Merge branch 'develop' into 2.0

# Conflicts:
#	src/system/CMakeLists.txt
#	src/system/detail/CMakeLists.txt
#	src/system/lite/CMakeLists.txt
#	src/util/inc/textbuffer.h
#	src/vnode/detail/inc/vnodeTagMgmt.h
#	src/vnode/detail/src/vnodeQueryImpl.c
#	src/vnode/detail/src/vnodeSupertableQuery.c

CMakeLists.txt

@@ -36,7 +36,7 @@ ENDIF ()
 IF (NOT DEFINED TD_CLUSTER)
   MESSAGE(STATUS "Build the Lite Version")
   SET(TD_CLUSTER FALSE)
-  SET(TD_LITE TRUE)
+  SET(TD_EDGE TRUE)
 
   SET(TD_COMMUNITY_DIR ${PROJECT_SOURCE_DIR})
   MESSAGE(STATUS "Community directory: " ${TD_COMMUNITY_DIR})
@@ -53,12 +53,18 @@ IF (NOT DEFINED TD_CLUSTER)
   SET(TD_MIPS_32 FALSE)
   SET(TD_DARWIN_64 FALSE)
   SET(TD_WINDOWS_64 FALSE)
+  SET(TD_PAGMODE_LITE FALSE)
+  
+  IF (${PAGMODE} MATCHES "lite")
+    SET(TD_PAGMODE_LITE TRUE)
+  ENDIF ()
 
   # if generate ARM version:
   #   cmake -DCPUTYPE=aarch32 .. or cmake -DCPUTYPE=aarch64
   IF (${CPUTYPE} MATCHES "aarch32")
     SET(TD_ARM TRUE)
     SET(TD_ARM_32 TRUE)
+    SET(TD_PAGMODE_LITE TRUE)
     ADD_DEFINITIONS(-D_TD_ARM_)
     ADD_DEFINITIONS(-D_TD_ARM_32_)
   ELSEIF (${CPUTYPE} MATCHES "aarch64")

README.md

@@ -130,3 +130,8 @@ The TDengine community has also kindly built some of their own connectors! Follo
 # Contribute to TDengine
 
 Please follow the [contribution guidelines](CONTRIBUTING.md) to contribute to the project.
+
+# Join TDengine WeChat Group
+
+Add WeChat “tdengine” to join the group，you can communicate with other users.
+

documentation/tdenginedocs-cn/super-table/index.html

@@ -32,7 +32,7 @@ tags (location binary(20), type int)</code></pre>
 <p>查看数据库内全部STable，及其相关信息，包括STable的名称、创建时间、列数量、标签（TAG）数量、通过该STable建表的数量。 </p></li>
 <li><p>删除超级表</p>
 <pre><code class="mysql language-mysql">DROP TABLE &lt;stable_name&gt;</code></pre>
-<p>Note: 删除STable不会级联删除通过STable创建的表；相反删除STable时要求通过该STable创建的表都已经被删除。</p></li>
+<p>Note: 删除STable时，所有通过该STable创建的表都将被删除。</p></li>
 <li><p>查看属于某STable并满足查询条件的表</p>
 <pre><code class="mysql language-mysql">SELECT TBNAME,[TAG_NAME,…] FROM &lt;stable_name&gt; WHERE &lt;tag_name&gt; &lt;[=|=&lt;|&gt;=|&lt;&gt;] values..&gt; ([AND|OR] …)</code></pre>
 <p>查看属于某STable并满足查询条件的表。说明：TBNAME为关键词，显示通过STable建立的子表表名，查询过程中可以使用针对标签的条件。</p>

documentation/tdenginedocs-en/super-table/index.html

@@ -73,7 +73,7 @@ INTERVAL(10M)</code></pre>
 <p>It lists the STable's schema and tags</p>
 <a class='anchor' id='Drop-a-STable'></a><h3>Drop a STable</h3>
 <pre><code class="mysql language-mysql">DROP TABLE &lt;stable_name&gt;</code></pre>
-<p>To delete a STable, all the tables created via this STable shall be deleted first, otherwise, it will fail.</p>
+<p>To delete a STable, all the tables created via this STable will be deleted.</p>
 <a class='anchor' id='List-the-Associated-Tables-of-a-STable'></a><h3>List the Associated Tables of a STable</h3>
 <pre><code class="mysql language-mysql">SELECT TBNAME,[TAG_NAME, ...] FROM &lt;stable_name&gt; WHERE &lt;tag_name&gt; &lt;[=|=&lt;|&gt;=|&lt;&gt;] values..&gt; ([AND|OR] ...)</code></pre>
 <p>It will list all the tables which satisfy the tag filter conditions. The tables are all created from this specific STable. TBNAME is a new keyword introduced, it is the table name associated with the STable. </p>

packaging/deb/makedeb.sh

@@ -72,10 +72,10 @@ sed -i "2c$debver" ${pkg_dir}/DEBIAN/control
 
 if [ "$verMode" == "cluster" ]; then
   debname="TDengine-server-"${tdengine_ver}-${osType}-${cpuType}
-elif [ "$verMode" == "lite" ]; then
+elif [ "$verMode" == "edge" ]; then
   debname="TDengine-server"-${tdengine_ver}-${osType}-${cpuType}
 else
-  echo "unknow verMode, nor cluster or lite"
+  echo "unknow verMode, nor cluster or edge"
   exit 1
 fi
 

packaging/release.sh

@@ -5,15 +5,20 @@
 set -e
 #set -x
 
-# releash.sh -v [cluster | lite]  -c [aarch32 | aarch64 | x64 | x86 | mips64 ...] -o [Linux | Kylin | Alpine | Raspberrypi | Darwin | Windows | ...]  -V [stable | beta]
+# releash.sh  -v [cluster | edge]  
+#             -c [aarch32 | aarch64 | x64 | x86 | mips64 ...] 
+#             -o [Linux | Kylin | Alpine | Raspberrypi | Darwin | Windows | ...]  
+#             -V [stable | beta]
+#             -l [full | lite]
 
 # set parameters by default value
-verMode=lite     # [cluster, lite]
+verMode=edge     # [cluster, edge]
 verType=stable   # [stable, beta]
 cpuType=x64      # [aarch32 | aarch64 | x64 | x86 | mips64 ...]
 osType=Linux     # [Linux | Kylin | Alpine | Raspberrypi | Darwin | Windows | ...]
+pagMode=full     # [full | lite]
 
-while getopts "hv:V:c:o:" arg
+while getopts "hv:V:c:o:l:" arg
 do
   case $arg in
     v)
@@ -28,12 +33,16 @@ do
       #echo "cpuType=$OPTARG"
       cpuType=$(echo $OPTARG)
       ;;
+    l)
+      #echo "pagMode=$OPTARG"
+      pagMode=$(echo $OPTARG)
+      ;;
     o)
       #echo "osType=$OPTARG"
       osType=$(echo $OPTARG)
       ;;
     h)
-      echo "Usage: `basename $0` -v [cluster | lite]  -c [aarch32 | aarch64 | x64 | x86 | mips64 ...] -o [Linux | Kylin | Alpine | Raspberrypi | Darwin | Windows | ...]  -V [stable | beta]"
+      echo "Usage: `basename $0` -v [cluster | edge]  -c [aarch32 | aarch64 | x64 | x86 | mips64 ...] -o [Linux | Kylin | Alpine | Raspberrypi | Darwin | Windows | ...]  -V [stable | beta] -l [full | lite]"
       exit 0
       ;;
     ?) #unknow option 
@@ -43,7 +52,7 @@ do
   esac
 done
 
-echo "verMode=${verMode} verType=${verType} cpuType=${cpuType} osType=${osType}"
+echo "verMode=${verMode} verType=${verType} cpuType=${cpuType} osType=${osType} pagMode=${pagMode}"
 
 curr_dir=$(pwd)
 
@@ -193,7 +202,7 @@ cd ${compile_dir}
 # check support cpu type
 if [[ "$cpuType" == "x64" ]] || [[ "$cpuType" == "aarch64" ]] || [[ "$cpuType" == "aarch32" ]] || [[ "$cpuType" == "mips64" ]] ; then
     if [ "$verMode" != "cluster" ]; then
-        cmake ../ -DCPUTYPE=${cpuType}
+      cmake ../ -DCPUTYPE=${cpuType} -DPAGMODE=${pagMode}
     else
       cmake ../../ -DCPUTYPE=${cpuType}
     fi
@@ -235,8 +244,8 @@ if [ "$osType" != "Darwin" ]; then
     echo "====do tar.gz package for all systems===="
     cd ${script_dir}/tools
     
-	${csudo} ./makepkg.sh    ${compile_dir} ${version} "${build_time}" ${cpuType} ${osType} ${verMode} ${verType}
+	${csudo} ./makepkg.sh    ${compile_dir} ${version} "${build_time}" ${cpuType} ${osType} ${verMode} ${verType} ${pagMode}
-	${csudo} ./makeclient.sh ${compile_dir} ${version} "${build_time}" ${cpuType} ${osType} ${verMode} ${verType}
+	${csudo} ./makeclient.sh ${compile_dir} ${version} "${build_time}" ${cpuType} ${osType} ${verMode} ${verType} ${pagMode}
 else
     cd ${script_dir}/tools
     ./makeclient.sh ${compile_dir} ${version} "${build_time}" ${cpuType} ${osType} ${verMode} ${verType}

packaging/rpm/makerpm.sh

@@ -65,10 +65,10 @@ cp_rpm_package ${pkg_dir}/RPMS
 
 if [ "$verMode" == "cluster" ]; then
   rpmname="TDengine-server-"${tdengine_ver}-${osType}-${cpuType}
-elif [ "$verMode" == "lite" ]; then
+elif [ "$verMode" == "edge" ]; then
   rpmname="TDengine-server"-${tdengine_ver}-${osType}-${cpuType}
 else
-  echo "unknow verMode, nor cluster or lite"
+  echo "unknow verMode, nor cluster or edge"
   exit 1
 fi
 

packaging/tools/install.sh

@@ -6,7 +6,8 @@
 set -e
 #set -x
 
-verMode=lite
+verMode=edge
+pagMode=full
 
 # -----------------------Variables definition---------------------
 script_dir=$(dirname $(readlink -f "$0"))
@@ -479,7 +480,9 @@ function update_TDengine() {
     install_log
     install_header
     install_lib
+    if [ "$pagMode" != "lite" ]; then
       install_connector
+    fi
     install_examples
     if [ -z $1 ]; then
         install_bin
@@ -554,7 +557,9 @@ function install_TDengine() {
     install_log 
     install_header
     install_lib
+    if [ "$pagMode" != "lite" ]; then
       install_connector
+    fi
     install_examples
 
     if [ -z $1 ]; then # install service and client

packaging/tools/install_client.sh

@@ -9,6 +9,7 @@ set -e
 # -----------------------Variables definition---------------------
 
 osType=Linux
+pagMode=full
 
 if [ "$osType" != "Darwin" ]; then
     script_dir=$(dirname $(readlink -f "$0"))
@@ -180,7 +181,9 @@ function update_TDengine() {
     install_log
     install_header
     install_lib
+    if [ "$pagMode" != "lite" ]; then
       install_connector
+    fi
     install_examples
     install_bin
     install_config
@@ -205,7 +208,9 @@ function install_TDengine() {
     install_log
     install_header
     install_lib
+    if [ "$pagMode" != "lite" ]; then
       install_connector
+    fi
     install_examples
     install_bin
     install_config

packaging/tools/makeclient.sh

@@ -12,6 +12,7 @@ cpuType=$4
 osType=$5
 verMode=$6
 verType=$7
+pagMode=$8
 
 if [ "$osType" != "Darwin" ]; then
     script_dir="$(dirname $(readlink -f $0))"
@@ -39,7 +40,13 @@ fi
 # Directories and files.
 
 if [ "$osType" != "Darwin" ]; then
+  if [ "$pagMode" == "lite" ]; then
+    strip ${build_dir}/bin/taosd 
+    strip ${build_dir}/bin/taos
+    bin_files="${build_dir}/bin/taos ${script_dir}/remove_client.sh"
+  else  
     bin_files="${build_dir}/bin/taos ${build_dir}/bin/taosdump ${script_dir}/remove_client.sh"
+  fi
   lib_files="${build_dir}/lib/libtaos.so.${version}"
 else
   bin_files="${build_dir}/bin/taos ${script_dir}/remove_client.sh"
@@ -74,18 +81,23 @@ if [ "$osType" == "Darwin" ]; then
     sed 's/osType=Linux/osType=Darwin/g' ${install_dir}/install_client.sh >> install_client_temp.sh
     mv install_client_temp.sh ${install_dir}/install_client.sh
 fi
+if [ "$pagMode" == "lite" ]; then
+    sed 's/pagMode=full/pagMode=lite/g' ${install_dir}/install_client.sh >> install_client_temp.sh
+    mv install_client_temp.sh ${install_dir}/install_client.sh
+fi
 chmod a+x ${install_dir}/install_client.sh
 
 # Copy example code
 mkdir -p ${install_dir}/examples
 examples_dir="${top_dir}/tests/examples"
 cp -r ${examples_dir}/c      ${install_dir}/examples
-cp -r ${examples_dir}/JDBC   ${install_dir}/examples
+if [[ "$pagMode" != "lite" ]] && [[ "$cpuType" != "aarch32" ]]; then
-cp -r ${examples_dir}/matlab ${install_dir}/examples
+  cp -r ${examples_dir}/JDBC   ${install_dir}/examples
-cp -r ${examples_dir}/python ${install_dir}/examples
+  cp -r ${examples_dir}/matlab ${install_dir}/examples
-cp -r ${examples_dir}/R      ${install_dir}/examples
+  cp -r ${examples_dir}/python ${install_dir}/examples
-cp -r ${examples_dir}/go     ${install_dir}/examples
+  cp -r ${examples_dir}/R      ${install_dir}/examples
-
+  cp -r ${examples_dir}/go     ${install_dir}/examples
+fi
 # Copy driver
 mkdir -p ${install_dir}/driver 
 cp ${lib_files} ${install_dir}/driver
@@ -94,13 +106,14 @@ cp ${lib_files} ${install_dir}/driver
 connector_dir="${code_dir}/connector"
 mkdir -p ${install_dir}/connector
 
-if [ "$osType" != "Darwin" ]; then
+if [[ "$pagMode" != "lite" ]] && [[ "$cpuType" != "aarch32" ]]; then
+  if [ "$osType" != "Darwin" ]; then
     cp ${build_dir}/lib/*.jar      ${install_dir}/connector
+  fi
+  cp -r ${connector_dir}/grafana ${install_dir}/connector/
+  cp -r ${connector_dir}/python  ${install_dir}/connector/
+  cp -r ${connector_dir}/go      ${install_dir}/connector
 fi
-cp -r ${connector_dir}/grafana ${install_dir}/connector/
-cp -r ${connector_dir}/python  ${install_dir}/connector/
-cp -r ${connector_dir}/go      ${install_dir}/connector
-
 # Copy release note
 # cp ${script_dir}/release_note ${install_dir}
 
@@ -110,13 +123,17 @@ cd ${release_dir}
 
 if [ "$verMode" == "cluster" ]; then
   pkg_name=${install_dir}-${version}-${osType}-${cpuType}
-elif [ "$verMode" == "lite" ]; then
+elif [ "$verMode" == "edge" ]; then
   pkg_name=${install_dir}-${version}-${osType}-${cpuType}
 else
-  echo "unknow verMode, nor cluster or lite"
+  echo "unknow verMode, nor cluster or edge"
   exit 1
 fi
 
+if [ "$pagMode" == "lite" ]; then
+  pkg_name=${pkg_name}-Lite
+fi
+
 if [ "$verType" == "beta" ]; then
   pkg_name=${pkg_name}-${verType}
 elif [ "$verType" == "stable" ]; then  

packaging/tools/makepkg.sh

@@ -13,6 +13,7 @@ cpuType=$4
 osType=$5
 verMode=$6
 verType=$7
+pagMode=$8
 
 script_dir="$(dirname $(readlink -f $0))"
 top_dir="$(readlink -f ${script_dir}/../..)"
@@ -30,7 +31,14 @@ else
 fi
 
 # Directories and files.
-bin_files="${build_dir}/bin/taosd ${build_dir}/bin/taos ${build_dir}/bin/taosdemo ${build_dir}/bin/taosdump ${script_dir}/remove.sh"
+if [ "$pagMode" == "lite" ]; then
+  strip ${build_dir}/bin/taosd 
+  strip ${build_dir}/bin/taos
+  bin_files="${build_dir}/bin/taosd ${build_dir}/bin/taos ${script_dir}/remove.sh"
+else 
+  bin_files="${build_dir}/bin/taosd ${build_dir}/bin/taos ${build_dir}/bin/taosdemo ${build_dir}/bin/taosdump ${script_dir}/remove.sh"
+fi
+
 lib_files="${build_dir}/lib/libtaos.so.${version}"
 header_files="${code_dir}/inc/taos.h ${code_dir}/inc/taoserror.h"
 cfg_dir="${top_dir}/packaging/cfg"
@@ -56,6 +64,9 @@ mkdir -p ${install_dir}/init.d && cp ${init_file_deb} ${install_dir}/init.d/taos
 mkdir -p ${install_dir}/init.d && cp ${init_file_rpm} ${install_dir}/init.d/taosd.rpm
 
 if [ "$verMode" == "cluster" ]; then
+    sed 's/verMode=edge/verMode=cluster/g' ${install_dir}/bin/remove.sh >> remove_temp.sh
+    mv remove_temp.sh ${install_dir}/bin/remove.sh
+    
     mkdir -p ${install_dir}/nginxd && cp -r ${nginx_dir}/* ${install_dir}/nginxd
     cp ${nginx_dir}/png/taos.png ${install_dir}/nginxd/admin/images/taos.png
     rm -rf ${install_dir}/nginxd/png
@@ -70,11 +81,20 @@ fi
 
 cd ${install_dir}
 tar -zcv -f taos.tar.gz * --remove-files  || :
+exitcode=$?
+if [ "$exitcode" != "0" ]; then
+    echo "tar taos.tar.gz error !!!"
+    exit $exitcode
+fi
 
 cd ${curr_dir}
 cp ${install_files} ${install_dir}
 if [ "$verMode" == "cluster" ]; then
-    sed 's/verMode=lite/verMode=cluster/g' ${install_dir}/install.sh >> install_temp.sh
+    sed 's/verMode=edge/verMode=cluster/g' ${install_dir}/install.sh >> install_temp.sh
+    mv install_temp.sh ${install_dir}/install.sh
+fi
+if [ "$pagMode" == "lite" ]; then
+    sed 's/pagMode=full/pagMode=lite/g' ${install_dir}/install.sh >> install_temp.sh
     mv install_temp.sh ${install_dir}/install.sh
 fi
 chmod a+x ${install_dir}/install.sh
@@ -82,13 +102,14 @@ chmod a+x ${install_dir}/install.sh
 # Copy example code
 mkdir -p ${install_dir}/examples
 examples_dir="${top_dir}/tests/examples"
-cp -r ${examples_dir}/c      ${install_dir}/examples
+  cp -r ${examples_dir}/c      ${install_dir}/examples
-cp -r ${examples_dir}/JDBC   ${install_dir}/examples
+if [[ "$pagMode" != "lite" ]] && [[ "$cpuType" != "aarch32" ]]; then
-cp -r ${examples_dir}/matlab ${install_dir}/examples
+  cp -r ${examples_dir}/JDBC   ${install_dir}/examples
-cp -r ${examples_dir}/python ${install_dir}/examples
+  cp -r ${examples_dir}/matlab ${install_dir}/examples
-cp -r ${examples_dir}/R      ${install_dir}/examples
+  cp -r ${examples_dir}/python ${install_dir}/examples
-cp -r ${examples_dir}/go     ${install_dir}/examples
+  cp -r ${examples_dir}/R      ${install_dir}/examples
-
+  cp -r ${examples_dir}/go     ${install_dir}/examples
+fi
 # Copy driver
 mkdir -p ${install_dir}/driver 
 cp ${lib_files} ${install_dir}/driver
@@ -96,11 +117,12 @@ cp ${lib_files} ${install_dir}/driver
 # Copy connector
 connector_dir="${code_dir}/connector"
 mkdir -p ${install_dir}/connector
-cp ${build_dir}/lib/*.jar      ${install_dir}/connector
+if [[ "$pagMode" != "lite" ]] && [[ "$cpuType" != "aarch32" ]]; then
-cp -r ${connector_dir}/grafana ${install_dir}/connector/
+  cp ${build_dir}/lib/*.jar      ${install_dir}/connector
-cp -r ${connector_dir}/python  ${install_dir}/connector/
+  cp -r ${connector_dir}/grafana ${install_dir}/connector/
-cp -r ${connector_dir}/go      ${install_dir}/connector
+  cp -r ${connector_dir}/python  ${install_dir}/connector/
-
+  cp -r ${connector_dir}/go      ${install_dir}/connector
+fi
 # Copy release note
 # cp ${script_dir}/release_note ${install_dir}
 
@@ -110,13 +132,17 @@ cd ${release_dir}
 
 if [ "$verMode" == "cluster" ]; then
   pkg_name=${install_dir}-${version}-${osType}-${cpuType}
-elif [ "$verMode" == "lite" ]; then
+elif [ "$verMode" == "edge" ]; then
   pkg_name=${install_dir}-${version}-${osType}-${cpuType}
 else
-  echo "unknow verMode, nor cluster or lite"
+  echo "unknow verMode, nor cluster or edge"
   exit 1
 fi
 
+if [ "$pagMode" == "lite" ]; then
+  pkg_name=${pkg_name}-Lite
+fi
+
 if [ "$verType" == "beta" ]; then
   pkg_name=${pkg_name}-${verType}
 elif [ "$verType" == "stable" ]; then 
@@ -127,5 +153,10 @@ else
 fi
 
 tar -zcv -f "$(basename ${pkg_name}).tar.gz" $(basename ${install_dir}) --remove-files || :
+exitcode=$?
+if [ "$exitcode" != "0" ]; then
+    echo "tar ${pkg_name}.tar.gz error !!!"
+    exit $exitcode
+fi
 
 cd ${curr_dir}

packaging/tools/remove.sh

@@ -5,7 +5,7 @@
 set -e
 #set -x
 
-verMode=lite
+verMode=edge
 
 RED='\033[0;31m'
 GREEN='\033[1;32m'

src/client/inc/tscSecondaryMerge.h

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

src/client/src/tscFunctionImpl.c

@@ -27,6 +27,7 @@
 #include "ttime.h"
 #include "ttypes.h"
 #include "tutil.h"
+#include "tpercentile.h"
 
 #define GET_INPUT_CHAR(x) (((char *)((x)->aInputElemBuf)) + ((x)->startOffset) * ((x)->inputBytes))
 #define GET_INPUT_CHAR_INDEX(x, y) (GET_INPUT_CHAR(x) + (y) * (x)->inputBytes)
@@ -220,7 +221,7 @@ int32_t getResultDataInfo(int32_t dataType, int32_t dataBytes, int32_t functionI
     } else if (functionId == TSDB_FUNC_TWA) {
       *type = TSDB_DATA_TYPE_DOUBLE;
       *bytes = sizeof(STwaInfo);
-      *intermediateResBytes = sizeof(STwaInfo);
+      *intermediateResBytes = *bytes;
       return TSDB_CODE_SUCCESS;
     }
   }
@@ -271,7 +272,6 @@ int32_t getResultDataInfo(int32_t dataType, int32_t dataBytes, int32_t functionI
   } else if (functionId == TSDB_FUNC_PERCT) {
     *type = (int16_t)TSDB_DATA_TYPE_DOUBLE;
     *bytes = (int16_t)sizeof(double);
-    //*intermediateResBytes = POINTER_BYTES;
     *intermediateResBytes = (int16_t)sizeof(double);
   } else if (functionId == TSDB_FUNC_LEASTSQR) {
     *type = TSDB_DATA_TYPE_BINARY;
@@ -1878,6 +1878,7 @@ static void do_top_function_add(STopBotInfo *pInfo, int32_t maxLen, void *pData,
   tVariantCreateFromBinary(&val, pData, tDataTypeDesc[type].nSize, type);
 
   tValuePair **pList = pInfo->res;
+  assert(pList != NULL);
   
   if (pInfo->num < maxLen) {
     if (pInfo->num == 0 ||
@@ -2416,7 +2417,7 @@ static bool percentile_function_setup(SQLFunctionCtx *pCtx) {
   SResultInfo *pResInfo = GET_RES_INFO(pCtx);
   SSchema      field[1] = {{pCtx->inputType, "dummyCol", 0, pCtx->inputBytes}};
 
-  tColModel *pModel = tColModelCreate(field, 1, 1000);
+  SColumnModel *pModel = createColumnModel(field, 1, 1000);
   int32_t    orderIdx = 0;
 
   // tOrderDesc object

src/client/src/tscParseInsert.c

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

src/client/src/tscSQLParser.c

@@ -72,7 +72,7 @@ static void setColumnOffsetValueInResultset(SQueryInfo* pQueryInfo);
 static int32_t parseGroupbyClause(SQueryInfo* pQueryInfo, tVariantList* pList, SSqlCmd* pCmd);
 
 static int32_t parseIntervalClause(SQueryInfo* pQueryInfo, SQuerySQL* pQuerySql);
-static int32_t setSlidingClause(SQueryInfo* pQueryInfo, SQuerySQL* pQuerySql);
+static int32_t parseSlidingClause(SQueryInfo* pQueryInfo, SQuerySQL* pQuerySql);
 
 static int32_t addProjectionExprAndResultField(SQueryInfo* pQueryInfo, tSQLExprItem* pItem);
 
@@ -657,14 +657,14 @@ int32_t parseIntervalClause(SQueryInfo* pQueryInfo, SQuerySQL* pQuerySql) {
     return ret;
   }
   
-  if (setSlidingClause(pQueryInfo, pQuerySql) != TSDB_CODE_SUCCESS) {
+  if (parseSlidingClause(pQueryInfo, pQuerySql) != TSDB_CODE_SUCCESS) {
     return TSDB_CODE_INVALID_SQL;
   }
   
   return TSDB_CODE_SUCCESS;
 }
 
-int32_t setSlidingClause(SQueryInfo* pQueryInfo, SQuerySQL* pQuerySql) {
+int32_t parseSlidingClause(SQueryInfo* pQueryInfo, SQuerySQL* pQuerySql) {
   const char* msg0 = "sliding value too small";
   const char* msg1 = "sliding value no larger than the interval value";
 
@@ -685,7 +685,7 @@ int32_t setSlidingClause(SQueryInfo* pQueryInfo, SQuerySQL* pQuerySql) {
       return invalidSqlErrMsg(pQueryInfo->msg, msg1);
     }
   } else {
-    pSliding->n = pQueryInfo->nAggTimeInterval;
+    pQueryInfo->nSlidingTime = -1;
   }
 
   return TSDB_CODE_SUCCESS;
@@ -5453,8 +5453,6 @@ int32_t doCheckForQuery(SSqlObj* pSql, SQuerySQL* pQuerySql, int32_t index) {
   const char* msg0 = "invalid table name";
   const char* msg1 = "table name too long";
   const char* msg2 = "point interpolation query needs timestamp";
-  const char* msg3 = "sliding value too small";
-  const char* msg4 = "sliding value no larger than the interval value";
   const char* msg5 = "fill only available for interval query";
   const char* msg6 = "start(end) time of query range required or time range too large";
   const char* msg7 = "illegal number of tables in from clause";
@@ -5588,30 +5586,6 @@ int32_t doCheckForQuery(SSqlObj* pSql, SQuerySQL* pQuerySql, int32_t index) {
     return invalidSqlErrMsg(pQueryInfo->msg, msg2);
   }
 
-//  // set sliding value, the query time range needs to be decide in the first place
-//  SSQLToken* pSliding = &pQuerySql->sliding;
-//  if (pSliding->n != 0) {
-//    if (!tscEmbedded && pCmd->inStream == 0 && hasDefaultQueryTimeRange(pQueryInfo)) {  // sliding only allowed in stream
-//      const char* msg = "time range expected for sliding window query";
-//      return invalidSqlErrMsg(tscGetErrorMsgPayload(pCmd), msg);
-//    }
-//
-//    getTimestampInUsFromStr(pSliding->z, pSliding->n, &pQueryInfo->nSlidingTime);
-//    if (pMeterMetaInfo->pMeterMeta->precision == TSDB_TIME_PRECISION_MILLI) {
-//      pQueryInfo->nSlidingTime /= 1000;
-//    }
-//
-//    if (pQueryInfo->nSlidingTime < tsMinSlidingTime) {
-//      return invalidSqlErrMsg(pQueryInfo->msg, msg3);
-//    }
-//
-//    if (pQueryInfo->nSlidingTime > pQueryInfo->nAggTimeInterval) {
-//      return invalidSqlErrMsg(pQueryInfo->msg, msg4);
-//    }
-//  } else {
-//    pQueryInfo->nSlidingTime = -1;
-//  }
-
   // in case of join query, time range is required.
   if (QUERY_IS_JOIN_QUERY(pQueryInfo->type)) {
     int64_t timeRange = labs(pQueryInfo->stime - pQueryInfo->etime);

src/client/src/tscSecondaryMerge.c

@@ -62,16 +62,19 @@ static void tscInitSqlContext(SSqlCmd *pCmd, SSqlRes *pRes, SLocalReducer *pRedu
   for (int32_t i = 0; i < pQueryInfo->fieldsInfo.numOfOutputCols; ++i) {
     SQLFunctionCtx *pCtx = &pReducer->pCtx[i];
 
-    pCtx->aOutputBuf = pReducer->pResultBuf->data + tscFieldInfoGetOffset(pQueryInfo, i) * pReducer->resColModel->maxCapacity;
+    pCtx->aOutputBuf = pReducer->pResultBuf->data + tscFieldInfoGetOffset(pQueryInfo, i) * pReducer->resColModel->capacity;
     pCtx->order = pQueryInfo->order.order;
     pCtx->functionId = pQueryInfo->exprsInfo.pExprs[i].functionId;
 
     // input buffer hold only one point data
-    pCtx->aInputElemBuf = pReducer->pTempBuffer->data + pDesc->pSchema->colOffset[i];
+    int16_t offset = getColumnModelOffset(pDesc->pColumnModel, i);
+    SSchema* pSchema = getColumnModelSchema(pDesc->pColumnModel, i);
+    
+    pCtx->aInputElemBuf = pReducer->pTempBuffer->data + offset;
 
     // input data format comes from pModel
-    pCtx->inputType = pDesc->pSchema->pFields[i].type;
+    pCtx->inputType = pSchema->type;
-    pCtx->inputBytes = pDesc->pSchema->pFields[i].bytes;
+    pCtx->inputBytes = pSchema->bytes;
 
     TAOS_FIELD *pField = tscFieldInfoGetField(pQueryInfo, i);
     // output data format yet comes from pCmd.
@@ -132,11 +135,11 @@ static void tscInitSqlContext(SSqlCmd *pCmd, SSqlRes *pRes, SLocalReducer *pRedu
  * todo release allocated memory process with async process
  */
 void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrderDescriptor *pDesc,
-                           tColModel *finalmodel, SSqlCmd *pCmd, SSqlRes *pRes) {
+                           SColumnModel *finalmodel, SSqlCmd *pCmd, SSqlRes *pRes) {
   // offset of cmd in SSqlObj structure
   char *pSqlObjAddr = (char *)pCmd - offsetof(SSqlObj, cmd);
 
-  if (pMemBuffer == NULL || pDesc->pSchema == NULL) {
+  if (pMemBuffer == NULL || pDesc->pColumnModel == NULL) {
     tscLocalReducerEnvDestroy(pMemBuffer, pDesc, finalmodel, numOfBuffer);
 
     tscError("%p no local buffer or intermediate result format model", pSqlObjAddr);
@@ -162,9 +165,9 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
     return;
   }
 
-  if (pDesc->pSchema->maxCapacity >= pMemBuffer[0]->nPageSize) {
+  if (pDesc->pColumnModel->capacity >= pMemBuffer[0]->pageSize) {
-    tscError("%p Invalid value of buffer capacity %d and page size %d ", pSqlObjAddr, pDesc->pSchema->maxCapacity,
+    tscError("%p Invalid value of buffer capacity %d and page size %d ", pSqlObjAddr, pDesc->pColumnModel->capacity,
-             pMemBuffer[0]->nPageSize);
+             pMemBuffer[0]->pageSize);
 
     tscLocalReducerEnvDestroy(pMemBuffer, pDesc, finalmodel, numOfBuffer);
     pRes->code = TSDB_CODE_APP_ERROR;
@@ -196,7 +199,7 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
     int32_t numOfFlushoutInFile = pMemBuffer[i]->fileMeta.flushoutData.nLength;
 
     for (int32_t j = 0; j < numOfFlushoutInFile; ++j) {
-      SLocalDataSource *pDS = (SLocalDataSource *)malloc(sizeof(SLocalDataSource) + pMemBuffer[0]->nPageSize);
+      SLocalDataSource *pDS = (SLocalDataSource *)malloc(sizeof(SLocalDataSource) + pMemBuffer[0]->pageSize);
       if (pDS == NULL) {
         tscError("%p failed to create merge structure", pSqlObjAddr);
         pRes->code = TSDB_CODE_CLI_OUT_OF_MEMORY;
@@ -219,7 +222,7 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
 
       tscGetSrcColumnInfo(colInfo, pQueryInfo);
 
-      tColModelDisplayEx(pDesc->pSchema, pDS->filePage.data, pDS->filePage.numOfElems, pMemBuffer[0]->numOfElemsPerPage,
+      tColModelDisplayEx(pDesc->pColumnModel, pDS->filePage.data, pDS->filePage.numOfElems, pMemBuffer[0]->numOfElemsPerPage,
                          colInfo);
 #endif
       if (pDS->filePage.numOfElems == 0) {  // no data in this flush
@@ -259,7 +262,7 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
   tscRestoreSQLFunctionForMetricQuery(pQueryInfo);
   tscFieldInfoCalOffset(pQueryInfo);
 
-  if (pReducer->rowSize > pMemBuffer[0]->nPageSize) {
+  if (pReducer->rowSize > pMemBuffer[0]->pageSize) {
     assert(false);  // todo fixed row size is larger than the minimum page size;
   }
 
@@ -274,15 +277,15 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
   pReducer->discardData = (tFilePage *)calloc(1, pReducer->rowSize + sizeof(tFilePage));
   pReducer->discard = false;
 
-  pReducer->nResultBufSize = pMemBuffer[0]->nPageSize * 16;
+  pReducer->nResultBufSize = pMemBuffer[0]->pageSize * 16;
   pReducer->pResultBuf = (tFilePage *)calloc(1, pReducer->nResultBufSize + sizeof(tFilePage));
   
   int32_t finalRowLength = tscGetResRowLength(pQueryInfo);
   pReducer->resColModel = finalmodel;
-  pReducer->resColModel->maxCapacity = pReducer->nResultBufSize / finalRowLength;
+  pReducer->resColModel->capacity = pReducer->nResultBufSize / finalRowLength;
   assert(finalRowLength <= pReducer->rowSize);
 
-  pReducer->pFinalRes = calloc(1, pReducer->rowSize * pReducer->resColModel->maxCapacity);
+  pReducer->pFinalRes = calloc(1, pReducer->rowSize * pReducer->resColModel->capacity);
   pReducer->pBufForInterpo = calloc(1, pReducer->nResultBufSize);
 
   if (pReducer->pTempBuffer == NULL|| pReducer->discardData == NULL  || pReducer->pResultBuf == NULL ||
@@ -304,8 +307,8 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
   tscCreateResPointerInfo(pRes, pQueryInfo);
   tscInitSqlContext(pCmd, pRes, pReducer, pDesc);
 
-  // we change the maxCapacity of schema to denote that there is only one row in temp buffer
+  // we change the capacity of schema to denote that there is only one row in temp buffer
-  pReducer->pDesc->pSchema->maxCapacity = 1;
+  pReducer->pDesc->pColumnModel->capacity = 1;
   
   //restore the limitation value at the last stage
   if (tscOrderedProjectionQueryOnSTable(pQueryInfo, 0)) {
@@ -333,7 +336,8 @@ void tscCreateLocalReducer(tExtMemBuffer **pMemBuffer, int32_t numOfBuffer, tOrd
   if (pQueryInfo->groupbyExpr.numOfGroupCols > 0) {
     pInterpoInfo->pTags[0] = (char *)pInterpoInfo->pTags + POINTER_BYTES * pQueryInfo->groupbyExpr.numOfGroupCols;
     for (int32_t i = 1; i < pQueryInfo->groupbyExpr.numOfGroupCols; ++i) {
-      pInterpoInfo->pTags[i] = pReducer->resColModel->pFields[startIndex + i - 1].bytes + pInterpoInfo->pTags[i - 1];
+      SSchema* pSchema = getColumnModelSchema(pReducer->resColModel, startIndex + i - 1);
+      pInterpoInfo->pTags[i] = pSchema->bytes + pInterpoInfo->pTags[i - 1];
     }
   } else {
     assert(pInterpoInfo->pTags == NULL);
@@ -346,16 +350,16 @@ static int32_t tscFlushTmpBufferImpl(tExtMemBuffer *pMemoryBuf, tOrderDescriptor
     return 0;
   }
 
-  assert(pPage->numOfElems <= pDesc->pSchema->maxCapacity);
+  assert(pPage->numOfElems <= pDesc->pColumnModel->capacity);
 
   // sort before flush to disk, the data must be consecutively put on tFilePage.
-  if (pDesc->orderIdx.numOfOrderedCols > 0) {
+  if (pDesc->orderIdx.numOfCols > 0) {
     tColDataQSort(pDesc, pPage->numOfElems, 0, pPage->numOfElems - 1, pPage->data, orderType);
   }
 
 #ifdef _DEBUG_VIEW
   printf("%" PRIu64 " rows data flushed to disk after been sorted:\n", pPage->numOfElems);
-  tColModelDisplay(pDesc->pSchema, pPage->data, pPage->numOfElems, pPage->numOfElems);
+  tColModelDisplay(pDesc->pColumnModel, pPage->data, pPage->numOfElems, pPage->numOfElems);
 #endif
 
   // write to cache after being sorted
@@ -383,18 +387,19 @@ int32_t tscFlushTmpBuffer(tExtMemBuffer *pMemoryBuf, tOrderDescriptor *pDesc, tF
 
 int32_t saveToBuffer(tExtMemBuffer *pMemoryBuf, tOrderDescriptor *pDesc, tFilePage *pPage, void *data,
                      int32_t numOfRows, int32_t orderType) {
-  if (pPage->numOfElems + numOfRows <= pDesc->pSchema->maxCapacity) {
+  SColumnModel *pModel = pDesc->pColumnModel;
-    tColModelAppend(pDesc->pSchema, pPage, data, 0, numOfRows, numOfRows);
+  
+  if (pPage->numOfElems + numOfRows <= pModel->capacity) {
+    tColModelAppend(pModel, pPage, data, 0, numOfRows, numOfRows);
     return 0;
   }
 
-  tColModel *pModel = pDesc->pSchema;
+  // current buffer is overflow, flush data to extensive buffer
-
+  int32_t numOfRemainEntries = pModel->capacity - pPage->numOfElems;
-  int32_t numOfRemainEntries = pDesc->pSchema->maxCapacity - pPage->numOfElems;
   tColModelAppend(pModel, pPage, data, 0, numOfRemainEntries, numOfRows);
 
-  /* current buffer is full, need to flushed to disk */
+  // current buffer is full, need to flushed to disk
-  assert(pPage->numOfElems == pDesc->pSchema->maxCapacity);
+  assert(pPage->numOfElems == pModel->capacity);
   int32_t ret = tscFlushTmpBuffer(pMemoryBuf, pDesc, pPage, orderType);
   if (ret != 0) {
     return -1;
@@ -404,15 +409,15 @@ int32_t saveToBuffer(tExtMemBuffer *pMemoryBuf, tOrderDescriptor *pDesc, tFilePa
 
   while (remain > 0) {
     int32_t numOfWriteElems = 0;
-    if (remain > pModel->maxCapacity) {
+    if (remain > pModel->capacity) {
-      numOfWriteElems = pModel->maxCapacity;
+      numOfWriteElems = pModel->capacity;
     } else {
       numOfWriteElems = remain;
     }
 
     tColModelAppend(pModel, pPage, data, numOfRows - remain, numOfWriteElems, numOfRows);
 
-    if (pPage->numOfElems == pModel->maxCapacity) {
+    if (pPage->numOfElems == pModel->capacity) {
       int32_t ret = tscFlushTmpBuffer(pMemoryBuf, pDesc, pPage, orderType);
       if (ret != 0) {
         return -1;
@@ -508,7 +513,7 @@ void tscDestroyLocalReducer(SSqlObj *pSql) {
   tscTrace("%p free local reducer finished", pSql);
 }
 
-static int32_t createOrderDescriptor(tOrderDescriptor **pOrderDesc, SSqlCmd *pCmd, tColModel *pModel) {
+static int32_t createOrderDescriptor(tOrderDescriptor **pOrderDesc, SSqlCmd *pCmd, SColumnModel *pModel) {
   int32_t numOfGroupByCols = 0;
   SQueryInfo* pQueryInfo = tscGetQueryInfoDetail(pCmd, pCmd->clauseIndex);
   
@@ -567,7 +572,7 @@ bool isSameGroup(SSqlCmd *pCmd, SLocalReducer *pReducer, char *pPrev, tFilePage
   }
 
   tOrderDescriptor *pOrderDesc = pReducer->pDesc;
-  int32_t           numOfCols = pOrderDesc->orderIdx.numOfOrderedCols;
+  int32_t           numOfCols = pOrderDesc->orderIdx.numOfCols;
 
   // no group by columns, all data belongs to one group
   if (numOfCols <= 0) {
@@ -577,25 +582,25 @@ bool isSameGroup(SSqlCmd *pCmd, SLocalReducer *pReducer, char *pPrev, tFilePage
   if (pOrderDesc->orderIdx.pData[numOfCols - 1] == PRIMARYKEY_TIMESTAMP_COL_INDEX) {  //<= 0
     // super table interval query
     assert(pQueryInfo->nAggTimeInterval > 0);
-    pOrderDesc->orderIdx.numOfOrderedCols -= 1;
+    pOrderDesc->orderIdx.numOfCols -= 1;
   } else {  // simple group by query
     assert(pQueryInfo->nAggTimeInterval == 0);
   }
 
   // only one row exists
   int32_t ret = compare_a(pOrderDesc, 1, 0, pPrev, 1, 0, tmpBuffer->data);
-  pOrderDesc->orderIdx.numOfOrderedCols = numOfCols;
+  pOrderDesc->orderIdx.numOfCols = numOfCols;
 
   return (ret == 0);
 }
 
 int32_t tscLocalReducerEnvCreate(SSqlObj *pSql, tExtMemBuffer ***pMemBuffer, tOrderDescriptor **pOrderDesc,
-                                 tColModel **pFinalModel, uint32_t nBufferSizes) {
+                                 SColumnModel **pFinalModel, uint32_t nBufferSizes) {
   SSqlCmd *pCmd = &pSql->cmd;
   SSqlRes *pRes = &pSql->res;
 
   SSchema *  pSchema = NULL;
-  tColModel *pModel = NULL;
+  SColumnModel *pModel = NULL;
   *pFinalModel = NULL;
 
   SQueryInfo* pQueryInfo = tscGetQueryInfoDetail(pCmd, pCmd->clauseIndex);
@@ -630,14 +635,10 @@ int32_t tscLocalReducerEnvCreate(SSqlObj *pSql, tExtMemBuffer ***pMemBuffer, tOr
     capacity = nBufferSizes / rlen;
   }
   
-  pModel = tColModelCreate(pSchema, pQueryInfo->fieldsInfo.numOfOutputCols, capacity);
+  pModel = createColumnModel(pSchema, pQueryInfo->fieldsInfo.numOfOutputCols, capacity);
 
   for (int32_t i = 0; i < pMeterMetaInfo->pMetricMeta->numOfVnodes; ++i) {
-    char tmpPath[512] = {0};
+    (*pMemBuffer)[i] = createExtMemBuffer(nBufferSizes, rlen, pModel);
-    getTmpfilePath("tv_bf_db", tmpPath);
-    tscTrace("%p create [%d](%d) tmp file for subquery:%s", pSql, pMeterMetaInfo->pMetricMeta->numOfVnodes, i, tmpPath);
-
-    tExtMemBufferCreate(&(*pMemBuffer)[i], nBufferSizes, rlen, tmpPath, pModel);
     (*pMemBuffer)[i]->flushModel = MULTIPLE_APPEND_MODEL;
   }
 
@@ -655,7 +656,7 @@ int32_t tscLocalReducerEnvCreate(SSqlObj *pSql, tExtMemBuffer ***pMemBuffer, tOr
     strcpy(pSchema[i].name, pField->name);
   }
 
-  *pFinalModel = tColModelCreate(pSchema, pQueryInfo->fieldsInfo.numOfOutputCols, capacity);
+  *pFinalModel = createColumnModel(pSchema, pQueryInfo->fieldsInfo.numOfOutputCols, capacity);
   tfree(pSchema);
 
   return TSDB_CODE_SUCCESS;
@@ -667,12 +668,12 @@ int32_t tscLocalReducerEnvCreate(SSqlObj *pSql, tExtMemBuffer ***pMemBuffer, tOr
  * @param pFinalModel
  * @param numOfVnodes
  */
-void tscLocalReducerEnvDestroy(tExtMemBuffer **pMemBuffer, tOrderDescriptor *pDesc, tColModel *pFinalModel,
+void tscLocalReducerEnvDestroy(tExtMemBuffer **pMemBuffer, tOrderDescriptor *pDesc, SColumnModel *pFinalModel,
                                int32_t numOfVnodes) {
-  tColModelDestroy(pFinalModel);
+  destroyColumnModel(pFinalModel);
   tOrderDescDestroy(pDesc);
   for (int32_t i = 0; i < numOfVnodes; ++i) {
-    tExtMemBufferDestroy(&pMemBuffer[i]);
+    pMemBuffer[i] = destoryExtMemBuffer(pMemBuffer[i]);
   }
 
   tfree(pMemBuffer);
@@ -697,8 +698,8 @@ int32_t loadNewDataFromDiskFor(SLocalReducer *pLocalReducer, SLocalDataSource *p
 
 #if defined(_DEBUG_VIEW)
     printf("new page load to buffer\n");
-    tColModelDisplay(pOneInterDataSrc->pMemBuffer->pColModel, pOneInterDataSrc->filePage.data,
+    tColModelDisplay(pOneInterDataSrc->pMemBuffer->pColumnModel, pOneInterDataSrc->filePage.data,
-                     pOneInterDataSrc->filePage.numOfElems, pOneInterDataSrc->pMemBuffer->pColModel->maxCapacity);
+                     pOneInterDataSrc->filePage.numOfElems, pOneInterDataSrc->pMemBuffer->pColumnModel->capacity);
 #endif
     *needAdjustLoserTree = true;
   } else {
@@ -759,7 +760,7 @@ void savePrevRecordAndSetupInterpoInfo(SLocalReducer *pLocalReducer, SQueryInfo*
   pLocalReducer->discard = true;
   pLocalReducer->discardData->numOfElems = 0;
 
-  tColModel *pModel = pLocalReducer->pDesc->pSchema;
+  SColumnModel *pModel = pLocalReducer->pDesc->pColumnModel;
   tColModelAppend(pModel, pLocalReducer->discardData, pLocalReducer->prevRowOfInput, 0, 1, 1);
 }
 
@@ -782,11 +783,12 @@ static void reversedCopyResultToDstBuf(SQueryInfo* pQueryInfo, SSqlRes *pRes, tF
 }
 
 static void reversedCopyFromInterpolationToDstBuf(SQueryInfo* pQueryInfo, SSqlRes *pRes, tFilePage **pResPages, SLocalReducer *pLocalReducer) {
+  assert(0);
   for (int32_t i = 0; i < pQueryInfo->exprsInfo.numOfExprs; ++i) {
     TAOS_FIELD *pField = tscFieldInfoGetField(pQueryInfo, i);
 
     int32_t offset = tscFieldInfoGetOffset(pQueryInfo, i);
-    assert(offset == pLocalReducer->resColModel->colOffset[i]);
+    assert(offset == getColumnModelOffset(pLocalReducer->resColModel, i));
 
     char *src = pResPages[i]->data + (pRes->numOfRows - 1) * pField->bytes;
     char *dst = pRes->data + pRes->numOfRows * offset;
@@ -880,7 +882,7 @@ static void doInterpolateResult(SSqlObj *pSql, SLocalReducer *pLocalReducer, boo
   tFilePage **pResPages = malloc(POINTER_BYTES * pQueryInfo->fieldsInfo.numOfOutputCols);
   for (int32_t i = 0; i < pQueryInfo->fieldsInfo.numOfOutputCols; ++i) {
     TAOS_FIELD *pField = tscFieldInfoGetField(pQueryInfo, i);
-    pResPages[i] = calloc(1, sizeof(tFilePage) + pField->bytes * pLocalReducer->resColModel->maxCapacity);
+    pResPages[i] = calloc(1, sizeof(tFilePage) + pField->bytes * pLocalReducer->resColModel->capacity);
   }
 
   char **  srcData = (char **)malloc((POINTER_BYTES + sizeof(int32_t)) * pQueryInfo->fieldsInfo.numOfOutputCols);
@@ -899,11 +901,11 @@ static void doInterpolateResult(SSqlObj *pSql, SLocalReducer *pLocalReducer, boo
     TSKEY etime = taosGetRevisedEndKey(actualETime, pQueryInfo->order.order, pQueryInfo->nAggTimeInterval, pQueryInfo->intervalTimeUnit,
                                        precision);
     int32_t nrows = taosGetNumOfResultWithInterpo(pInterpoInfo, pPrimaryKeys, remains, pQueryInfo->nAggTimeInterval, etime,
-                                                  pLocalReducer->resColModel->maxCapacity);
+                                                  pLocalReducer->resColModel->capacity);
 
     int32_t newRows = taosDoInterpoResult(pInterpoInfo, pQueryInfo->interpoType, pResPages, remains, nrows,
                                           pQueryInfo->nAggTimeInterval, pPrimaryKeys, pLocalReducer->resColModel, srcData,
-                                          pQueryInfo->defaultVal, functions, pLocalReducer->resColModel->maxCapacity);
+                                          pQueryInfo->defaultVal, functions, pLocalReducer->resColModel->capacity);
     assert(newRows <= nrows);
 
     if (pQueryInfo->limit.offset < newRows) {
@@ -960,11 +962,10 @@ static void doInterpolateResult(SSqlObj *pSql, SLocalReducer *pLocalReducer, boo
     if (pQueryInfo->order.order == TSQL_SO_ASC) {
       for (int32_t i = 0; i < pQueryInfo->fieldsInfo.numOfOutputCols; ++i) {
         TAOS_FIELD *pField = tscFieldInfoGetField(pQueryInfo, i);
-
+        int16_t offset = getColumnModelOffset(pLocalReducer->resColModel, i);
-        memcpy(pRes->data + pLocalReducer->resColModel->colOffset[i] * pRes->numOfRows, pResPages[i]->data,
+        memcpy(pRes->data + offset * pRes->numOfRows, pResPages[i]->data, pField->bytes * pRes->numOfRows);
-               pField->bytes * pRes->numOfRows);
       }
-    } else {
+    } else {//todo bug??
       reversedCopyFromInterpolationToDstBuf(pQueryInfo, pRes, pResPages, pLocalReducer);
     }
   }
@@ -979,13 +980,15 @@ static void doInterpolateResult(SSqlObj *pSql, SLocalReducer *pLocalReducer, boo
 }
 
 static void savePreviousRow(SLocalReducer *pLocalReducer, tFilePage *tmpBuffer) {
-  tColModel *pColModel = pLocalReducer->pDesc->pSchema;
+  SColumnModel *pColumnModel = pLocalReducer->pDesc->pColumnModel;
-  assert(pColModel->maxCapacity == 1 && tmpBuffer->numOfElems == 1);
+  assert(pColumnModel->capacity == 1 && tmpBuffer->numOfElems == 1);
 
   // copy to previous temp buffer
-  for (int32_t i = 0; i < pLocalReducer->pDesc->pSchema->numOfCols; ++i) {
+  for (int32_t i = 0; i < pColumnModel->numOfCols; ++i) {
-    memcpy(pLocalReducer->prevRowOfInput + pColModel->colOffset[i], tmpBuffer->data + pColModel->colOffset[i],
+    SSchema* pSchema = getColumnModelSchema(pColumnModel, i);
-           pColModel->pFields[i].bytes);
+    int16_t offset = getColumnModelOffset(pColumnModel, i);
+    
+    memcpy(pLocalReducer->prevRowOfInput + offset, tmpBuffer->data + offset, pSchema->bytes);
   }
 
   tmpBuffer->numOfElems = 0;
@@ -1127,7 +1130,7 @@ bool needToMerge(SQueryInfo* pQueryInfo, SLocalReducer *pLocalReducer, tFilePage
     ret = 1;                                                            // disable merge procedure
   } else {
     tOrderDescriptor *pDesc = pLocalReducer->pDesc;
-    if (pDesc->orderIdx.numOfOrderedCols > 0) {
+    if (pDesc->orderIdx.numOfCols > 0) {
       if (pDesc->tsOrder == TSQL_SO_ASC) {  // asc
         // todo refactor comparator
         ret = compare_a(pLocalReducer->pDesc, 1, 0, pLocalReducer->prevRowOfInput, 1, 0, tmpBuffer->data);
@@ -1177,7 +1180,7 @@ bool doGenerateFinalResults(SSqlObj *pSql, SLocalReducer *pLocalReducer, bool no
   
   SQueryInfo* pQueryInfo = tscGetQueryInfoDetail(pCmd, pCmd->clauseIndex);
   tFilePage *pResBuf = pLocalReducer->pResultBuf;
-  tColModel *pModel = pLocalReducer->resColModel;
+  SColumnModel *pModel = pLocalReducer->resColModel;
 
   pRes->code = TSDB_CODE_SUCCESS;
 
@@ -1192,7 +1195,7 @@ bool doGenerateFinalResults(SSqlObj *pSql, SLocalReducer *pLocalReducer, bool no
     return false;
   }
 
-  tColModelCompact(pModel, pResBuf, pModel->maxCapacity);
+  tColModelCompact(pModel, pResBuf, pModel->capacity);
   memcpy(pLocalReducer->pBufForInterpo, pResBuf->data, pLocalReducer->nResultBufSize);
 
 #ifdef _DEBUG_VIEW
@@ -1204,9 +1207,11 @@ bool doGenerateFinalResults(SSqlObj *pSql, SLocalReducer *pLocalReducer, bool no
   int32_t             startIndex = pQueryInfo->fieldsInfo.numOfOutputCols - pQueryInfo->groupbyExpr.numOfGroupCols;
 
   for (int32_t i = 0; i < pQueryInfo->groupbyExpr.numOfGroupCols; ++i) {
+    int16_t offset = getColumnModelOffset(pModel, startIndex + i);
+    SSchema* pSchema = getColumnModelSchema(pModel, startIndex + i);
+    
     memcpy(pInterpoInfo->pTags[i],
-           pLocalReducer->pBufForInterpo + pModel->colOffset[startIndex + i] * pResBuf->numOfElems,
+           pLocalReducer->pBufForInterpo + offset * pResBuf->numOfElems, pSchema->bytes);
-           pModel->pFields[startIndex + i].bytes);
   }
 
   taosInterpoSetStartInfo(&pLocalReducer->interpolationInfo, pResBuf->numOfElems, pQueryInfo->interpoType);
@@ -1218,7 +1223,7 @@ bool doGenerateFinalResults(SSqlObj *pSql, SLocalReducer *pLocalReducer, bool no
 void resetOutputBuf(SQueryInfo* pQueryInfo, SLocalReducer *pLocalReducer) {  // reset output buffer to the beginning
   for (int32_t i = 0; i < pQueryInfo->fieldsInfo.numOfOutputCols; ++i) {
     pLocalReducer->pCtx[i].aOutputBuf =
-        pLocalReducer->pResultBuf->data + tscFieldInfoGetOffset(pQueryInfo, i) * pLocalReducer->resColModel->maxCapacity;
+        pLocalReducer->pResultBuf->data + tscFieldInfoGetOffset(pQueryInfo, i) * pLocalReducer->resColModel->capacity;
   }
 
   memset(pLocalReducer->pResultBuf, 0, pLocalReducer->nResultBufSize + sizeof(tFilePage));
@@ -1270,7 +1275,7 @@ static bool doInterpolationForCurrentGroup(SSqlObj *pSql) {
     int32_t remain = taosNumOfRemainPoints(pInterpoInfo);
     TSKEY   ekey = taosGetRevisedEndKey(etime, pQueryInfo->order.order, pQueryInfo->nAggTimeInterval, pQueryInfo->intervalTimeUnit, p);
     int32_t rows = taosGetNumOfResultWithInterpo(pInterpoInfo, (TSKEY *)pLocalReducer->pBufForInterpo, remain,
-                                                 pQueryInfo->nAggTimeInterval, ekey, pLocalReducer->resColModel->maxCapacity);
+                                                 pQueryInfo->nAggTimeInterval, ekey, pLocalReducer->resColModel->capacity);
     if (rows > 0) {  // do interpo
       doInterpolateResult(pSql, pLocalReducer, false);
     }
@@ -1302,7 +1307,7 @@ static bool doHandleLastRemainData(SSqlObj *pSql) {
 
       etime = taosGetRevisedEndKey(etime, pQueryInfo->order.order, pQueryInfo->nAggTimeInterval, pQueryInfo->intervalTimeUnit, precision);
       int32_t rows = taosGetNumOfResultWithInterpo(pInterpoInfo, NULL, 0, pQueryInfo->nAggTimeInterval, etime,
-                                                   pLocalReducer->resColModel->maxCapacity);
+                                                   pLocalReducer->resColModel->capacity);
       if (rows > 0) {  // do interpo
         doInterpolateResult(pSql, pLocalReducer, true);
       }
@@ -1391,7 +1396,7 @@ int32_t tscDoLocalreduce(SSqlObj *pSql) {
 
   // clear buffer
   handleUnprocessedRow(pCmd, pLocalReducer, tmpBuffer);
-  tColModel *pModel = pLocalReducer->pDesc->pSchema;
+  SColumnModel *pModel = pLocalReducer->pDesc->pColumnModel;
 
   while (1) {
     if (isAllSourcesCompleted(pLocalReducer)) {
@@ -1408,14 +1413,14 @@ int32_t tscDoLocalreduce(SSqlObj *pSql) {
     SLocalDataSource *pOneDataSrc = pLocalReducer->pLocalDataSrc[pTree->pNode[0].index];
 
     tColModelAppend(pModel, tmpBuffer, pOneDataSrc->filePage.data, pOneDataSrc->rowIdx, 1,
-                    pOneDataSrc->pMemBuffer->pColModel->maxCapacity);
+                    pOneDataSrc->pMemBuffer->pColumnModel->capacity);
 
 #if defined(_DEBUG_VIEW)
     printf("chosen row:\t");
     SSrcColumnInfo colInfo[256] = {0};
     tscGetSrcColumnInfo(colInfo, pQueryInfo);
 
-    tColModelDisplayEx(pModel, tmpBuffer->data, tmpBuffer->numOfElems, pModel->maxCapacity, colInfo);
+    tColModelDisplayEx(pModel, tmpBuffer->data, tmpBuffer->numOfElems, pModel->capacity, colInfo);
 #endif
 
     if (pLocalReducer->discard) {
@@ -1470,7 +1475,7 @@ int32_t tscDoLocalreduce(SSqlObj *pSql) {
          * continue to process results instead of return results.
          */
         if ((!sameGroup && pResBuf->numOfElems > 0) ||
-            (pResBuf->numOfElems == pLocalReducer->resColModel->maxCapacity)) {
+            (pResBuf->numOfElems == pLocalReducer->resColModel->capacity)) {
           // does not belong to the same group
           bool notSkipped = doGenerateFinalResults(pSql, pLocalReducer, !sameGroup);
 

src/client/src/tscServer.c

@@ -901,7 +901,7 @@ int tscLaunchSTableSubqueries(SSqlObj *pSql) {
 
   tExtMemBuffer **  pMemoryBuf = NULL;
   tOrderDescriptor *pDesc = NULL;
-  tColModel *       pModel = NULL;
+  SColumnModel *       pModel = NULL;
 
   pRes->qhandle = 1;  // hack the qhandle check
 
@@ -1181,7 +1181,7 @@ void tscRetrieveFromVnodeCallBack(void *param, TAOS_RES *tres, int numOfRows) {
     tscTrace("%p sub:%p retrieve numOfRows:%d totalNumOfRows:%d from ip:%u,vid:%d,orderOfSub:%d", pPObj, pSql,
              pRes->numOfRows, pState->numOfRetrievedRows, pSvd->ip, pSvd->vnode, idx);
     
-    if (num > tsMaxNumOfOrderedResults) {
+    if (num > tsMaxNumOfOrderedResults && tscIsProjectionQueryOnSTable(pQueryInfo, 0)) {
       tscError("%p sub:%p num of OrderedRes is too many, max allowed:%" PRId64 " , current:%" PRId64,
           pPObj, pSql, tsMaxNumOfOrderedResults, num);
       tscAbortFurtherRetryRetrieval(trsupport, tres, TSDB_CODE_SORTED_RES_TOO_MANY);
@@ -1194,7 +1194,7 @@ void tscRetrieveFromVnodeCallBack(void *param, TAOS_RES *tres, int numOfRows) {
     SSrcColumnInfo colInfo[256] = {0};
 
     tscGetSrcColumnInfo(colInfo, pQueryInfo);
-    tColModelDisplayEx(pDesc->pSchema, pRes->data, pRes->numOfRows, pRes->numOfRows, colInfo);
+    tColModelDisplayEx(pDesc->pColumnModel, pRes->data, pRes->numOfRows, pRes->numOfRows, colInfo);
 #endif
     if (tsTotalTmpDirGB != 0 && tsAvailTmpDirGB < tsMinimalTmpDirGB) {
       tscError("%p sub:%p client disk space remain %.3f GB, need at least %.3f GB, stop query", pPObj, pSql,
@@ -1202,6 +1202,7 @@ void tscRetrieveFromVnodeCallBack(void *param, TAOS_RES *tres, int numOfRows) {
       tscAbortFurtherRetryRetrieval(trsupport, tres, TSDB_CODE_CLI_NO_DISKSPACE);
       return;
     }
+    
     int32_t ret = saveToBuffer(trsupport->pExtMemBuffer[idx], pDesc, trsupport->localBuffer, pRes->data,
                                pRes->numOfRows, pQueryInfo->groupbyExpr.orderType);
     if (ret < 0) {
@@ -1214,17 +1215,17 @@ void tscRetrieveFromVnodeCallBack(void *param, TAOS_RES *tres, int numOfRows) {
 
   } else {  // all data has been retrieved to client
     /* data in from current vnode is stored in cache and disk */
-    uint32_t numOfRowsFromVnode = trsupport->pExtMemBuffer[idx]->numOfAllElems + trsupport->localBuffer->numOfElems;
+    uint32_t numOfRowsFromVnode = trsupport->pExtMemBuffer[idx]->numOfTotalElems + trsupport->localBuffer->numOfElems;
     tscTrace("%p sub:%p all data retrieved from ip:%u,vid:%d, numOfRows:%d, orderOfSub:%d", pPObj, pSql, pSvd->ip,
              pSvd->vnode, numOfRowsFromVnode, idx);
 
-    tColModelCompact(pDesc->pSchema, trsupport->localBuffer, pDesc->pSchema->maxCapacity);
+    tColModelCompact(pDesc->pColumnModel, trsupport->localBuffer, pDesc->pColumnModel->capacity);
 
 #ifdef _DEBUG_VIEW
     printf("%" PRIu64 " rows data flushed to disk:\n", trsupport->localBuffer->numOfElems);
     SSrcColumnInfo colInfo[256] = {0};
     tscGetSrcColumnInfo(colInfo, pQueryInfo);
-    tColModelDisplayEx(pDesc->pSchema, trsupport->localBuffer->data, trsupport->localBuffer->numOfElems,
+    tColModelDisplayEx(pDesc->pColumnModel, trsupport->localBuffer->data, trsupport->localBuffer->numOfElems,
                        trsupport->localBuffer->numOfElems, colInfo);
 #endif
     
@@ -1256,7 +1257,7 @@ void tscRetrieveFromVnodeCallBack(void *param, TAOS_RES *tres, int numOfRows) {
     }
 
     // all sub-queries are returned, start to local merge process
-    pDesc->pSchema->maxCapacity = trsupport->pExtMemBuffer[idx]->numOfElemsPerPage;
+    pDesc->pColumnModel->capacity = trsupport->pExtMemBuffer[idx]->numOfElemsPerPage;
 
     tscTrace("%p retrieve from %d vnodes completed.final NumOfRows:%d,start to build loser tree", pPObj,
              pState->numOfTotal, pState->numOfRetrievedRows);
@@ -1516,7 +1517,7 @@ void tscUpdateVnodeInQueryMsg(SSqlObj *pSql, char *buf) {
   char *          pStart = buf + tsRpcHeadSize;
   SQueryMeterMsg *pQueryMsg = (SQueryMeterMsg *)pStart;
 
-  if (UTIL_METER_IS_NOMRAL_METER(pMeterMetaInfo)) {  // pSchema == NULL, query on meter
+  if (UTIL_METER_IS_NOMRAL_METER(pMeterMetaInfo)) {  // pColumnModel == NULL, query on meter
     SMeterMeta *pMeterMeta = pMeterMetaInfo->pMeterMeta;
     pQueryMsg->vnode = htons(pMeterMeta->vpeerDesc[pSql->index].vnode);
   } else {  // query on metric

src/client/src/tscStream.c

@@ -396,7 +396,9 @@ static void tscSetSlidingWindowInfo(SSqlObj *pSql, SSqlStream *pStream) {
   int64_t minSlidingTime =
       (pStream->precision == TSDB_TIME_PRECISION_MICRO) ? tsMinSlidingTime * 1000L : tsMinSlidingTime;
 
-  if (pQueryInfo->nSlidingTime < minSlidingTime) {
+  if (pQueryInfo->nSlidingTime == -1) {
+    pQueryInfo->nSlidingTime = pQueryInfo->nAggTimeInterval;
+  } else if (pQueryInfo->nSlidingTime < minSlidingTime) {
     tscWarn("%p stream:%p, original sliding value:%" PRId64 " too small, reset to:%" PRId64 "", pSql, pStream,
         pQueryInfo->nSlidingTime, minSlidingTime);
 

src/client/src/tscUtil.c

@@ -676,7 +676,7 @@ int32_t tscGetDataBlockFromList(void* pHashList, SDataBlockList* pDataBlockList,
                                 STableDataBlocks** dataBlocks) {
   *dataBlocks = NULL;
 
-  STableDataBlocks** t1 = (STableDataBlocks**)taosGetDataFromHash(pHashList, (const char*)&id, sizeof(id));
+  STableDataBlocks** t1 = (STableDataBlocks**)taosGetDataFromHashTable(pHashList, (const char*)&id, sizeof(id));
   if (t1 != NULL) {
     *dataBlocks = *t1;
   }

src/dnode/src/dnodeShell.c

@@ -442,7 +442,6 @@ void vnodeExecuteRetrieveReq(SSchedMsg *pSched) {
 
   if (code == TSDB_CODE_SUCCESS) {
     size = vnodeGetResultSize((void *)(pRetrieve->qhandle), &numOfRows);
-  }
   
     // buffer size for progress information, including meter count,
     // and for each meter, including 'uid' and 'TSKEY'.
@@ -457,6 +456,7 @@ void vnodeExecuteRetrieveReq(SSchedMsg *pSched) {
       taosSendSimpleRsp(pObj->thandle, TSDB_MSG_TYPE_RETRIEVE_RSP, TSDB_CODE_SERV_OUT_OF_MEMORY);
       goto _exit;
     }
+  }
 
   pMsg = pStart;
 
@@ -485,27 +485,29 @@ void vnodeExecuteRetrieveReq(SSchedMsg *pSched) {
 
   // write the progress information of each meter to response
   // this is required by subscriptions
+  if (numOfRows > 0 && code == TSDB_CODE_SUCCESS) {
     if (pQInfo->pMeterQuerySupporter != NULL && pQInfo->pMeterQuerySupporter->pMeterSidExtInfo != NULL) {
-    *((int32_t*)pMsg) = htonl(pQInfo->pMeterQuerySupporter->numOfMeters);
+      *((int32_t *)pMsg) = htonl(pQInfo->pMeterQuerySupporter->numOfMeters);
       pMsg += sizeof(int32_t);
       for (int32_t i = 0; i < pQInfo->pMeterQuerySupporter->numOfMeters; i++) {
-      *((int64_t*)pMsg) = htobe64(pQInfo->pMeterQuerySupporter->pMeterSidExtInfo[i]->uid);
+        *((int64_t *)pMsg) = htobe64(pQInfo->pMeterQuerySupporter->pMeterSidExtInfo[i]->uid);
         pMsg += sizeof(int64_t);
-      *((TSKEY*)pMsg) = htobe64(pQInfo->pMeterQuerySupporter->pMeterSidExtInfo[i]->key);
+        *((TSKEY *)pMsg) = htobe64(pQInfo->pMeterQuerySupporter->pMeterSidExtInfo[i]->key);
         pMsg += sizeof(TSKEY);
       }
     } else if (pQInfo->pObj != NULL) {
-    *((int32_t*)pMsg) = htonl(1);
+      *((int32_t *)pMsg) = htonl(1);
       pMsg += sizeof(int32_t);
-    *((int64_t*)pMsg) = htobe64(pQInfo->pObj->uid);
+      *((int64_t *)pMsg) = htobe64(pQInfo->pObj->uid);
       pMsg += sizeof(int64_t);
       if (pQInfo->pointsRead > 0) {
-      *((TSKEY*)pMsg) = htobe64(pQInfo->query.lastKey + 1);
+        *((TSKEY *)pMsg) = htobe64(pQInfo->query.lastKey + 1);
       } else {
-      *((TSKEY*)pMsg) = htobe64(pQInfo->query.lastKey);
+        *((TSKEY *)pMsg) = htobe64(pQInfo->query.lastKey);
       }
       pMsg += sizeof(TSKEY);
     }
+  }
 
   msgLen = pMsg - pStart;
 

src/inc/tpercentile.h

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

src/inc/tresultBuf.h

@@ -0,0 +1,104 @@
+#ifndef TDENGINE_VNODEQUERYUTIL_H
+#define TDENGINE_VNODEQUERYUTIL_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "os.h"
+#include "textbuffer.h"
+
+typedef struct SIDList {
+  uint32_t alloc;
+  int32_t  size;
+  int32_t* pData;
+} SIDList;
+
+typedef struct SQueryResultBuf {
+  int32_t  numOfRowsPerPage;
+  int32_t  numOfPages;
+  int64_t  totalBufSize;
+  int32_t  fd;                  // data file fd
+  int32_t  allocateId;          // allocated page id
+  int32_t  incStep;             // minimum allocated pages
+  char*    pBuf;                // mmap buffer pointer
+  char*    path;                // file path
+  
+  uint32_t numOfAllocGroupIds;  // number of allocated id list
+  void*    idsTable;            // id hash table
+  SIDList* list;                // for each id, there is a page id list
+} SQueryResultBuf;
+
+/**
+ * create disk-based result buffer
+ * @param pResultBuf
+ * @param size
+ * @param rowSize
+ * @return
+ */
+int32_t createResultBuf(SQueryResultBuf** pResultBuf, int32_t size, int32_t rowSize);
+
+/**
+ *
+ * @param pResultBuf
+ * @param groupId
+ * @param pageId
+ * @return
+ */
+tFilePage* getNewDataBuf(SQueryResultBuf* pResultBuf, int32_t groupId, int32_t* pageId);
+
+/**
+ *
+ * @param pResultBuf
+ * @return
+ */
+int32_t getNumOfRowsPerPage(SQueryResultBuf* pResultBuf);
+
+/**
+ *
+ * @param pResultBuf
+ * @param groupId
+ * @return
+ */
+SIDList getDataBufPagesIdList(SQueryResultBuf* pResultBuf, int32_t groupId);
+
+/**
+ * get the specified buffer page by id
+ * @param pResultBuf
+ * @param id
+ * @return
+ */
+tFilePage* getResultBufferPageById(SQueryResultBuf* pResultBuf, int32_t id);
+
+/**
+ * get the total buffer size in the format of disk file
+ * @param pResultBuf
+ * @return
+ */
+int32_t getResBufSize(SQueryResultBuf* pResultBuf);
+
+/**
+ * get the number of groups in the result buffer
+ * @param pResultBuf
+ * @return
+ */
+int32_t getNumOfResultBufGroupId(SQueryResultBuf* pResultBuf);
+
+/**
+ * destroy result buffer
+ * @param pResultBuf
+ */
+void destroyResultBuf(SQueryResultBuf* pResultBuf);
+
+/**
+ *
+ * @param pList
+ * @return
+ */
+int32_t getLastPageId(SIDList *pList);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // TDENGINE_VNODEQUERYUTIL_H

src/kit/shell/CMakeLists.txt

@@ -11,7 +11,13 @@ IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
   LIST(REMOVE_ITEM SRC ./src/shellWindows.c)
   LIST(REMOVE_ITEM SRC ./src/shellDarwin.c)
   ADD_EXECUTABLE(shell ${SRC})  
+  
+  IF (TD_PAGMODE_LITE)
+    TARGET_LINK_LIBRARIES(shell taos)
+  ELSE ()
     TARGET_LINK_LIBRARIES(shell taos_static)
+  ENDIF ()
+  
   SET_TARGET_PROPERTIES(shell PROPERTIES OUTPUT_NAME taos)
 ELSEIF (TD_WINDOWS_64)
   INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/deps/pthread)

src/kit/taosdemo/CMakeLists.txt

@@ -9,5 +9,11 @@ INCLUDE_DIRECTORIES(inc)
 IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
   AUX_SOURCE_DIRECTORY(. SRC)
   ADD_EXECUTABLE(taosdemo ${SRC})
+  
+  IF (TD_PAGMODE_LITE)
+    TARGET_LINK_LIBRARIES(taosdemo taos)
+  ELSE ()
     TARGET_LINK_LIBRARIES(taosdemo taos_static)
+  ENDIF ()
+  
 ENDIF ()

src/kit/taosdump/CMakeLists.txt

@@ -9,5 +9,12 @@ INCLUDE_DIRECTORIES(inc)
 IF ((TD_LINUX_64) OR (TD_LINUX_32 AND TD_ARM))
   AUX_SOURCE_DIRECTORY(. SRC)
   ADD_EXECUTABLE(taosdump ${SRC})
+  
+  IF (TD_PAGMODE_LITE)
+    TARGET_LINK_LIBRARIES(taosdump taos)
+  ELSE ()
     TARGET_LINK_LIBRARIES(taosdump taos_static)
+  ENDIF ()  
+  
+  
 ENDIF ()

src/mnode/src/mgmtDnodeInt.c

@@ -211,7 +211,7 @@ char *mgmtBuildCreateMeterIe(STabObj *pTable, char *pMsg, int vnode) {
 
   for (int i = 0; i < pTable->numOfColumns; ++i) {
     pCreateMeter->schema[i].type = pSchema[i].type;
-    /* strcpy(pCreateMeter->schema[i].name, pSchema[i].name); */
+    /* strcpy(pCreateMeter->schema[i].name, pColumnModel[i].name); */
     pCreateMeter->schema[i].bytes = htons(pSchema[i].bytes);
     pCreateMeter->schema[i].colId = htons(pSchema[i].colId);
   }

src/system/CMakeLists.txt

@@ -0,0 +1,8 @@
+CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
+PROJECT(TDengine)
+
+ADD_SUBDIRECTORY(detail)
+
+IF (TD_EDGE)
+  ADD_SUBDIRECTORY(lite)
+ENDIF ()

src/system/detail/CMakeLists.txt

@@ -0,0 +1,46 @@
+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/client/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/modules/http/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/modules/monitor/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/inc)
+  INCLUDE_DIRECTORIES(${TD_ENTERPRISE_DIR}/src/util/cluster/inc)
+  INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
+  INCLUDE_DIRECTORIES(inc)
+  AUX_SOURCE_DIRECTORY(./src SRC)
+  LIST(REMOVE_ITEM SRC ./src/vnodeFileUtil.c)
+  LIST(REMOVE_ITEM SRC ./src/taosGrant.c)
+
+  ADD_EXECUTABLE(taosd ${SRC})
+
+  IF (TD_PAGMODE_LITE)
+    TARGET_LINK_LIBRARIES(taosd taos trpc tutil sdb monitor pthread http)
+  ELSE ()
+    TARGET_LINK_LIBRARIES(taosd taos_static trpc tutil sdb monitor pthread http)
+  ENDIF ()
+
+  IF (TD_EDGE)
+    TARGET_LINK_LIBRARIES(taosd taosd_edge)
+  ELSE ()
+    TARGET_LINK_LIBRARIES(taosd taosd_cluster)
+  ENDIF ()
+
+  SET(PREPARE_ENV_CMD "prepare_env_cmd")
+  SET(PREPARE_ENV_TARGET "prepare_env_target")
+  ADD_CUSTOM_COMMAND(OUTPUT ${PREPARE_ENV_CMD}
+    POST_BUILD 
+    COMMAND echo "make test directory"
+    DEPENDS taosd
+    COMMAND ${CMAKE_COMMAND} -E make_directory ${TD_TESTS_OUTPUT_DIR}/cfg/
+    COMMAND ${CMAKE_COMMAND} -E make_directory ${TD_TESTS_OUTPUT_DIR}/log/
+    COMMAND ${CMAKE_COMMAND} -E make_directory ${TD_TESTS_OUTPUT_DIR}/data/
+    COMMAND ${CMAKE_COMMAND} -E echo dataDir ${TD_TESTS_OUTPUT_DIR}/data > ${TD_TESTS_OUTPUT_DIR}/cfg/taos.cfg
+    COMMAND ${CMAKE_COMMAND} -E echo logDir ${TD_TESTS_OUTPUT_DIR}/log  >> ${TD_TESTS_OUTPUT_DIR}/cfg/taos.cfg
+    COMMAND ${CMAKE_COMMAND} -E echo charset UTF-8  >> ${TD_TESTS_OUTPUT_DIR}/cfg/taos.cfg
+    COMMENT "prepare taosd environment")
+  ADD_CUSTOM_TARGET(${PREPARE_ENV_TARGET} ALL WORKING_DIRECTORY ${TD_EXECUTABLE_OUTPUT_PATH} DEPENDS ${PREPARE_ENV_CMD})
+
+ENDIF ()

src/system/lite/CMakeLists.txt

@@ -0,0 +1,17 @@
+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/client/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/modules/http/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/modules/monitor/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/src/system/detail/inc)
+  INCLUDE_DIRECTORIES(${TD_COMMUNITY_DIR}/inc)
+  INCLUDE_DIRECTORIES(${TD_ENTERPRISE_DIR}/src/util/cluster/inc)
+  INCLUDE_DIRECTORIES(${TD_OS_DIR}/inc)
+  INCLUDE_DIRECTORIES(inc)
+
+  AUX_SOURCE_DIRECTORY(./src SRC)
+  ADD_LIBRARY(taosd_edge ${SRC})
+ENDIF ()

src/util/inc/hash.h

@@ -43,10 +43,10 @@ typedef struct SHashEntry {
 
 typedef struct HashObj {
   SHashEntry **hashList;
-  uint32_t     capacity;
+  uint32_t     capacity;         // number of slots
-  int          size;
+  int          size;             // number of elements in hash table
-  _hash_fn_t   hashFp;
+  _hash_fn_t   hashFp;           // hash function
-  bool         multithreadSafe;  // enable lock
+  bool         multithreadSafe;  // enable lock or not
 
 #if defined LINUX
   pthread_rwlock_t lock;
@@ -57,11 +57,13 @@ typedef struct HashObj {
 } HashObj;
 
 void *taosInitHashTable(uint32_t capacity, _hash_fn_t fn, bool multithreadSafe);
-
-int32_t taosAddToHashTable(HashObj *pObj, const char *key, uint32_t keyLen, void *data, uint32_t size);
 void  taosDeleteFromHashTable(HashObj *pObj, const char *key, uint32_t keyLen);
 
-char *taosGetDataFromHash(HashObj *pObj, const char *key, uint32_t keyLen);
+int32_t taosAddToHashTable(HashObj *pObj, const char *key, uint32_t keyLen, void *data, uint32_t size);
+int32_t taosNumElemsInHashTable(HashObj *pObj);
+
+char *taosGetDataFromHashTable(HashObj *pObj, const char *key, uint32_t keyLen);
+
 
 void taosCleanUpHashTable(void *handle);
 

src/util/inc/textbuffer.h

@@ -19,21 +19,15 @@
 extern "C" {
 #endif
 
-#include <stdio.h>
+#include "os.h"
-#include <stdlib.h>
-#include <string.h>
-
-#include "tutil.h"
 #include "taosmsg.h"
+#include "tutil.h"
 
 #define DEFAULT_PAGE_SIZE 16384  // 16k larger than the SHistoInfo
 #define MIN_BUFFER_SIZE (1 << 19)
 #define MAX_TMPFILE_PATH_LENGTH PATH_MAX
 #define INITIAL_ALLOCATION_BUFFER_SIZE 64
 
-// forward declare
-struct tTagSchema;
-
 typedef enum EXT_BUFFER_FLUSH_MODEL {
   /*
    * all data that have been flushed to disk is belonged to the same group
@@ -61,12 +55,12 @@ typedef struct tFlushoutData {
   tFlushoutInfo *pFlushoutInfo;
 } tFlushoutData;
 
-typedef struct tFileMeta {
+typedef struct SFileInfo {
   uint32_t      nFileSize;  // in pages
-  uint32_t      nPageSize;
+  uint32_t      pageSize;
   uint32_t      numOfElemsInFile;
   tFlushoutData flushoutData;
-} tFileMeta;
+} SFileInfo;
 
 typedef struct tFilePage {
   uint64_t numOfElems;
@@ -78,51 +72,50 @@ typedef struct tFilePagesItem {
   tFilePage              item;
 } tFilePagesItem;
 
-typedef struct tColModel {
+typedef struct SSchemaEx {
-  int32_t         maxCapacity;
+  struct SSchema field;
-  int32_t         numOfCols;
+  int16_t        offset;
-  int16_t *       colOffset;
+} SSchemaEx;
-  struct SSchema *pFields;
-} tColModel;
 
-typedef struct tOrderIdx {
+typedef struct SColumnModel {
-  int32_t numOfOrderedCols;
+  int32_t    capacity;
+  int32_t    numOfCols;
+  int16_t    rowSize;
+  SSchemaEx *pFields;
+} SColumnModel;
+
+typedef struct SColumnOrderInfo {
+  int32_t numOfCols;
   int16_t pData[];
-} tOrderIdx;
+} SColumnOrderInfo;
 
 typedef struct tOrderDescriptor {
-  union {
+  SColumnModel *   pColumnModel;
-    struct tTagSchema *pTagSchema;
-    tColModel *        pSchema;
-  };
   int32_t          tsOrder;  // timestamp order type if exists
-  tOrderIdx orderIdx;
+  SColumnOrderInfo orderIdx;
 } tOrderDescriptor;
 
 typedef struct tExtMemBuffer {
-  int32_t nMaxSizeInPages;
+  int32_t inMemCapacity;
-
   int32_t nElemSize;
-  int32_t nPageSize;
+  int32_t pageSize;
-
+  int32_t numOfTotalElems;
-  int32_t numOfAllElems;
   int32_t numOfElemsInBuffer;
   int32_t numOfElemsPerPage;
+  int16_t numOfInMemPages;
 
-  int16_t         numOfPagesInMem;
   tFilePagesItem *pHead;
   tFilePagesItem *pTail;
 
-  tFileMeta fileMeta;
+  char *    path;
-
+  FILE *    file;
-  char  dataFilePath[MAX_TMPFILE_PATH_LENGTH];
+  SFileInfo fileMeta;
-  FILE *dataFile;
-
-  tColModel *pColModel;
 
+  SColumnModel *         pColumnModel;
   EXT_BUFFER_FLUSH_MODEL flushModel;
 } tExtMemBuffer;
 
+<<<<<<< HEAD:src/util/inc/textbuffer.h
 typedef struct tTagSchema {
   struct SSchema *pSchema;
   int32_t         numOfCols;
@@ -139,20 +132,32 @@ typedef struct tSidSet {
   tOrderIdx   orderIdx;
 } tSidSet;
 
+=======
+/**
+ *
+ * @param fileNamePattern
+ * @param dstPath
+ */
+>>>>>>> develop:src/inc/textbuffer.h
 void getTmpfilePath(const char *fileNamePattern, char *dstPath);
 
-/*
+/**
- * create ext-memory buffer
+ *
+ * @param inMemSize
+ * @param elemSize
+ * @param pModel
+ * @return
  */
-void tExtMemBufferCreate(tExtMemBuffer **pMemBuffer, int32_t numOfBufferSize, int32_t elemSize,
+tExtMemBuffer *createExtMemBuffer(int32_t inMemSize, int32_t elemSize, SColumnModel *pModel);
-                         const char *tmpDataFilePath, tColModel *pModel);
 
-/*
+/**
- * destroy ext-memory buffer
+ *
+ * @param pMemBuffer
+ * @return
  */
-void tExtMemBufferDestroy(tExtMemBuffer **pMemBuffer);
+void *destoryExtMemBuffer(tExtMemBuffer *pMemBuffer);
 
-/*
+/**
  * @param pMemBuffer
  * @param data       input data pointer
  * @param numOfRows  number of rows in data
@@ -161,12 +166,15 @@ void tExtMemBufferDestroy(tExtMemBuffer **pMemBuffer);
  */
 int16_t tExtMemBufferPut(tExtMemBuffer *pMemBuffer, void *data, int32_t numOfRows);
 
-/*
+/**
- * flush all data into disk and release all in-memory buffer
+ *
+ * @param pMemBuffer
+ * @return
  */
 bool tExtMemBufferFlush(tExtMemBuffer *pMemBuffer);
 
-/*
+/**
+ *
  * remove all data that has been put into buffer, including in buffer or
  * ext-buffer(disk)
  */
@@ -179,11 +187,44 @@ void tExtMemBufferClear(tExtMemBuffer *pMemBuffer);
  */
 bool tExtMemBufferLoadData(tExtMemBuffer *pMemBuffer, tFilePage *pFilePage, int32_t flushIdx, int32_t pageIdx);
 
+/**
+ *
+ * @param pMemBuffer
+ * @return
+ */
 bool tExtMemBufferIsAllDataInMem(tExtMemBuffer *pMemBuffer);
 
-tColModel *tColModelCreate(SSchema *field, int32_t numOfCols, int32_t maxCapacity);
+/**
+ *
+ * @param fields
+ * @param numOfCols
+ * @param blockCapacity
+ * @return
+ */
+SColumnModel *createColumnModel(SSchema *fields, int32_t numOfCols, int32_t blockCapacity);
 
-void tColModelDestroy(tColModel *pModel);
+/**
+ *
+ * @param pSrc
+ * @return
+ */
+SColumnModel *cloneColumnModel(SColumnModel *pSrc);
+
+/**
+ *
+ * @param pModel
+ */
+void destroyColumnModel(SColumnModel *pModel);
+
+/*
+ * compress data into consecutive block without hole in data
+ */
+void tColModelCompact(SColumnModel *pModel, tFilePage *inputBuffer, int32_t maxElemsCapacity);
+
+void     tColModelErase(SColumnModel *pModel, tFilePage *inputBuffer, int32_t maxCapacity, int32_t s, int32_t e);
+SSchema *getColumnModelSchema(SColumnModel *pColumnModel, int32_t index);
+
+int16_t getColumnModelOffset(SColumnModel *pColumnModel, int32_t index);
 
 typedef struct SSrcColumnInfo {
   int32_t functionId;
@@ -193,68 +234,18 @@ typedef struct SSrcColumnInfo {
 /*
  * display data in column format model for debug purpose only
  */
-void tColModelDisplay(tColModel *pModel, void *pData, int32_t numOfRows, int32_t maxCount);
+void tColModelDisplay(SColumnModel *pModel, void *pData, int32_t numOfRows, int32_t maxCount);
 
-void tColModelDisplayEx(tColModel *pModel, void *pData, int32_t numOfRows, int32_t maxCount, SSrcColumnInfo *pInfo);
+void tColModelDisplayEx(SColumnModel *pModel, void *pData, int32_t numOfRows, int32_t maxCount, SSrcColumnInfo *pInfo);
 
-/*
+tOrderDescriptor *tOrderDesCreate(const int32_t *orderColIdx, int32_t numOfOrderCols, SColumnModel *pModel,
- * compress data into consecutive block without hole in data
+                                  int32_t tsOrderType);
- */
-void tColModelCompact(tColModel *pModel, tFilePage *inputBuffer, int32_t maxElemsCapacity);
-
-void tColModelErase(tColModel *pModel, tFilePage *inputBuffer, int32_t maxCapacity, int32_t s, int32_t e);
-
-tOrderDescriptor *tOrderDesCreate(int32_t *orderColIdx, int32_t numOfOrderCols, tColModel *pModel, int32_t tsOrderType);
 
 void tOrderDescDestroy(tOrderDescriptor *pDesc);
 
-void tColModelAppend(tColModel *dstModel, tFilePage *dstPage, void *srcData, int32_t srcStartRows,
+void tColModelAppend(SColumnModel *dstModel, tFilePage *dstPage, void *srcData, int32_t srcStartRows,
                      int32_t numOfRowsToWrite, int32_t srcCapacity);
 
-///////////////////////////////////////////////////////////////////////////////////////////////////////
-typedef struct MinMaxEntry {
-  union {
-    double  dMinVal;
-    int32_t iMinVal;
-    int64_t i64MinVal;
-  };
-  union {
-    double  dMaxVal;
-    int32_t iMaxVal;
-    int64_t i64MaxVal;
-  };
-} MinMaxEntry;
-
-typedef struct tMemBucketSegment {
-  int32_t         numOfSlots;
-  MinMaxEntry *   pBoundingEntries;
-  tExtMemBuffer **pBuffer;
-} tMemBucketSegment;
-
-typedef struct tMemBucket {
-  int16_t numOfSegs;
-  int16_t nTotalSlots;
-  int16_t nSlotsOfSeg;
-  int16_t dataType;
-
-  int16_t nElemSize;
-  int32_t numOfElems;
-
-  int32_t nTotalBufferSize;
-  int32_t maxElemsCapacity;
-
-  int16_t nPageSize;
-  int16_t numOfTotalPages;
-  int16_t numOfAvailPages; /* remain available buffer pages */
-
-  tMemBucketSegment *pSegs;
-  tOrderDescriptor * pOrderDesc;
-
-  MinMaxEntry nRange;
-
-  void (*HashFunc)(struct tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx);
-} tMemBucket;
-
 typedef int (*__col_compar_fn_t)(tOrderDescriptor *, int32_t numOfRows, int32_t idx1, int32_t idx2, char *data);
 
 void tColDataQSort(tOrderDescriptor *, int32_t numOfRows, int32_t start, int32_t end, char *data, int32_t orderType);
@@ -269,19 +260,6 @@ int32_t compare_a(tOrderDescriptor *, int32_t numOfRow1, int32_t s1, char *data1
 int32_t compare_d(tOrderDescriptor *, int32_t numOfRow1, int32_t s1, char *data1, int32_t numOfRow2, int32_t s2,
                   char *data2);
 
-tMemBucket* tMemBucketCreate(int32_t totalSlots, int32_t nBufferSize, int16_t nElemSize,
-                      int16_t dataType, tOrderDescriptor *pDesc);
-
-void tMemBucketDestroy(tMemBucket *pBucket);
-
-void tMemBucketPut(tMemBucket *pBucket, void *data, int32_t numOfRows);
-
-double getPercentile(tMemBucket *pMemBucket, double percent);
-
-void tBucketIntHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx);
-
-void tBucketDoubleHash(tMemBucket *pBucket, void *value, int16_t *segIdx, int16_t *slotIdx);
-
 #ifdef __cplusplus
 }
 #endif

src/util/inc/thistogram.h

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

src/util/inc/tinterpolation.h

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

src/util/src/hash.c

@@ -346,6 +346,14 @@ static void doAddToHashTable(HashObj *pObj, SHashNode *pNode) {
 //  pTrace("key:%s %p add to hash table", key, pNode);
 }
 
+int32_t taosNumElemsInHashTable(HashObj *pObj) {
+  if (pObj == NULL) {
+    return 0;
+  }
+  
+  return pObj->size;
+}
+
 /**
  * add data node into hash table
  * @param pObj    hash object
@@ -392,7 +400,7 @@ int32_t taosAddToHashTable(HashObj *pObj, const char *key, uint32_t keyLen, void
   return 0;
 }
 
-char *taosGetDataFromHash(HashObj *pObj, const char *key, uint32_t keyLen) {
+char *taosGetDataFromHashTable(HashObj *pObj, const char *key, uint32_t keyLen) {
   if (pObj->multithreadSafe) {
     __rd_lock(&pObj->lock);
   }

src/util/src/tinterpolation.c

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

src/util/src/tpercentile.c

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

src/util/src/tresultBuf.c

@@ -0,0 +1,225 @@
+#include "hash.h"
+#include "taoserror.h"
+#include "textbuffer.h"
+#include "tlog.h"
+#include "tsqlfunction.h"
+#include "tresultBuf.h"
+
+#define DEFAULT_INTERN_BUF_SIZE 16384L
+
+int32_t createResultBuf(SQueryResultBuf** pResultBuf, int32_t size, int32_t rowSize) {
+  SQueryResultBuf* pResBuf = calloc(1, sizeof(SQueryResultBuf));
+  pResBuf->numOfRowsPerPage = (DEFAULT_INTERN_BUF_SIZE - sizeof(tFilePage)) / rowSize;
+  pResBuf->numOfPages = size;
+
+  pResBuf->totalBufSize = pResBuf->numOfPages * DEFAULT_INTERN_BUF_SIZE;
+  pResBuf->incStep = 4;
+
+  // init id hash table
+  pResBuf->idsTable = taosInitHashTable(size, taosGetDefaultHashFunction(TSDB_DATA_TYPE_INT), false);
+  pResBuf->list = calloc(size, sizeof(SIDList));
+  pResBuf->numOfAllocGroupIds = size;
+
+  char path[4096] = {0};
+  getTmpfilePath("tsdb_q_buf", path);
+  pResBuf->path = strdup(path);
+
+  pResBuf->fd = open(pResBuf->path, O_CREAT | O_RDWR, 0666);
+  
+  memset(path, 0, tListLen(path));
+
+  if (!FD_VALID(pResBuf->fd)) {
+    pError("failed to create tmp file: %s on disk. %s", pResBuf->path, strerror(errno));
+    return TSDB_CODE_CLI_NO_DISKSPACE;
+  }
+
+  int32_t ret = ftruncate(pResBuf->fd, pResBuf->numOfPages * DEFAULT_INTERN_BUF_SIZE);
+  if (ret != TSDB_CODE_SUCCESS) {
+    pError("failed to create tmp file: %s on disk. %s", pResBuf->path, strerror(errno));
+    return TSDB_CODE_CLI_NO_DISKSPACE;
+  }
+
+  pResBuf->pBuf = mmap(NULL, pResBuf->totalBufSize, PROT_READ | PROT_WRITE, MAP_SHARED, pResBuf->fd, 0);
+  if (pResBuf->pBuf == MAP_FAILED) {
+    pError("QInfo:%p failed to map temp file: %s. %s", pResBuf->path, strerror(errno));
+    return TSDB_CODE_CLI_OUT_OF_MEMORY; // todo change error code
+  }
+
+  pTrace("create tmp file for output result, %s, " PRId64 "bytes", pResBuf->path, pResBuf->totalBufSize);
+  *pResultBuf = pResBuf;
+  return TSDB_CODE_SUCCESS;
+}
+
+tFilePage* getResultBufferPageById(SQueryResultBuf* pResultBuf, int32_t id) {
+  assert(id < pResultBuf->numOfPages && id >= 0);
+
+  return (tFilePage*)(pResultBuf->pBuf + DEFAULT_INTERN_BUF_SIZE * id);
+}
+
+int32_t getNumOfResultBufGroupId(SQueryResultBuf* pResultBuf) { return taosNumElemsInHashTable(pResultBuf->idsTable); }
+
+int32_t getResBufSize(SQueryResultBuf* pResultBuf) { return pResultBuf->totalBufSize; }
+
+static int32_t extendDiskFileSize(SQueryResultBuf* pResultBuf, int32_t numOfPages) {
+  assert(pResultBuf->numOfPages * DEFAULT_INTERN_BUF_SIZE == pResultBuf->totalBufSize);
+
+  int32_t ret = munmap(pResultBuf->pBuf, pResultBuf->totalBufSize);
+  pResultBuf->numOfPages += numOfPages;
+
+  /*
+   * disk-based output buffer is exhausted, try to extend the disk-based buffer, the available disk space may
+   * be insufficient
+   */
+  ret = ftruncate(pResultBuf->fd, pResultBuf->numOfPages * DEFAULT_INTERN_BUF_SIZE);
+  if (ret != 0) {
+    //    dError("QInfo:%p failed to create intermediate result output file:%s. %s", pQInfo, pSupporter->extBufFile,
+    //           strerror(errno));
+    return -TSDB_CODE_SERV_NO_DISKSPACE;
+  }
+
+  pResultBuf->totalBufSize = pResultBuf->numOfPages * DEFAULT_INTERN_BUF_SIZE;
+  pResultBuf->pBuf = mmap(NULL, pResultBuf->totalBufSize, PROT_READ | PROT_WRITE, MAP_SHARED, pResultBuf->fd, 0);
+
+  if (pResultBuf->pBuf == MAP_FAILED) {
+    //    dError("QInfo:%p failed to map temp file: %s. %s", pQInfo, pSupporter->extBufFile, strerror(errno));
+    return -TSDB_CODE_SERV_OUT_OF_MEMORY;
+  }
+
+  return TSDB_CODE_SUCCESS;
+}
+
+static bool noMoreAvailablePages(SQueryResultBuf* pResultBuf) {
+  return (pResultBuf->allocateId == pResultBuf->numOfPages - 1);
+}
+
+static int32_t getGroupIndex(SQueryResultBuf* pResultBuf, int32_t groupId) {
+  assert(pResultBuf != NULL);
+
+  char* p = taosGetDataFromHashTable(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 < pResultBuf->numOfAllocGroupIds);
+
+  return slot;
+}
+
+static int32_t addNewGroupId(SQueryResultBuf* pResultBuf, int32_t groupId) {
+  int32_t num = getNumOfResultBufGroupId(pResultBuf); // the num is the newest allocated group id slot
+
+  if (pResultBuf->numOfAllocGroupIds <= num) {
+    size_t n = pResultBuf->numOfAllocGroupIds << 1u;
+
+    SIDList* p = (SIDList*)realloc(pResultBuf->list, sizeof(SIDList) * n);
+    assert(p != NULL);
+
+    memset(&p[pResultBuf->numOfAllocGroupIds], 0, sizeof(SIDList) * pResultBuf->numOfAllocGroupIds);
+    
+    pResultBuf->list = p;
+    pResultBuf->numOfAllocGroupIds = n;
+  }
+
+  taosAddToHashTable(pResultBuf->idsTable, (const char*)&groupId, sizeof(int32_t), &num, sizeof(int32_t));
+  return num;
+}
+
+static int32_t doRegisterId(SIDList* pList, int32_t id) {
+  if (pList->size >= pList->alloc) {
+    int32_t s = 0;
+    if (pList->alloc == 0) {
+      s = 4;
+      assert(pList->pData == NULL);
+    } else {
+      s = pList->alloc << 1u;
+    }
+    
+    int32_t* c = realloc(pList->pData, s * sizeof(int32_t));
+    assert(c);
+    
+    memset(&c[pList->alloc], 0, sizeof(int32_t) * pList->alloc);
+
+    pList->pData = c;
+    pList->alloc = s;
+  }
+
+  pList->pData[pList->size++] = id;
+  return 0;
+}
+
+static void registerPageId(SQueryResultBuf* pResultBuf, int32_t groupId, int32_t pageId) {
+  int32_t slot = getGroupIndex(pResultBuf, groupId);
+  if (slot < 0) {
+    slot = addNewGroupId(pResultBuf, groupId);
+  }
+
+  SIDList* pList = &pResultBuf->list[slot];
+  doRegisterId(pList, pageId);
+}
+
+tFilePage* getNewDataBuf(SQueryResultBuf* pResultBuf, int32_t groupId, int32_t* pageId) {
+  if (noMoreAvailablePages(pResultBuf)) {
+    if (extendDiskFileSize(pResultBuf, pResultBuf->incStep) != TSDB_CODE_SUCCESS) {
+      return NULL;
+    }
+  }
+
+  // register new id in this group
+  *pageId = (pResultBuf->allocateId++);
+  registerPageId(pResultBuf, groupId, *pageId);
+
+  tFilePage* page = getResultBufferPageById(pResultBuf, *pageId);
+  
+  // clear memory for the new page
+  memset(page, 0, DEFAULT_INTERN_BUF_SIZE);
+  
+  return page;
+}
+
+int32_t getNumOfRowsPerPage(SQueryResultBuf* pResultBuf) { return pResultBuf->numOfRowsPerPage; }
+
+SIDList getDataBufPagesIdList(SQueryResultBuf* pResultBuf, int32_t groupId) {
+  SIDList list = {0};
+  int32_t slot = getGroupIndex(pResultBuf, groupId);
+  if (slot < 0) {
+    return list;
+  } else {
+    return pResultBuf->list[slot];
+  }
+}
+
+void destroyResultBuf(SQueryResultBuf* pResultBuf) {
+  if (pResultBuf == NULL) {
+    return;
+  }
+
+  if (FD_VALID(pResultBuf->fd)) {
+    close(pResultBuf->fd);
+  }
+
+  pTrace("disk-based output buffer closed, %" PRId64 " bytes, file:%s", pResultBuf->totalBufSize, pResultBuf->path);
+  munmap(pResultBuf->pBuf, pResultBuf->totalBufSize);
+  unlink(pResultBuf->path);
+  
+  tfree(pResultBuf->path);
+
+  for (int32_t i = 0; i < pResultBuf->numOfAllocGroupIds; ++i) {
+    SIDList* pList = &pResultBuf->list[i];
+    tfree(pList->pData);
+  }
+
+  tfree(pResultBuf->list);
+  taosCleanUpHashTable(pResultBuf->idsTable);
+  
+  tfree(pResultBuf);
+}
+
+int32_t getLastPageId(SIDList *pList) {
+  if (pList == NULL && pList->size <= 0) {
+    return -1;
+  }
+  
+  return pList->pData[pList->size - 1];
+}
+

src/util/src/ttokenizer.c

@@ -274,7 +274,7 @@ int tSQLKeywordCode(const char* z, int n) {
     }
   }
 
-  SKeyword** pKey = (SKeyword**)taosGetDataFromHash(KeywordHashTable, key, n);
+  SKeyword** pKey = (SKeyword**)taosGetDataFromHashTable(KeywordHashTable, key, n);
   if (pKey != NULL) {
     return (*pKey)->type;
   } else {

src/vnode/detail/inc/vnodeQueryImpl.h

@@ -13,8 +13,8 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
 
-#ifndef TDENGINE_VNODEQUERYUTIL_H
+#ifndef TDENGINE_VNODEQUERYIMPL_H
-#define TDENGINE_VNODEQUERYUTIL_H
+#define TDENGINE_VNODEQUERYIMPL_H
 
 #ifdef __cplusplus
 extern "C" {
@@ -120,7 +120,7 @@ typedef enum {
 typedef int (*__block_search_fn_t)(char* data, int num, int64_t key, int order);
 
 static FORCE_INLINE SMeterObj* getMeterObj(void* hashHandle, int32_t sid) {
-  return *(SMeterObj**)taosGetDataFromHash(hashHandle, (const char*) &sid, sizeof(sid));
+  return *(SMeterObj**)taosGetDataFromHashTable(hashHandle, (const char*) &sid, sizeof(sid));
 }
 
 bool isQueryKilled(SQuery* pQuery);
@@ -209,7 +209,7 @@ int32_t vnodeGetHeaderFile(SQueryRuntimeEnv *pRuntimeEnv, int32_t fileIndex);
  * @param ekey
  * @return
  */
-SMeterQueryInfo* createMeterQueryInfo(SQuery* pQuery, TSKEY skey, TSKEY ekey);
+SMeterQueryInfo* createMeterQueryInfo(SQuery* pQuery, int32_t sid, TSKEY skey, TSKEY ekey);
 
 /**
  * Destroy meter query info
@@ -224,7 +224,7 @@ void destroyMeterQueryInfo(SMeterQueryInfo *pMeterQueryInfo, int32_t numOfCols);
  * @param skey
  * @param ekey
  */
-void changeMeterQueryInfoForSuppleQuery(SMeterQueryInfo *pMeterQueryInfo, TSKEY skey, TSKEY ekey);
+void changeMeterQueryInfoForSuppleQuery(SQueryResultBuf* pResultBuf, SMeterQueryInfo *pMeterQueryInfo, TSKEY skey, TSKEY ekey);
 
 /**
  * add the new allocated disk page to meter query info
@@ -276,11 +276,11 @@ void displayInterResult(SData** pdata, SQuery* pQuery, int32_t numOfRows);
 
 void vnodePrintQueryStatistics(SMeterQuerySupportObj* pSupporter);
 
-void clearGroupResultBuf(SOutputRes* pOneOutputRes, int32_t nOutputCols);
+void clearGroupResultBuf(SQueryRuntimeEnv *pRuntimeEnv, SOutputRes *pOneOutputRes);
-void copyGroupResultBuf(SOutputRes* dst, const SOutputRes* src, int32_t nOutputCols);
+void copyGroupResultBuf(SQueryRuntimeEnv *pRuntimeEnv, SOutputRes* dst, const SOutputRes* src);
 
-void resetSlidingWindowInfo(SSlidingWindowInfo* pSlidingWindowInfo, int32_t numOfCols);
+void resetSlidingWindowInfo(SQueryRuntimeEnv *pRuntimeEnv, SSlidingWindowInfo* pSlidingWindowInfo);
-void clearCompletedSlidingWindows(SSlidingWindowInfo* pSlidingWindowInfo, int32_t numOfCols);
+void clearCompletedSlidingWindows(SQueryRuntimeEnv* pRuntimeEnv);
 int32_t numOfClosedSlidingWindow(SSlidingWindowInfo* pSlidingWindowInfo);
 void closeSlidingWindow(SSlidingWindowInfo* pSlidingWindowInfo, int32_t slot);
 void closeAllSlidingWindow(SSlidingWindowInfo* pSlidingWindowInfo);
@@ -289,4 +289,4 @@ void closeAllSlidingWindow(SSlidingWindowInfo* pSlidingWindowInfo);
 }
 #endif
 
-#endif  // TDENGINE_VNODEQUERYUTIL_H
+#endif  // TDENGINE_VNODEQUERYIMPL_H

src/vnode/detail/inc/vnodeRead.h

@@ -21,6 +21,7 @@ extern "C" {
 #endif
 
 #include "os.h"
+#include "tresultBuf.h"
 
 #include "tinterpolation.h"
 #include "vnodeTagMgmt.h"
@@ -84,10 +85,15 @@ typedef struct SQueryCostSummary {
   int64_t tmpBufferInDisk;  // size of buffer for intermediate result
 } SQueryCostSummary;
 
+typedef struct SPosInfo {
+  int64_t pageId;
+  int32_t rowId;
+} SPosInfo;
+
 typedef struct SOutputRes {
   uint16_t     numOfRows;
   int32_t      nAlloc;
-  tFilePage**  result;
+  SPosInfo     pos;
   SResultInfo* resultInfo;
 } SOutputRes;
 
@@ -174,16 +180,17 @@ typedef struct SQueryRuntimeEnv {
    * So we keep a copy of the support structure as well as the cache block data itself.
    */
   SCacheBlock         cacheBlock;
+  
+  SQueryResultBuf*    pResultBuf;
+  bool                stableQuery;  // is super table query or not
 } SQueryRuntimeEnv;
 
-/* intermediate result during multimeter query involves interval */
+/* intermediate pos during multimeter query involves interval */
 typedef struct SMeterQueryInfo {
   int64_t      lastKey;
   int64_t      skey;
   int64_t      ekey;
   int32_t      numOfRes;
-  uint32_t     numOfPages;
-  uint32_t     numOfAlloc;
   int32_t      reverseIndex;    // reversed output indicator, start from (numOfRes-1)
   int16_t      reverseFillRes;  // denote if reverse fill the results in supplementary scan required or not
   int16_t      queryRangeSet;   // denote if the query range is set, only available for interval query
@@ -191,7 +198,7 @@ typedef struct SMeterQueryInfo {
   int64_t      tag;
   STSCursor    cur;
   SResultInfo* resultInfo;
-  uint32_t*    pageList;
+  int32_t      sid; // for retrieve the page id list
 } SMeterQueryInfo;
 
 typedef struct SMeterDataInfo {
@@ -235,16 +242,8 @@ typedef struct SMeterQuerySupportObj {
    */
   int32_t meterIdx;
 
-  int32_t meterOutputFd;
-  int32_t lastPageId;
-  int32_t numOfPages;
   int32_t numOfGroupResultPages;
   int32_t groupResultSize;
-
-  char*   meterOutputMMapBuf;
-  int64_t bufSize;
-  char    extBufFile[256];  // external file name
-
   SMeterDataInfo* pMeterDataInfo;
 
   TSKEY* tsList;

src/vnode/detail/inc/vnodeTagMgmt.h

@@ -32,14 +32,21 @@ extern "C" {
  * Note:
  * 1. we implement a quick sort algorithm, may remove it later.
  */
+typedef struct tSidSet {
+  int32_t            numOfSids;
+  int32_t            numOfSubSet;
+  SMeterSidExtInfo **pSids;
+  int32_t *          starterPos;  // position of each subgroup, generated according to
+
+  SColumnModel      *pColumnModel;
+  SColumnOrderInfo   orderIdx;
+} tSidSet;
 
 typedef int32_t (*__ext_compar_fn_t)(const void *p1, const void *p2, void *param);
 
 tSidSet *tSidSetCreate(struct SMeterSidExtInfo **pMeterSidExtInfo, int32_t numOfMeters, SSchema *pSchema,
                        int32_t numOfTags, SColIndexEx *colList, int32_t numOfOrderCols);
 
-tTagSchema *tCreateTagSchema(SSchema *pSchema, int32_t numOfTagCols);
-
 int32_t *calculateSubGroup(void **pSids, int32_t numOfMeters, int32_t *numOfSubset, tOrderDescriptor *pOrderDesc,
                            __ext_compar_fn_t compareFn);
 

src/vnode/detail/src/vnodeFile.c

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

src/vnode/detail/src/vnodeQueryProcess.c

@@ -132,7 +132,7 @@ static void queryOnMultiDataCache(SQInfo *pQInfo, SMeterDataInfo *pMeterInfo) {
       pRuntimeEnv->pMeterObj = pMeterObj;
 
       if (pMeterInfo[k].pMeterQInfo == NULL) {
-        pMeterInfo[k].pMeterQInfo = createMeterQueryInfo(pQuery, pSupporter->rawSKey, pSupporter->rawEKey);
+        pMeterInfo[k].pMeterQInfo = createMeterQueryInfo(pQuery, pMeterObj->sid, pSupporter->rawSKey, pSupporter->rawEKey);
       }
 
       if (pMeterInfo[k].pMeterObj == NULL) {  // no data in disk for this meter, set its pointer
@@ -683,7 +683,7 @@ static void vnodeSTableSeqProcessor(SQInfo *pQInfo) {
     }
   
     resetCtxOutputBuf(pRuntimeEnv);
-    resetSlidingWindowInfo(&pRuntimeEnv->swindowResInfo, pQuery->numOfOutputCols);
+    resetSlidingWindowInfo(pRuntimeEnv, &pRuntimeEnv->swindowResInfo);
     
     while (pSupporter->meterIdx < pSupporter->numOfMeters) {
       int32_t k = pSupporter->meterIdx;
@@ -858,7 +858,9 @@ static void doOrderedScan(SQInfo *pQInfo) {
 static void setupMeterQueryInfoForSupplementQuery(SMeterQuerySupportObj *pSupporter) {
   for (int32_t i = 0; i < pSupporter->numOfMeters; ++i) {
     SMeterQueryInfo *pMeterQueryInfo = pSupporter->pMeterDataInfo[i].pMeterQInfo;
-    changeMeterQueryInfoForSuppleQuery(pMeterQueryInfo, pSupporter->rawSKey, pSupporter->rawEKey);
+    SQueryResultBuf* pResultBuf = pSupporter->runtimeEnv.pResultBuf;
+    
+    changeMeterQueryInfoForSuppleQuery(pResultBuf, pMeterQueryInfo, pSupporter->rawSKey, pSupporter->rawEKey);
   }
 }
 
@@ -1086,7 +1088,7 @@ static void vnodeSingleMeterIntervalMainLooper(SMeterQuerySupportObj *pSupporter
            (pQuery->skey >= pQuery->ekey && !QUERY_IS_ASC_QUERY(pQuery)));
 
     initCtxOutputBuf(pRuntimeEnv);
-    clearCompletedSlidingWindows(&pRuntimeEnv->swindowResInfo, pQuery->numOfOutputCols);
+    clearCompletedSlidingWindows(pRuntimeEnv);
     
     vnodeScanAllData(pRuntimeEnv);
     if (isQueryKilled(pQuery)) {

src/vnode/detail/src/vnodeSupertableQuery.c

@@ -77,7 +77,7 @@ static int32_t tabObjResultComparator(const void* p1, const void* p2, void* para
   STabObj* pNode1 = (STabObj*)p1;
   STabObj* pNode2 = (STabObj*)p2;
 
-  for (int32_t i = 0; i < pOrderDesc->orderIdx.numOfOrderedCols; ++i) {
+  for (int32_t i = 0; i < pOrderDesc->orderIdx.numOfCols; ++i) {
     int32_t colIdx = pOrderDesc->orderIdx.pData[i];
 
     char* f1 = NULL;
@@ -91,9 +91,11 @@ static int32_t tabObjResultComparator(const void* p1, const void* p2, void* para
       schema.type = TSDB_DATA_TYPE_BINARY;
       schema.bytes = TSDB_TABLE_ID_LEN;
     } else {
-      f1 = mgmtTableGetTag(pNode1, colIdx, NULL);
+      f1 = mgmtMeterGetTag(pNode1, colIdx, NULL);
-      f2 = mgmtTableGetTag(pNode2, colIdx, &schema);
+      f2 = mgmtMeterGetTag(pNode2, colIdx, &schema);
-      assert(schema.type == pOrderDesc->pTagSchema->pSchema[colIdx].type);
+      
+      SSchema* pSchema = getColumnModelSchema(pOrderDesc->pColumnModel, colIdx);
+      assert(schema.type == pSchema->type);
     }
 
     int32_t ret = doCompare(f1, f2, schema.type, schema.bytes);
@@ -116,7 +118,7 @@ static int32_t tabObjResultComparator(const void* p1, const void* p2, void* para
  * @param pOrderIndexInfo
  * @param numOfTags
  */
-static void mgmtUpdateOrderTagColIndex(SSuperTableMetaMsg* pSuperTableMetaMsg, int32_t tableIndex, tOrderIdx* pOrderIndexInfo,
+static void mgmtUpdateOrderTagColIndex(SMetricMetaMsg* pMetricMetaMsg, int32_t tableIndex, SColumnOrderInfo* pOrderIndexInfo,
      int32_t numOfTags) {
   SMetricMetaElemMsg* pElem = (SMetricMetaElemMsg*)((char*)pSuperTableMetaMsg + pSuperTableMetaMsg->metaElem[tableIndex]);
   SColIndexEx* groupColumnList = (SColIndexEx*)((char*)pSuperTableMetaMsg + pElem->groupbyTagColumnList);
@@ -130,7 +132,7 @@ static void mgmtUpdateOrderTagColIndex(SSuperTableMetaMsg* pSuperTableMetaMsg, i
     }
   }
   
-  pOrderIndexInfo->numOfOrderedCols = numOfGroupbyTags;
+  pOrderIndexInfo->numOfCols = numOfGroupbyTags;
 }
 
 // todo merge sort function with losertree used
@@ -150,14 +152,14 @@ void mgmtReorganizeMetersInMetricMeta(SSuperTableMetaMsg* pSuperTableMetaMsg, in
    */
   tOrderDescriptor* descriptor =
       (tOrderDescriptor*)calloc(1, sizeof(tOrderDescriptor) + sizeof(int32_t) * pElem->numOfGroupCols);
-  descriptor->pTagSchema = tCreateTagSchema(pTagSchema, pMetric->numOfTags);
+  descriptor->pColumnModel = createColumnModel(pTagSchema, pMetric->numOfTags, 1);
-  descriptor->orderIdx.numOfOrderedCols = pElem->numOfGroupCols;
+  descriptor->orderIdx.numOfCols = pElem->numOfGroupCols;
 
   int32_t* startPos = NULL;
   int32_t  numOfSubset = 1;
   
-  mgmtUpdateOrderTagColIndex(pSuperTableMetaMsg, tableIndex, &descriptor->orderIdx, pMetric->numOfTags);
+  mgmtUpdateOrderTagColIndex(pMetricMetaMsg, tableIndex, &descriptor->orderIdx, pMetric->numOfTags);
-  if (descriptor->orderIdx.numOfOrderedCols > 0) {
+  if (descriptor->orderIdx.numOfCols > 0) {
     tQSortEx(pRes->pRes, POINTER_BYTES, 0, pRes->num - 1, descriptor, tabObjResultComparator);
     startPos = calculateSubGroup(pRes->pRes, pRes->num, &numOfSubset, descriptor, tabObjResultComparator);
   } else {
@@ -173,7 +175,7 @@ void mgmtReorganizeMetersInMetricMeta(SSuperTableMetaMsg* pSuperTableMetaMsg, in
    */
   qsort(pRes->pRes, (size_t)pRes->num, POINTER_BYTES, tabObjVGIDComparator);
 
-  free(descriptor->pTagSchema);
+  free(descriptor->pColumnModel);
   free(descriptor);
   free(startPos);
 }
@@ -298,15 +300,15 @@ static void orderResult(SSuperTableMetaMsg* pSuperTableMetaMsg, tQueryResultset*
   STabObj* pMetric = mgmtGetTable(pElem->meterId);
   SSchema* pTagSchema = (SSchema*)(pMetric->schema + pMetric->numOfColumns * sizeof(SSchema));
 
-  descriptor->pTagSchema = tCreateTagSchema(pTagSchema, pMetric->numOfTags);
+  descriptor->pColumnModel = createColumnModel(pTagSchema, pMetric->numOfTags, 1);
 
   descriptor->orderIdx.pData[0] = colIndex;
-  descriptor->orderIdx.numOfOrderedCols = 1;
+  descriptor->orderIdx.numOfCols = 1;
 
   // sort results list
   tQSortEx(pRes->pRes, POINTER_BYTES, 0, pRes->num - 1, descriptor, tabObjResultComparator);
 
-  free(descriptor->pTagSchema);
+  free(descriptor->pColumnModel);
   free(descriptor);
 }
 

src/vnode/detail/src/vnodeTagMgmt.c

@@ -24,10 +24,10 @@
 #include "tast.h"
 #include "vnodeTagMgmt.h"
 
-#define GET_TAG_VAL_POINTER(s, col, sc, t) ((t *)(&((s)->tags[(sc)->colOffset[(col)]])))
+#define GET_TAG_VAL_POINTER(s, col, sc, t) ((t *)(&((s)->tags[getColumnModelOffset(sc, col)])))
 #define GET_TAG_VAL(s, col, sc, t) (*GET_TAG_VAL_POINTER(s, col, sc, t))
 
-static void tTagsPrints(SMeterSidExtInfo *pMeterInfo, tTagSchema *pSchema, tOrderIdx *pOrder);
+static void tTagsPrints(SMeterSidExtInfo *pMeterInfo, SColumnModel *pSchema, SColumnOrderInfo *pOrder);
 
 static void tSidSetDisplay(tSidSet *pSets);
 
@@ -65,7 +65,7 @@ int32_t meterSidComparator(const void *p1, const void *p2, void *param) {
   SMeterSidExtInfo *s1 = (SMeterSidExtInfo *)p1;
   SMeterSidExtInfo *s2 = (SMeterSidExtInfo *)p2;
 
-  for (int32_t i = 0; i < pOrderDesc->orderIdx.numOfOrderedCols; ++i) {
+  for (int32_t i = 0; i < pOrderDesc->orderIdx.numOfCols; ++i) {
     int32_t colIdx = pOrderDesc->orderIdx.pData[i];
 
     char *  f1 = NULL;
@@ -79,9 +79,9 @@ int32_t meterSidComparator(const void *p1, const void *p2, void *param) {
       type = TSDB_DATA_TYPE_BINARY;
       bytes = TSDB_METER_NAME_LEN;
     } else {
-      f1 = GET_TAG_VAL_POINTER(s1, colIdx, pOrderDesc->pTagSchema, char);
+      f1 = GET_TAG_VAL_POINTER(s1, colIdx, pOrderDesc->pColumnModel, char);
-      f2 = GET_TAG_VAL_POINTER(s2, colIdx, pOrderDesc->pTagSchema, char);
+      f2 = GET_TAG_VAL_POINTER(s2, colIdx, pOrderDesc->pColumnModel, char);
-      SSchema *pSchema = &pOrderDesc->pTagSchema->pSchema[colIdx];
+      SSchema *pSchema = getColumnModelSchema(pOrderDesc->pColumnModel, colIdx);
       type = pSchema->type;
       bytes = pSchema->bytes;
     }
@@ -116,9 +116,9 @@ static void median(void **pMeterSids, size_t size, int32_t s1, int32_t s2, tOrde
          compareFn(pMeterSids[s1], pMeterSids[s2], pOrderDesc) <= 0);
 
 #ifdef _DEBUG_VIEW
-  tTagsPrints(pMeterSids[s1], pOrderDesc->pTagSchema, &pOrderDesc->orderIdx);
+  tTagsPrints(pMeterSids[s1], pOrderDesc->pColumnModel, &pOrderDesc->orderIdx);
-  tTagsPrints(pMeterSids[midIdx], pOrderDesc->pTagSchema, &pOrderDesc->orderIdx);
+  tTagsPrints(pMeterSids[midIdx], pOrderDesc->pColumnModel, &pOrderDesc->orderIdx);
-  tTagsPrints(pMeterSids[s2], pOrderDesc->pTagSchema, &pOrderDesc->orderIdx);
+  tTagsPrints(pMeterSids[s2], pOrderDesc->pColumnModel, &pOrderDesc->orderIdx);
 #endif
 }
 
@@ -241,25 +241,6 @@ int32_t *calculateSubGroup(void **pSids, int32_t numOfMeters, int32_t *numOfSubs
   return starterPos;
 }
 
-tTagSchema *tCreateTagSchema(SSchema *pSchema, int32_t numOfTagCols) {
-  if (numOfTagCols == 0 || pSchema == NULL) {
-    return NULL;
-  }
-
-  tTagSchema *pTagSchema =
-      (tTagSchema *)calloc(1, sizeof(tTagSchema) + numOfTagCols * sizeof(int32_t) + sizeof(SSchema) * numOfTagCols);
-
-  pTagSchema->colOffset[0] = 0;
-  pTagSchema->numOfCols = numOfTagCols;
-  for (int32_t i = 1; i < numOfTagCols; ++i) {
-    pTagSchema->colOffset[i] = (pTagSchema->colOffset[i - 1] + pSchema[i - 1].bytes);
-  }
-
-  pTagSchema->pSchema = (SSchema *)&(pTagSchema->colOffset[numOfTagCols]);
-  memcpy(pTagSchema->pSchema, pSchema, sizeof(SSchema) * numOfTagCols);
-  return pTagSchema;
-}
-
 tSidSet *tSidSetCreate(struct SMeterSidExtInfo **pMeterSidExtInfo, int32_t numOfMeters, SSchema *pSchema,
                        int32_t numOfTags, SColIndexEx *colList, int32_t numOfCols) {
   tSidSet *pSidSet = (tSidSet *)calloc(1, sizeof(tSidSet) + numOfCols * sizeof(int16_t));
@@ -269,8 +250,8 @@ tSidSet *tSidSetCreate(struct SMeterSidExtInfo **pMeterSidExtInfo, int32_t numOf
 
   pSidSet->numOfSids = numOfMeters;
   pSidSet->pSids = pMeterSidExtInfo;
-  pSidSet->pTagSchema = tCreateTagSchema(pSchema, numOfTags);
+  pSidSet->pColumnModel = createColumnModel(pSchema, numOfTags, 1);
-  pSidSet->orderIdx.numOfOrderedCols = numOfCols;
+  pSidSet->orderIdx.numOfCols = numOfCols;
 
   /*
    * in case of "group by tbname,normal_col", the normal_col is ignored
@@ -282,7 +263,7 @@ tSidSet *tSidSetCreate(struct SMeterSidExtInfo **pMeterSidExtInfo, int32_t numOf
     }
   }
 
-  pSidSet->orderIdx.numOfOrderedCols = numOfTagCols;
+  pSidSet->orderIdx.numOfCols = numOfTagCols;
 
   pSidSet->starterPos = NULL;
   return pSidSet;
@@ -291,19 +272,19 @@ tSidSet *tSidSetCreate(struct SMeterSidExtInfo **pMeterSidExtInfo, int32_t numOf
 void tSidSetDestroy(tSidSet **pSets) {
   if ((*pSets) != NULL) {
     tfree((*pSets)->starterPos);
-    tfree((*pSets)->pTagSchema)(*pSets)->pSids = NULL;
+    tfree((*pSets)->pColumnModel)(*pSets)->pSids = NULL;
     tfree(*pSets);
   }
 }
 
-void tTagsPrints(SMeterSidExtInfo *pMeterInfo, tTagSchema *pSchema, tOrderIdx *pOrder) {
+void tTagsPrints(SMeterSidExtInfo *pMeterInfo, SColumnModel *pSchema, SColumnOrderInfo *pOrder) {
   if (pSchema == NULL) {
     return;
   }
 
   printf("sid: %-5d tags(", pMeterInfo->sid);
 
-  for (int32_t i = 0; i < pOrder->numOfOrderedCols; ++i) {
+  for (int32_t i = 0; i < pOrder->numOfCols; ++i) {
     int32_t colIndex = pOrder->pData[i];
 
     // it is the tbname column
@@ -312,7 +293,9 @@ void tTagsPrints(SMeterSidExtInfo *pMeterInfo, tTagSchema *pSchema, tOrderIdx *p
       continue;
     }
 
-    switch (pSchema->pSchema[colIndex].type) {
+    SSchema* s = getColumnModelSchema(pSchema, colIndex);
+    
+    switch (s->type) {
       case TSDB_DATA_TYPE_INT:
         printf("%d, ", GET_TAG_VAL(pMeterInfo, colIndex, pSchema, int32_t));
         break;
@@ -336,9 +319,9 @@ void tTagsPrints(SMeterSidExtInfo *pMeterInfo, tTagSchema *pSchema, tOrderIdx *p
         break;
       case TSDB_DATA_TYPE_NCHAR: {
         char *data = GET_TAG_VAL_POINTER(pMeterInfo, colIndex, pSchema, char);
-        char  buffer[512] = {0};
         
-        taosUcs4ToMbs(data, pSchema->pSchema[colIndex].bytes, buffer);
+        char  buffer[512] = {0};
+        taosUcs4ToMbs(data, s->bytes, buffer);
         printf("%s, ", buffer);
         break;
       }
@@ -370,16 +353,16 @@ static void UNUSED_FUNC tSidSetDisplay(tSidSet *pSets) {
 
     printf("the %d-th subgroup: \n", i + 1);
     for (int32_t j = s; j < e; ++j) {
-      tTagsPrints(pSets->pSids[j], pSets->pTagSchema, &pSets->orderIdx);
+      tTagsPrints(pSets->pSids[j], pSets->pColumnModel, &pSets->orderIdx);
     }
   }
 }
 
 void tSidSetSort(tSidSet *pSets) {
   pTrace("number of meters in sort: %d", pSets->numOfSids);
-  tOrderIdx *pOrderIdx = &pSets->orderIdx;
+  SColumnOrderInfo *pOrderIdx = &pSets->orderIdx;
 
-  if (pOrderIdx->numOfOrderedCols == 0 || pSets->numOfSids <= 1 || pSets->pTagSchema == NULL) { // no group by tags clause
+  if (pOrderIdx->numOfCols == 0 || pSets->numOfSids <= 1 || pSets->pColumnModel == NULL) { // no group by tags clause
     pSets->numOfSubSet = 1;
     pSets->starterPos = (int32_t *)malloc(sizeof(int32_t) * (pSets->numOfSubSet + 1));
     pSets->starterPos[0] = 0;
@@ -390,11 +373,11 @@ void tSidSetSort(tSidSet *pSets) {
 #endif
   } else {
     tOrderDescriptor *descriptor =
-        (tOrderDescriptor *)calloc(1, sizeof(tOrderDescriptor) + sizeof(int16_t) * pSets->orderIdx.numOfOrderedCols);
+        (tOrderDescriptor *)calloc(1, sizeof(tOrderDescriptor) + sizeof(int16_t) * pSets->orderIdx.numOfCols);
-    descriptor->pTagSchema = pSets->pTagSchema;
+    descriptor->pColumnModel = pSets->pColumnModel;
     descriptor->orderIdx = pSets->orderIdx;
 
-    memcpy(descriptor->orderIdx.pData, pOrderIdx->pData, sizeof(int16_t) * pSets->orderIdx.numOfOrderedCols);
+    memcpy(descriptor->orderIdx.pData, pOrderIdx->pData, sizeof(int16_t) * pSets->orderIdx.numOfCols);
 
     tQSortEx((void **)pSets->pSids, POINTER_BYTES, 0, pSets->numOfSids - 1, descriptor, meterSidComparator);
     pSets->starterPos =

src/vnode/detail/src/vnodeUtil.c

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