  tFilePage* t1 = getResBufPage(pResultBuf, pageId);
+  ASSERT_TRUE(t1 == pBufPage1);
+  ASSERT_TRUE(pageId == 1);
+
+  tFilePage* pBufPage2 = getNewDataBuf(pResultBuf, groupId, &pageId);
+  tFilePage* t2 = getResBufPage(pResultBuf, pageId);
+  ASSERT_TRUE(t2 == pBufPage2);
+  ASSERT_TRUE(pageId == 2);
+
+  tFilePage* pBufPage3 = getNewDataBuf(pResultBuf, groupId, &pageId);
+  tFilePage* t3 = getResBufPage(pResultBuf, pageId);
+  ASSERT_TRUE(t3 == pBufPage3);
+  ASSERT_TRUE(pageId == 3);
+
+  tFilePage* pBufPage4 = getNewDataBuf(pResultBuf, groupId, &pageId);
+  tFilePage* t4 = getResBufPage(pResultBuf, pageId);
+  ASSERT_TRUE(t4 == pBufPage4);
+  ASSERT_TRUE(pageId == 4);
+  releaseResBufPage(pResultBuf, t4);
+
+  // flush the written page to disk, and read it out again
+  tFilePage* pBufPagex = getResBufPage(pResultBuf, writePageId);
+  ASSERT_EQ(*(int32_t*)pBufPagex->data, nx);
+
+  SArray* pa = getDataBufPagesIdList(pResultBuf, groupId);
+  ASSERT_EQ(taosArrayGetSize(pa), 5);
+
+  destroyResultBuf(pResultBuf);
+}
+
+void recyclePageTest() {
+  SDiskbasedResultBuf* pResultBuf = NULL;
+  int32_t ret = createDiskbasedResultBuffer(&pResultBuf, 1000, 64, 1024, 4, NULL);
+
+  int32_t pageId = 0;
+  int32_t writePageId = 0;
+  int32_t groupId = 0;
+  int32_t nx = 12345;
+
+  tFilePage* pBufPage = getNewDataBuf(pResultBuf, groupId, &pageId);
+  ASSERT_TRUE(pBufPage != NULL);
+  releaseResBufPage(pResultBuf, pBufPage);
+
+  tFilePage* pBufPage1 = getNewDataBuf(pResultBuf, groupId, &pageId);
+  tFilePage* t1 = getResBufPage(pResultBuf, pageId);
+  ASSERT_TRUE(t1 == pBufPage1);
+  ASSERT_TRUE(pageId == 1);
+
+  tFilePage* pBufPage2 = getNewDataBuf(pResultBuf, groupId, &pageId);
+  tFilePage* t2 = getResBufPage(pResultBuf, pageId);
+  ASSERT_TRUE(t2 == pBufPage2);
+  ASSERT_TRUE(pageId == 2);
+
+  tFilePage* pBufPage3 = getNewDataBuf(pResultBuf, groupId, &pageId);
+  tFilePage* t3 = getResBufPage(pResultBuf, pageId);
+  ASSERT_TRUE(t3 == pBufPage3);
+  ASSERT_TRUE(pageId == 3);
+
+  tFilePage* pBufPage4 = getNewDataBuf(pResultBuf, groupId, &pageId);
+  tFilePage* t4 = getResBufPage(pResultBuf, pageId);
+  ASSERT_TRUE(t4 == pBufPage4);
+  ASSERT_TRUE(pageId == 4);
+  releaseResBufPage(pResultBuf, t4);
+  releaseResBufPage(pResultBuf, t4);
+
+  tFilePage* pBufPage5 = getNewDataBuf(pResultBuf, groupId, &pageId);
+  tFilePage* t5 = getResBufPage(pResultBuf, pageId);
+  ASSERT_TRUE(t5 == pBufPage5);
+  ASSERT_TRUE(pageId == 5);
+
+  // flush the written page to disk, and read it out again
+  tFilePage* pBufPagex = getResBufPage(pResultBuf, writePageId);
+  *(int32_t*)(pBufPagex->data) = nx;
+  writePageId = pageId;   // update the data
+  releaseResBufPage(pResultBuf, pBufPagex);
+
+  tFilePage* pBufPagex1 = getResBufPage(pResultBuf, 1);
+
+  SArray* pa = getDataBufPagesIdList(pResultBuf, groupId);
+  ASSERT_EQ(taosArrayGetSize(pa), 6);
+
+  destroyResultBuf(pResultBuf);
 }
 } // namespace
 
+
 TEST(testCase, resultBufferTest) {
+  srand(time(NULL));
   simpleTest();
+  writeDownTest();
+  recyclePageTest();
 }

src/rpc/CMakeLists.txt

@@ -1,8 +1,8 @@
 CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
 PROJECT(TDengine)
 
-INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/inc)
+INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/os/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/util/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/common/inc)
 INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/deps/cJson/inc)
@@ -12,10 +12,11 @@ INCLUDE_DIRECTORIES(inc)
 IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
   AUX_SOURCE_DIRECTORY(./src SRC)
 ELSEIF (TD_DARWIN_64)
-  LIST(APPEND SRC ./src/thaship.c)
+  #LIST(APPEND SRC ./src/thaship.c)
-  LIST(APPEND SRC ./src/trpc.c)
+  #LIST(APPEND SRC ./src/trpc.c)
-  LIST(APPEND SRC ./src/tstring.c)
+  #LIST(APPEND SRC ./src/tstring.c)
-  LIST(APPEND SRC ./src/tudp.c)
+  #LIST(APPEND SRC ./src/tudp.c)
+  AUX_SOURCE_DIRECTORY(src SRC)
 ELSEIF (TD_WINDOWS_64)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/deps/pthread)
   LIST(APPEND SRC ./src/thaship.c)