Merge branch '3.0' of github.com:taosdata/TDengine into szhou/select-tags-scan

cmake/apr-util_CMakeLists.txt.in

@@ -0,0 +1,19 @@
+# apr-util
+ExternalProject_Add(aprutil-1
+        URL https://dlcdn.apache.org//apr/apr-util-1.6.3.tar.gz
+        URL_HASH SHA256=2b74d8932703826862ca305b094eef2983c27b39d5c9414442e9976a9acf1983
+        DOWNLOAD_NO_PROGRESS 1
+        DOWNLOAD_DIR "${TD_CONTRIB_DIR}/deps-download"
+        #GIT_REPOSITORY https://github.com/apache/apr-util.git
+	#GIT_TAG 1.5.4
+        SOURCE_DIR "${TD_CONTRIB_DIR}/apr-util"
+        #BINARY_DIR ""
+        BUILD_IN_SOURCE TRUE
+        BUILD_ALWAYS 1
+        #UPDATE_COMMAND ""
+        CONFIGURE_COMMAND ./configure --prefix=$ENV{HOME}/.cos-local.1/ --with-apr=$ENV{HOME}/.cos-local.1
+        #CONFIGURE_COMMAND ./configure --with-apr=/usr/local/apr
+        BUILD_COMMAND make
+        INSTALL_COMMAND make install
+        TEST_COMMAND ""
+)

cmake/apr_CMakeLists.txt.in

@@ -0,0 +1,19 @@
+# apr
+ExternalProject_Add(apr-1
+        URL https://dlcdn.apache.org//apr/apr-1.7.4.tar.gz
+        URL_HASH SHA256=a4137dd82a185076fa50ba54232d920a17c6469c30b0876569e1c2a05ff311d9
+        DOWNLOAD_NO_PROGRESS 1
+        DOWNLOAD_DIR "${TD_CONTRIB_DIR}/deps-download"
+        #GIT_REPOSITORY https://github.com/apache/apr.git
+	#GIT_TAG 1.5.2
+        SOURCE_DIR "${TD_CONTRIB_DIR}/apr"
+        BUILD_IN_SOURCE TRUE
+        UPDATE_DISCONNECTED TRUE
+        BUILD_ALWAYS 1
+        #UPDATE_COMMAND ""
+        CONFIGURE_COMMAND ./configure --prefix=$ENV{HOME}/.cos-local.1/ --enable-shared=no
+        #CONFIGURE_COMMAND ./configure
+        BUILD_COMMAND make
+        INSTALL_COMMAND make install
+        TEST_COMMAND ""
+)

cmake/cmake.define

@@ -78,6 +78,12 @@ ELSE ()
     SET(TD_TAOS_TOOLS TRUE)
 ENDIF ()
 
+IF (${TD_WINDOWS})
+    SET(TAOS_LIB taos_static)
+ELSE ()
+    SET(TAOS_LIB taos)
+ENDIF ()
+
 IF (TD_WINDOWS)
     MESSAGE("${Yellow} set compiler flag for Windows! ${ColourReset}")
     SET(COMMON_FLAGS "/w /D_WIN32 /DWIN32 /Zi /MTd")

cmake/cmake.options

@@ -125,6 +125,16 @@ option(
     ON
 )
 
+IF(${TD_LINUX})
+
+option(
+    BUILD_WITH_COS
+    "If build with cos"
+    ON
+)
+
+ENDIF ()
+
 option(
     BUILD_WITH_SQLITE
     "If build with sqlite" 

cmake/cos_CMakeLists.txt.in

@@ -0,0 +1,12 @@
+# cos
+ExternalProject_Add(cos
+        GIT_REPOSITORY https://github.com/tencentyun/cos-c-sdk-v5.git
+	GIT_TAG v5.0.16
+        SOURCE_DIR "${TD_CONTRIB_DIR}/cos-c-sdk-v5"
+        BINARY_DIR ""
+        #BUILD_IN_SOURCE TRUE
+        CONFIGURE_COMMAND ""
+        BUILD_COMMAND ""
+        INSTALL_COMMAND ""
+        TEST_COMMAND ""
+)

cmake/curl_CMakeLists.txt.in

@@ -0,0 +1,17 @@
+# curl
+ExternalProject_Add(curl
+        URL  https://curl.se/download/curl-8.2.1.tar.gz
+        DOWNLOAD_NO_PROGRESS 1
+        DOWNLOAD_DIR "${TD_CONTRIB_DIR}/deps-download"
+        #GIT_REPOSITORY https://github.com/curl/curl.git
+        #GIT_TAG curl-7_88_1
+        SOURCE_DIR "${TD_CONTRIB_DIR}/curl"
+        BUILD_IN_SOURCE TRUE
+        BUILD_ALWAYS 1
+        #UPDATE_COMMAND ""
+        CONFIGURE_COMMAND ./configure --prefix=$ENV{HOME}/.cos-local.1 --without-ssl --enable-shared=no --disable-ldap --disable-ldaps --without-brotli
+        #CONFIGURE_COMMAND ./configure --without-ssl
+        BUILD_COMMAND make
+        INSTALL_COMMAND make install
+        TEST_COMMAND ""
+)

cmake/mxml_CMakeLists.txt.in

@@ -0,0 +1,14 @@
+# cos
+ExternalProject_Add(mxml
+        GIT_REPOSITORY https://github.com/michaelrsweet/mxml.git
+	GIT_TAG release-2.10
+        SOURCE_DIR "${TD_CONTRIB_DIR}/mxml"
+        #BINARY_DIR ""
+        BUILD_IN_SOURCE TRUE
+        #UPDATE_COMMAND ""
+        CONFIGURE_COMMAND ./configure --prefix=$ENV{HOME}/.cos-local.1 --enable-shared=no
+        #CONFIGURE_COMMAND ./configure
+        BUILD_COMMAND make
+        INSTALL_COMMAND make install
+        TEST_COMMAND ""
+)

contrib/CMakeLists.txt

@@ -6,6 +6,39 @@ function(cat IN_FILE OUT_FILE)
     file(APPEND ${OUT_FILE} "${CONTENTS}")
 endfunction(cat IN_FILE OUT_FILE)
 
+if(${TD_LINUX})
+
+set(CONTRIB_TMP_FILE3 "${CMAKE_BINARY_DIR}/deps_tmp_CMakeLists.txt.in3")
+configure_file("${TD_SUPPORT_DIR}/deps_CMakeLists.txt.in" ${CONTRIB_TMP_FILE3})
+
+if(${BUILD_WITH_COS})
+    file(MAKE_DIRECTORY $ENV{HOME}/.cos-local.1/)
+    cat("${TD_SUPPORT_DIR}/mxml_CMakeLists.txt.in" ${CONTRIB_TMP_FILE3})
+    cat("${TD_SUPPORT_DIR}/apr_CMakeLists.txt.in" ${CONTRIB_TMP_FILE3})
+    cat("${TD_SUPPORT_DIR}/curl_CMakeLists.txt.in" ${CONTRIB_TMP_FILE3})
+endif(${BUILD_WITH_COS})
+
+configure_file(${CONTRIB_TMP_FILE3} "${TD_CONTRIB_DIR}/deps-download/CMakeLists.txt")
+execute_process(COMMAND "${CMAKE_COMMAND}" -G "${CMAKE_GENERATOR}" .
+        WORKING_DIRECTORY "${TD_CONTRIB_DIR}/deps-download")
+execute_process(COMMAND "${CMAKE_COMMAND}" --build .
+        WORKING_DIRECTORY "${TD_CONTRIB_DIR}/deps-download")
+
+set(CONTRIB_TMP_FILE2 "${CMAKE_BINARY_DIR}/deps_tmp_CMakeLists.txt.in2")
+configure_file("${TD_SUPPORT_DIR}/deps_CMakeLists.txt.in" ${CONTRIB_TMP_FILE2})
+
+if(${BUILD_WITH_COS})
+    cat("${TD_SUPPORT_DIR}/apr-util_CMakeLists.txt.in" ${CONTRIB_TMP_FILE2})
+endif(${BUILD_WITH_COS})
+
+configure_file(${CONTRIB_TMP_FILE2} "${TD_CONTRIB_DIR}/deps-download/CMakeLists.txt")
+execute_process(COMMAND "${CMAKE_COMMAND}" -G "${CMAKE_GENERATOR}" .
+        WORKING_DIRECTORY "${TD_CONTRIB_DIR}/deps-download")
+execute_process(COMMAND "${CMAKE_COMMAND}" --build .
+        WORKING_DIRECTORY "${TD_CONTRIB_DIR}/deps-download")
+
+endif(${TD_LINUX})
+
 set(CONTRIB_TMP_FILE "${CMAKE_BINARY_DIR}/deps_tmp_CMakeLists.txt.in")
 configure_file("${TD_SUPPORT_DIR}/deps_CMakeLists.txt.in" ${CONTRIB_TMP_FILE})
 
@@ -122,6 +155,16 @@ if(${BUILD_WITH_SQLITE})
     cat("${TD_SUPPORT_DIR}/sqlite_CMakeLists.txt.in" ${CONTRIB_TMP_FILE})
 endif(${BUILD_WITH_SQLITE})
 
+# cos
+if(${BUILD_WITH_COS})
+    #cat("${TD_SUPPORT_DIR}/mxml_CMakeLists.txt.in" ${CONTRIB_TMP_FILE})
+    #cat("${TD_SUPPORT_DIR}/apr_CMakeLists.txt.in" ${CONTRIB_TMP_FILE})
+    #cat("${TD_SUPPORT_DIR}/apr-util_CMakeLists.txt.in" ${CONTRIB_TMP_FILE})
+    #cat("${TD_SUPPORT_DIR}/curl_CMakeLists.txt.in" ${CONTRIB_TMP_FILE})
+    cat("${TD_SUPPORT_DIR}/cos_CMakeLists.txt.in" ${CONTRIB_TMP_FILE})
+    add_definitions(-DUSE_COS)
+endif(${BUILD_WITH_COS})
+
 # lucene
 if(${BUILD_WITH_LUCENE})
     cat("${TD_SUPPORT_DIR}/lucene_CMakeLists.txt.in" ${CONTRIB_TMP_FILE})
@@ -347,6 +390,31 @@ if (${BUILD_WITH_ROCKSDB})
     endif()
 endif()
 
+# cos
+if(${BUILD_WITH_COS})
+  if(${TD_LINUX})
+    set(CMAKE_PREFIX_PATH $ENV{HOME}/.cos-local.1)
+    #ADD_DEFINITIONS(-DMINIXML_LIBRARY=${CMAKE_BINARY_DIR}/build/lib/libxml.a)
+    option(ENABLE_TEST "Enable the tests" OFF)
+    INCLUDE_DIRECTORIES($ENV{HOME}/.cos-local.1/include)
+    MESSAGE("$ENV{HOME}/.cos-local.1/include")
+
+    set(CMAKE_BUILD_TYPE debug)
+    set(ORIG_CMAKE_PROJECT_NAME ${CMAKE_PROJECT_NAME})
+    set(CMAKE_PROJECT_NAME cos_c_sdk)
+
+    add_subdirectory(cos-c-sdk-v5 EXCLUDE_FROM_ALL)
+    target_include_directories(
+      cos_c_sdk
+      PUBLIC $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/cos-c-sdk-v5/cos_c_sdk>
+      )
+
+    set(CMAKE_PROJECT_NAME ${ORIG_CMAKE_PROJECT_NAME})
+  else()
+
+  endif(${TD_LINUX})
+endif(${BUILD_WITH_COS})
+
 # lucene
 # To support build on ubuntu: sudo apt-get install libboost-all-dev
 if(${BUILD_WITH_LUCENE})

contrib/test/CMakeLists.txt

@@ -3,6 +3,11 @@ if(${BUILD_WITH_ROCKSDB})
     add_subdirectory(rocksdb)
 endif(${BUILD_WITH_ROCKSDB})
 
+# cos
+if(${BUILD_WITH_COS})
+    add_subdirectory(cos)
+endif(${BUILD_WITH_COS})
+
 if(${BUILD_WITH_LUCENE})
     add_subdirectory(lucene)
 endif(${BUILD_WITH_LUCENE})

contrib/test/cos/CMakeLists.txt

@@ -0,0 +1,49 @@
+add_executable(cosTest "")
+target_sources(cosTest
+  PRIVATE
+  "${CMAKE_CURRENT_SOURCE_DIR}/main.c"
+  )
+
+#find_path(APR_INCLUDE_DIR apr-1/apr_time.h)
+#find_path(APR_UTIL_INCLUDE_DIR apr/include/apr-1/apr_md5.h)
+#find_path(MINIXML_INCLUDE_DIR mxml.h)
+#find_path(CURL_INCLUDE_DIR curl/curl.h)
+
+#include_directories (${MINIXML_INCLUDE_DIR})
+#include_directories (${CURL_INCLUDE_DIR})
+FIND_PROGRAM(APR_CONFIG_BIN NAMES apr-config apr-1-config PATHS /usr/bin /usr/local/bin /usr/local/apr/bin/)
+#FIND_PROGRAM(APU_CONFIG_BIN NAMES apu-config apu-1-config PATHS /usr/bin /usr/local/bin /usr/local/apr/bin/)
+
+IF (APR_CONFIG_BIN)
+  EXECUTE_PROCESS(
+    COMMAND ${APR_CONFIG_BIN} --includedir
+    OUTPUT_VARIABLE APR_INCLUDE_DIR
+    OUTPUT_STRIP_TRAILING_WHITESPACE
+    )
+ENDIF()
+#IF (APU_CONFIG_BIN)
+#  EXECUTE_PROCESS(
+#    COMMAND ${APU_CONFIG_BIN} --includedir
+#    OUTPUT_VARIABLE APR_UTIL_INCLUDE_DIR
+#    OUTPUT_STRIP_TRAILING_WHITESPACE
+#    )
+#ENDIF()
+
+include_directories (${APR_INCLUDE_DIR})
+#include_directories (${APR_UTIL_INCLUDE_DIR})
+
+target_include_directories(
+  cosTest
+  PUBLIC "${TD_SOURCE_DIR}/contrib/cos-c-sdk-v5/cos_c_sdk"
+  )
+
+#find_library(APR_LIBRARY apr-1 PATHS /usr/local/apr/lib/)
+#find_library(APR_UTIL_LIBRARY aprutil-1 PATHS /usr/local/apr/lib/)
+#find_library(MINIXML_LIBRARY mxml)
+#find_library(CURL_LIBRARY curl)
+
+target_link_libraries(cosTest cos_c_sdk)
+target_link_libraries(cosTest apr-1})
+target_link_libraries(cosTest aprutil-1})
+target_link_libraries(cosTest mxml)
+target_link_libraries(cosTest curl)

docs/en/07-develop/07-tmq.mdx

@@ -23,7 +23,20 @@ By subscribing to a topic, a consumer can obtain the latest data in that topic i
 
 To implement these features, TDengine indexes its write-ahead log (WAL) file for fast random access and provides configurable methods for replacing and retaining this file. You can define a retention period and size for this file. For information, see the CREATE DATABASE statement. In this way, the WAL file is transformed into a persistent storage engine that remembers the order in which events occur. However, note that configuring an overly long retention period for your WAL files makes database compression inefficient. TDengine then uses the WAL file instead of the time-series database as its storage engine for queries in the form of topics. TDengine reads the data from the WAL file; uses a unified query engine instance to perform filtering, transformations, and other operations; and finally pushes the data to consumers.
 
-Tips: Data subscription is to consume data from the wal. If some wal files are deleted according to WAL retention policy, the deleted data can't be consumed any more. So you need to set a reasonable value for parameter `WAL_RETENTION_PERIOD` or `WAL_RETENTION_SIZE` when creating the database and make sure your application consume the data in a timely way to make sure there is no data loss. This behavior is similar to Kafka and other widely used message queue products.
+Tips:(c interface for example)
+1. A consumption group consumes all data under the same topic, and different consumption groups are independent of each other;
+2. A consumption group consumes all vgroups of the same topic, which can be composed of multiple consumers, but a vgroup is only consumed by one consumer. If the number of consumers exceeds the number of vgroups, the excess consumers do not consume data;
+3. On the server side, only one offset is saved for each vgroup, and the offsets for each vgroup are monotonically increasing, but not necessarily continuous. There is no correlation between the offsets of various vgroups;
+4. Each poll server will return a result block, which belongs to a vgroup and may contain data from multiple versions of wal. This block can be accessed through tmq_get_vgroup_offset. The offset interface obtains the offset of the first record in the block;
+5. If a consumer group has never committed an offset, when its member consumers restart and pull data again, they start consuming from the set value of the parameter auto.offset.reset; In a consumer lifecycle, the client locally records the offset of the most recent pull data and will not pull duplicate data;
+6. If a consumer terminates abnormally (without calling tmq_close), they need to wait for about 12 seconds to trigger their consumer group rebalance. The consumer's status on the server will change to LOST, and after about 1 day, the consumer will be automatically deleted; Exit normally, and after exiting, the consumer will be deleted; Add a new consumer, wait for about 2 seconds to trigger Rebalance, and the consumer's status on the server will change to ready;
+7. The consumer group Rebalance will reassign Vgroups to all consumer members in the ready state of the group, and consumers can only assign/see/commit/poll operations to the Vgroups they are responsible for;
+8. Consumers can tmq_position to obtain the offset of the current consumption, seek to the specified offset, and consume again;
+9. Seek points the position to the specified offset without executing the commit operation. Once the seek is successful, it can poll the specified offset and subsequent data;
+10. Before the seek operation, tmq must be call tmq_get_topic_assignment, The assignment interface obtains the vgroup ID and offset range of the consumer. The seek operation will detect whether the vgroup ID and offset are legal, and if they are illegal, an error will be reported;
+11. Due to the existence of a WAL expiration deletion mechanism, even if the seek operation is successful, it is possible that the offset has expired when polling data. If the offset of poll is less than the WAL minimum version number, it will be consumed from the WAL minimum version number;
+12. The tmq_get_vgroup_offset interface obtains the offset of the first data in the result block where the record is located. When seeking to this offset, it will consume all the data in this block. Refer to point four;
+13. Data subscription is to consume data from the wal. If some wal files are deleted according to WAL retention policy, the deleted data can't be consumed any more. So you need to set a reasonable value for parameter `WAL_RETENTION_PERIOD` or `WAL_RETENTION_SIZE` when creating the database and make sure your application consume the data in a timely way to make sure there is no data loss. This behavior is similar to Kafka and other widely used message queue products.
 
 ## Data Schema and API
 
@@ -37,35 +50,54 @@ typedef struct tmq_t      tmq_t;
     typedef struct tmq_conf_t tmq_conf_t;
     typedef struct tmq_list_t tmq_list_t;
 
-typedef void(tmq_commit_cb(tmq_t *, int32_t code, void *param));
+    typedef void(tmq_commit_cb(tmq_t *tmq, int32_t code, void *param));
 
-DLL_EXPORT tmq_list_t *tmq_list_new();
+    typedef enum tmq_conf_res_t {
-DLL_EXPORT int32_t     tmq_list_append(tmq_list_t *, const char *);
-DLL_EXPORT void        tmq_list_destroy(tmq_list_t *);
-DLL_EXPORT tmq_t *tmq_consumer_new(tmq_conf_t *conf, char *errstr, int32_t errstrLen);
-DLL_EXPORT const char *tmq_err2str(int32_t code);
-
-DLL_EXPORT int32_t   tmq_subscribe(tmq_t *tmq, const tmq_list_t *topic_list);
-DLL_EXPORT int32_t   tmq_unsubscribe(tmq_t *tmq);
-DLL_EXPORT TAOS_RES *tmq_consumer_poll(tmq_t *tmq, int64_t timeout);
-DLL_EXPORT int32_t   tmq_consumer_close(tmq_t *tmq);
-DLL_EXPORT int32_t   tmq_commit_sync(tmq_t *tmq, const TAOS_RES *msg);
-DLL_EXPORT void      tmq_commit_async(tmq_t *tmq, const TAOS_RES *msg, tmq_commit_cb *cb, void *param);
-
-enum tmq_conf_res_t {
     TMQ_CONF_UNKNOWN = -2,
     TMQ_CONF_INVALID = -1,
     TMQ_CONF_OK = 0,
-};
+} tmq_conf_res_t;
-typedef enum tmq_conf_res_t tmq_conf_res_t;
+
+    typedef struct tmq_topic_assignment {
+    int32_t vgId;
+    int64_t currentOffset;
+    int64_t begin;
+    int64_t end;
+} tmq_topic_assignment;
 
     DLL_EXPORT tmq_conf_t    *tmq_conf_new();
     DLL_EXPORT tmq_conf_res_t tmq_conf_set(tmq_conf_t *conf, const char *key, const char *value);
     DLL_EXPORT void           tmq_conf_destroy(tmq_conf_t *conf);
     DLL_EXPORT void           tmq_conf_set_auto_commit_cb(tmq_conf_t *conf, tmq_commit_cb *cb, void *param);
-```
 
-For more information, see [C/C++ Connector](/reference/connector/cpp).
+    DLL_EXPORT tmq_list_t *tmq_list_new();
+    DLL_EXPORT int32_t     tmq_list_append(tmq_list_t *, const char *);
+    DLL_EXPORT void        tmq_list_destroy(tmq_list_t *);
+    DLL_EXPORT int32_t     tmq_list_get_size(const tmq_list_t *);
+    DLL_EXPORT char      **tmq_list_to_c_array(const tmq_list_t *);
+
+    DLL_EXPORT tmq_t    *tmq_consumer_new(tmq_conf_t *conf, char *errstr, int32_t errstrLen);
+    DLL_EXPORT int32_t   tmq_subscribe(tmq_t *tmq, const tmq_list_t *topic_list);
+    DLL_EXPORT int32_t   tmq_unsubscribe(tmq_t *tmq);
+    DLL_EXPORT int32_t   tmq_subscription(tmq_t *tmq, tmq_list_t **topics);
+    DLL_EXPORT TAOS_RES *tmq_consumer_poll(tmq_t *tmq, int64_t timeout);
+    DLL_EXPORT int32_t   tmq_consumer_close(tmq_t *tmq);
+    DLL_EXPORT int32_t   tmq_commit_sync(tmq_t *tmq, const TAOS_RES *msg);
+    DLL_EXPORT void      tmq_commit_async(tmq_t *tmq, const TAOS_RES *msg, tmq_commit_cb *cb, void *param);
+    DLL_EXPORT int32_t   tmq_commit_offset_sync(tmq_t *tmq, const char *pTopicName, int32_t vgId, int64_t offset);
+    DLL_EXPORT void      tmq_commit_offset_async(tmq_t *tmq, const char *pTopicName, int32_t vgId, int64_t offset, tmq_commit_cb *cb, void *param);
+    DLL_EXPORT int32_t   tmq_get_topic_assignment(tmq_t *tmq, const char *pTopicName, tmq_topic_assignment **assignment,int32_t *numOfAssignment);
+    DLL_EXPORT void      tmq_free_assignment(tmq_topic_assignment* pAssignment);
+    DLL_EXPORT int32_t   tmq_offset_seek(tmq_t *tmq, const char *pTopicName, int32_t vgId, int64_t offset);
+    DLL_EXPORT int64_t   tmq_position(tmq_t *tmq, const char *pTopicName, int32_t vgId);
+    DLL_EXPORT int64_t   tmq_committed(tmq_t *tmq, const char *pTopicName, int32_t vgId);
+
+    DLL_EXPORT const char *tmq_get_topic_name(TAOS_RES *res);
+    DLL_EXPORT const char *tmq_get_db_name(TAOS_RES *res);
+    DLL_EXPORT int32_t     tmq_get_vgroup_id(TAOS_RES *res);
+    DLL_EXPORT int64_t     tmq_get_vgroup_offset(TAOS_RES* res);
+    DLL_EXPORT const char *tmq_err2str(int32_t code);DLL_EXPORT void           tmq_conf_set_auto_commit_cb(tmq_conf_t *conf, tmq_commit_cb *cb, void *param);
+```
 
 The following example is based on the smart meter table described in Data Models. For complete sample code, see the C language section below.
 

docs/en/14-reference/03-connector/04-java.mdx

@@ -1091,8 +1091,6 @@ public abstract class ConsumerLoop {
         config.setProperty("client.id", "1");
         config.setProperty("value.deserializer", "com.taosdata.jdbc.tmq.ConsumerTest.ConsumerLoop$ResultDeserializer");
         config.setProperty("value.deserializer.encoding", "UTF-8");
-        config.setProperty("experimental.snapshot.enable", "true");
-
 
         this.consumer = new TaosConsumer<>(config);
         this.topics = Collections.singletonList("topic_speed");
@@ -1176,7 +1174,6 @@ public abstract class ConsumerLoop {
         config.setProperty("client.id", "1");
         config.setProperty("value.deserializer", "com.taosdata.jdbc.tmq.ConsumerTest.ConsumerLoop$ResultDeserializer");
         config.setProperty("value.deserializer.encoding", "UTF-8");
-        config.setProperty("experimental.snapshot.enable", "true");
 
         this.consumer = new TaosConsumer<>(config);
         this.topics = Collections.singletonList("topic_speed");

docs/examples/c/tmq_example.c

@@ -227,12 +227,6 @@ tmq_t* build_consumer() {
     return NULL;
   }
 
-  code = tmq_conf_set(conf, "experimental.snapshot.enable", "false");
-  if (TMQ_CONF_OK != code) {
-    tmq_conf_destroy(conf);
-    return NULL;
-  }
-
   tmq_conf_set_auto_commit_cb(conf, tmq_commit_cb_print, NULL);
 
   tmq_t* tmq = tmq_consumer_new(conf, NULL, 0);

docs/examples/go/sub/main.go

@@ -35,7 +35,6 @@ func main() {
 		"td.connect.port":              "6030",
 		"client.id":                    "test_tmq_client",
 		"enable.auto.commit":           "false",
-		"experimental.snapshot.enable": "true",
 		"msg.with.table.name":          "true",
 	})
 	if err != nil {

docs/zh/07-develop/07-tmq.mdx

@@ -25,7 +25,20 @@ import CDemo from "./_sub_c.mdx";
 
 本文档不对消息队列本身的基础知识做介绍，如果需要了解，请自行搜索。
 
-注意：数据订阅是从 WAL 消费数据，如果一些 WAL 文件被基于 WAL 保留策略删除，则已经删除的 WAL 文件中的数据就无法再消费到。需要根据业务需要在创建数据库时合理设置 `WAL_RETENTION_PERIOD` 或 `WAL_RETENTION_SIZE` ，并确保应用及时消费数据，这样才不会产生数据丢失的现象。数据订阅的行为与 Kafka 等广泛使用的消息队列类产品的行为相似。
+说明（以c接口为例）：
+1. 一个消费组消费同一个topic下的所有数据，不同消费组之间相互独立；
+2. 一个消费组消费同一个topic所有的vgroup，消费组可由多个消费者组成，但一个vgroup仅被一个消费者消费，如果消费者数量超过了vgroup数量，多余的消费者不消费数据；
+3. 在服务端每个vgroup仅保存一个offset，每个vgroup的offset是单调递增的，但不一定连续。各个vgroup的offset之间没有关联；
+4. 每次poll服务端会返回一个结果block，该block属于一个vgroup，可能包含多个wal版本的数据，可以通过 tmq_get_vgroup_offset 接口获得是该block第一条记录的offset；
+5. 一个消费组如果从未commit过offset，当其成员消费者重启重新拉取数据时，均从参数auto.offset.reset设定值开始消费；在一个消费者生命周期中，客户端本地记录了最近一次拉取数据的offset，不会拉取重复数据；
+6. 消费者如果异常终止（没有调用tmq_close），需等约12秒后触发其所属消费组rebalance，该消费者在服务端状态变为LOST，约1天后该消费者自动被删除；正常退出，退出后就会删除消费者；新增消费者，需等约2秒触发rebalance，该消费者在服务端状态变为ready；
+7. 消费组rebalance会对该组所有ready状态的消费者成员重新进行vgroup分配，消费者仅能对自己负责的vgroup进行assignment/seek/commit/poll操作；
+8. 消费者可利用 tmq_position 获得当前消费的offset，并seek到指定offset，重新消费；
+9. seek将position指向指定offset，不执行commit操作，一旦seek成功，可poll拉取指定offset及以后的数据；
+10. seek 操作之前须调用 tmq_get_topic_assignment 接口获取该consumer的vgroup ID和offset范围。seek 操作会检测vgroup ID 和 offset是否合法，如非法将报错；
+11. tmq_get_vgroup_offset接口获取的是记录所在结果block块里的第一条数据的offset，当seek至该offset时，将消费到这个block里的全部数据。参见第四点；
+12. 由于存在 WAL 过期删除机制，即使seek 操作成功，poll数据时有可能offset已失效。如果poll 的offset 小于 WAL 最小版本号，将会从WAL最小版本号消费；
+13. 数据订阅是从 WAL 消费数据，如果一些 WAL 文件被基于 WAL 保留策略删除，则已经删除的 WAL 文件中的数据就无法再消费到。需要根据业务需要在创建数据库时合理设置 `WAL_RETENTION_PERIOD` 或 `WAL_RETENTION_SIZE` ，并确保应用及时消费数据，这样才不会产生数据丢失的现象。数据订阅的行为与 Kafka 等广泛使用的消息队列类产品的行为相似；
 
 ## 主要数据结构和 API
 
@@ -39,35 +52,56 @@ typedef struct tmq_t      tmq_t;
     typedef struct tmq_conf_t tmq_conf_t;
     typedef struct tmq_list_t tmq_list_t;
 
-typedef void(tmq_commit_cb(tmq_t *, int32_t code, void *param));
+    typedef void(tmq_commit_cb(tmq_t *tmq, int32_t code, void *param));
 
-DLL_EXPORT tmq_list_t *tmq_list_new();
+    typedef enum tmq_conf_res_t {
-DLL_EXPORT int32_t     tmq_list_append(tmq_list_t *, const char *);
-DLL_EXPORT void        tmq_list_destroy(tmq_list_t *);
-DLL_EXPORT tmq_t *tmq_consumer_new(tmq_conf_t *conf, char *errstr, int32_t errstrLen);
-DLL_EXPORT const char *tmq_err2str(int32_t code);
-
-DLL_EXPORT int32_t   tmq_subscribe(tmq_t *tmq, const tmq_list_t *topic_list);
-DLL_EXPORT int32_t   tmq_unsubscribe(tmq_t *tmq);
-DLL_EXPORT TAOS_RES *tmq_consumer_poll(tmq_t *tmq, int64_t timeout);
-DLL_EXPORT int32_t   tmq_consumer_close(tmq_t *tmq);
-DLL_EXPORT int32_t   tmq_commit_sync(tmq_t *tmq, const TAOS_RES *msg);
-DLL_EXPORT void      tmq_commit_async(tmq_t *tmq, const TAOS_RES *msg, tmq_commit_cb *cb, void *param);
-
-enum tmq_conf_res_t {
     TMQ_CONF_UNKNOWN = -2,
     TMQ_CONF_INVALID = -1,
     TMQ_CONF_OK = 0,
-};
+} tmq_conf_res_t;
-typedef enum tmq_conf_res_t tmq_conf_res_t;
+
+    typedef struct tmq_topic_assignment {
+    int32_t vgId;
+    int64_t currentOffset;
+    int64_t begin;
+    int64_t end;
+} tmq_topic_assignment;
 
     DLL_EXPORT tmq_conf_t    *tmq_conf_new();
     DLL_EXPORT tmq_conf_res_t tmq_conf_set(tmq_conf_t *conf, const char *key, const char *value);
     DLL_EXPORT void           tmq_conf_destroy(tmq_conf_t *conf);
     DLL_EXPORT void           tmq_conf_set_auto_commit_cb(tmq_conf_t *conf, tmq_commit_cb *cb, void *param);
+
+    DLL_EXPORT tmq_list_t *tmq_list_new();
+    DLL_EXPORT int32_t     tmq_list_append(tmq_list_t *, const char *);
+    DLL_EXPORT void        tmq_list_destroy(tmq_list_t *);
+    DLL_EXPORT int32_t     tmq_list_get_size(const tmq_list_t *);
+    DLL_EXPORT char      **tmq_list_to_c_array(const tmq_list_t *);
+
+    DLL_EXPORT tmq_t    *tmq_consumer_new(tmq_conf_t *conf, char *errstr, int32_t errstrLen);
+    DLL_EXPORT int32_t   tmq_subscribe(tmq_t *tmq, const tmq_list_t *topic_list);
+    DLL_EXPORT int32_t   tmq_unsubscribe(tmq_t *tmq);
+    DLL_EXPORT int32_t   tmq_subscription(tmq_t *tmq, tmq_list_t **topics);
+    DLL_EXPORT TAOS_RES *tmq_consumer_poll(tmq_t *tmq, int64_t timeout);
+    DLL_EXPORT int32_t   tmq_consumer_close(tmq_t *tmq);
+    DLL_EXPORT int32_t   tmq_commit_sync(tmq_t *tmq, const TAOS_RES *msg);
+    DLL_EXPORT void      tmq_commit_async(tmq_t *tmq, const TAOS_RES *msg, tmq_commit_cb *cb, void *param);
+    DLL_EXPORT int32_t   tmq_commit_offset_sync(tmq_t *tmq, const char *pTopicName, int32_t vgId, int64_t offset);
+    DLL_EXPORT void      tmq_commit_offset_async(tmq_t *tmq, const char *pTopicName, int32_t vgId, int64_t offset, tmq_commit_cb *cb, void *param);
+    DLL_EXPORT int32_t   tmq_get_topic_assignment(tmq_t *tmq, const char *pTopicName, tmq_topic_assignment **assignment,int32_t *numOfAssignment);
+    DLL_EXPORT void      tmq_free_assignment(tmq_topic_assignment* pAssignment);
+    DLL_EXPORT int32_t   tmq_offset_seek(tmq_t *tmq, const char *pTopicName, int32_t vgId, int64_t offset);
+    DLL_EXPORT int64_t   tmq_position(tmq_t *tmq, const char *pTopicName, int32_t vgId);
+    DLL_EXPORT int64_t   tmq_committed(tmq_t *tmq, const char *pTopicName, int32_t vgId);
+
+    DLL_EXPORT const char *tmq_get_topic_name(TAOS_RES *res);
+    DLL_EXPORT const char *tmq_get_db_name(TAOS_RES *res);
+    DLL_EXPORT int32_t     tmq_get_vgroup_id(TAOS_RES *res);
+    DLL_EXPORT int64_t     tmq_get_vgroup_offset(TAOS_RES* res);
+    DLL_EXPORT const char *tmq_err2str(int32_t code);DLL_EXPORT void           tmq_conf_set_auto_commit_cb(tmq_conf_t *conf, tmq_commit_cb *cb, void *param);
 ```
 
-这些 API 的文档请见 [C/C++ Connector](../../connector/cpp)，下面介绍一下它们的具体用法（超级表和子表结构请参考“数据建模”一节），完整的示例代码请见下面 C 语言的示例代码。
+下面介绍一下它们的具体用法（超级表和子表结构请参考“数据建模”一节），完整的示例代码请见下面 C 语言的示例代码。
 
 </TabItem>
 <TabItem value="java" label="Java">

examples/c/tmq.c

@@ -228,11 +228,6 @@ tmq_t* build_consumer() {
     tmq_conf_destroy(conf);
     return NULL;
   }
-  code = tmq_conf_set(conf, "experimental.snapshot.enable", "false");
-  if (TMQ_CONF_OK != code) {
-    tmq_conf_destroy(conf);
-    return NULL;
-  }
 
   tmq_conf_set_auto_commit_cb(conf, tmq_commit_cb_print, NULL);
   tmq = tmq_consumer_new(conf, NULL, 0);

include/client/taos.h

@@ -260,7 +260,25 @@ typedef struct tmq_t      tmq_t;
 typedef struct tmq_conf_t tmq_conf_t;
 typedef struct tmq_list_t tmq_list_t;
 
-typedef void(tmq_commit_cb(tmq_t *, int32_t code, void *param));
+typedef void(tmq_commit_cb(tmq_t *tmq, int32_t code, void *param));
+
+typedef enum tmq_conf_res_t {
+  TMQ_CONF_UNKNOWN = -2,
+  TMQ_CONF_INVALID = -1,
+  TMQ_CONF_OK = 0,
+} tmq_conf_res_t;
+
+typedef struct tmq_topic_assignment {
+  int32_t vgId;
+  int64_t currentOffset;
+  int64_t begin;
+  int64_t end;
+} tmq_topic_assignment;
+
+DLL_EXPORT tmq_conf_t    *tmq_conf_new();
+DLL_EXPORT tmq_conf_res_t tmq_conf_set(tmq_conf_t *conf, const char *key, const char *value);
+DLL_EXPORT void           tmq_conf_destroy(tmq_conf_t *conf);
+DLL_EXPORT void           tmq_conf_set_auto_commit_cb(tmq_conf_t *conf, tmq_commit_cb *cb, void *param);
 
 DLL_EXPORT tmq_list_t *tmq_list_new();
 DLL_EXPORT int32_t     tmq_list_append(tmq_list_t *, const char *);
@@ -269,17 +287,6 @@ DLL_EXPORT int32_t     tmq_list_get_size(const tmq_list_t *);
 DLL_EXPORT char      **tmq_list_to_c_array(const tmq_list_t *);
 
 DLL_EXPORT tmq_t    *tmq_consumer_new(tmq_conf_t *conf, char *errstr, int32_t errstrLen);
-
-DLL_EXPORT const char *tmq_err2str(int32_t code);
-
-/* ------------------------TMQ CONSUMER INTERFACE------------------------ */
-typedef struct tmq_topic_assignment {
-  int32_t vgId;
-  int64_t currentOffset;
-  int64_t begin;
-  int64_t end;
-} tmq_topic_assignment;
-
 DLL_EXPORT int32_t   tmq_subscribe(tmq_t *tmq, const tmq_list_t *topic_list);
 DLL_EXPORT int32_t   tmq_unsubscribe(tmq_t *tmq);
 DLL_EXPORT int32_t   tmq_subscription(tmq_t *tmq, tmq_list_t **topics);
@@ -289,34 +296,17 @@ DLL_EXPORT int32_t   tmq_commit_sync(tmq_t *tmq, const TAOS_RES *msg);
 DLL_EXPORT void      tmq_commit_async(tmq_t *tmq, const TAOS_RES *msg, tmq_commit_cb *cb, void *param);
 DLL_EXPORT int32_t   tmq_commit_offset_sync(tmq_t *tmq, const char *pTopicName, int32_t vgId, int64_t offset);
 DLL_EXPORT void      tmq_commit_offset_async(tmq_t *tmq, const char *pTopicName, int32_t vgId, int64_t offset, tmq_commit_cb *cb, void *param);
-DLL_EXPORT int32_t   tmq_get_topic_assignment(tmq_t *tmq, const char *pTopicName, tmq_topic_assignment **assignment,
+DLL_EXPORT int32_t   tmq_get_topic_assignment(tmq_t *tmq, const char *pTopicName, tmq_topic_assignment **assignment,int32_t *numOfAssignment);
-                                              int32_t *numOfAssignment);
 DLL_EXPORT void      tmq_free_assignment(tmq_topic_assignment* pAssignment);
 DLL_EXPORT int32_t   tmq_offset_seek(tmq_t *tmq, const char *pTopicName, int32_t vgId, int64_t offset);
+DLL_EXPORT int64_t   tmq_position(tmq_t *tmq, const char *pTopicName, int32_t vgId);
+DLL_EXPORT int64_t   tmq_committed(tmq_t *tmq, const char *pTopicName, int32_t vgId);
 
 DLL_EXPORT const char *tmq_get_topic_name(TAOS_RES *res);
 DLL_EXPORT const char *tmq_get_db_name(TAOS_RES *res);
 DLL_EXPORT int32_t     tmq_get_vgroup_id(TAOS_RES *res);
 DLL_EXPORT int64_t     tmq_get_vgroup_offset(TAOS_RES* res);
-DLL_EXPORT int64_t     tmq_position(tmq_t *tmq, const char *pTopicName, int32_t vgId);
+DLL_EXPORT const char *tmq_err2str(int32_t code);
-DLL_EXPORT int64_t     tmq_committed(tmq_t *tmq, const char *pTopicName, int32_t vgId);
-
-/* ----------------------TMQ CONFIGURATION INTERFACE---------------------- */
-
-enum tmq_conf_res_t {
-  TMQ_CONF_UNKNOWN = -2,
-  TMQ_CONF_INVALID = -1,
-  TMQ_CONF_OK = 0,
-};
-
-typedef enum tmq_conf_res_t tmq_conf_res_t;
-
-DLL_EXPORT tmq_conf_t    *tmq_conf_new();
-DLL_EXPORT tmq_conf_res_t tmq_conf_set(tmq_conf_t *conf, const char *key, const char *value);
-DLL_EXPORT void           tmq_conf_destroy(tmq_conf_t *conf);
-DLL_EXPORT void           tmq_conf_set_auto_commit_cb(tmq_conf_t *conf, tmq_commit_cb *cb, void *param);
-
-/* -------------------------TMQ MSG HANDLE INTERFACE---------------------- */
 
 /* ------------------------------ TAOSX -----------------------------------*/
 // note: following apis are unstable

include/libs/executor/storageapi.h

@@ -98,6 +98,16 @@ typedef struct SMTbCursor {
   int8_t      paused;
 } SMTbCursor;
 
+typedef struct SMCtbCursor {
+  SMeta   *pMeta;
+  void    *pCur;
+  tb_uid_t suid;
+  void    *pKey;
+  void    *pVal;
+  int      kLen;
+  int      vLen;
+} SMCtbCursor;
+
 typedef struct SRowBuffPos {
   void* pRowBuff;
   void* pKey;
@@ -278,13 +288,15 @@ typedef struct SStoreMeta {
   void (*getBasicInfo)(void* pVnode, const char** dbname, int32_t* vgId, int64_t* numOfTables,
                        int64_t* numOfNormalTables);  // vnodeGetInfo(void *pVnode, const char **dbname, int32_t *vgId) &
                                                      // metaGetTbNum(SMeta *pMeta) & metaGetNtbNum(SMeta *pMeta);
-
   int64_t (*getNumOfRowsInMem)(void* pVnode);
   /**
 int32_t vnodeGetCtbIdList(void *pVnode, int64_t suid, SArray *list);
 int32_t vnodeGetCtbIdListByFilter(void *pVnode, int64_t suid, SArray *list, bool (*filter)(void *arg), void *arg);
 int32_t vnodeGetStbIdList(void *pVnode, int64_t suid, SArray *list);
  */
+  SMCtbCursor* (*openCtbCursor)(void *pVnode, tb_uid_t uid, int lock);
+  void         (*closeCtbCursor)(SMCtbCursor *pCtbCur, int lock);
+  tb_uid_t     (*ctbCursorNext)(SMCtbCursor* pCur);
 } SStoreMeta;
 
 typedef struct SStoreMetaReader {

include/libs/nodes/nodes.h

@@ -169,7 +169,7 @@ typedef enum ENodeType {
   QUERY_NODE_REVOKE_STMT,
   QUERY_NODE_SHOW_DNODES_STMT,
   QUERY_NODE_SHOW_MNODES_STMT,
-  QUERY_NODE_SHOW_MODULES_STMT,
+//  QUERY_NODE_SHOW_MODULES_STMT,
   QUERY_NODE_SHOW_QNODES_STMT,
   QUERY_NODE_SHOW_SNODES_STMT,
   QUERY_NODE_SHOW_BNODES_STMT,

include/libs/nodes/plannodes.h

@@ -107,6 +107,7 @@ typedef struct SScanLogicNode {
   bool          sortPrimaryKey;
   bool          igLastNull;
   bool          groupOrderScan;
+  bool          onlyMetaCtbIdx; // for tag scan with no tbname
 } SScanLogicNode;
 
 typedef struct SJoinLogicNode {
@@ -334,7 +335,11 @@ typedef struct SScanPhysiNode {
   bool       groupOrderScan;
 } SScanPhysiNode;
 
-typedef SScanPhysiNode STagScanPhysiNode;
+typedef struct STagScanPhysiNode {
+  SScanPhysiNode scan;
+  bool       onlyMetaCtbIdx; //no tbname, tag index not used.
+} STagScanPhysiNode;
+
 typedef SScanPhysiNode SBlockDistScanPhysiNode;
 
 typedef struct SLastRowScanPhysiNode {

include/libs/wal/wal.h

@@ -153,7 +153,6 @@ struct SWalReader {
   int64_t        capacity;
   TdThreadMutex  mutex;
   SWalFilterCond cond;
-  // TODO remove it
   SWalCkHead *pHead;
 };
 
@@ -207,10 +206,9 @@ void        walReaderValidVersionRange(SWalReader *pReader, int64_t *sver, int64
 void        walReaderVerifyOffset(SWalReader *pWalReader, STqOffsetVal* pOffset);
 
 // only for tq usage
-void    walSetReaderCapacity(SWalReader *pRead, int32_t capacity);
+int32_t walFetchHead(SWalReader *pRead, int64_t ver);
-int32_t walFetchHead(SWalReader *pRead, int64_t ver, SWalCkHead *pHead);
+int32_t walFetchBody(SWalReader *pRead);
-int32_t walFetchBody(SWalReader *pRead, SWalCkHead **ppHead);
+int32_t walSkipFetchBody(SWalReader *pRead);
-int32_t walSkipFetchBody(SWalReader *pRead, const SWalCkHead *pHead);
 
 void walRefFirstVer(SWal *, SWalRef *);
 void walRefLastVer(SWal *, SWalRef *);

include/os/osFile.h

@@ -76,7 +76,7 @@ int32_t taosUnLockFile(TdFilePtr pFile);
 
 int32_t taosUmaskFile(int32_t maskVal);
 
-int32_t taosStatFile(const char *path, int64_t *size, int32_t *mtime);
+int32_t taosStatFile(const char *path, int64_t *size, int32_t *mtime, int32_t *atime);
 int32_t taosDevInoFile(TdFilePtr pFile, int64_t *stDev, int64_t *stIno);
 int32_t taosFStatFile(TdFilePtr pFile, int64_t *size, int32_t *mtime);
 bool    taosCheckExistFile(const char *pathname);

source/client/CMakeLists.txt

@@ -16,7 +16,7 @@ target_include_directories(
 target_link_libraries(
     taos
     INTERFACE api
-    PRIVATE os util common transport nodes parser command planner catalog scheduler function qcom
+    PRIVATE os util common transport nodes parser command planner catalog scheduler function qcom geometry
 )
 
 if(TD_DARWIN_ARM64)
@@ -57,7 +57,7 @@ target_include_directories(
 target_link_libraries(
     taos_static
     INTERFACE api
-    PRIVATE os util common transport nodes parser command planner catalog scheduler function qcom
+    PRIVATE os util common transport nodes parser command planner catalog scheduler function qcom geometry
 )
 
 if(${BUILD_TEST})

source/client/inc/clientSml.h

@@ -33,6 +33,7 @@ extern "C" {
 #include "ttime.h"
 #include "ttypes.h"
 #include "cJSON.h"
+#include "geosWrapper.h"
 
 #if (defined(__GNUC__) && (__GNUC__ >= 3)) || (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 800)) || defined(__clang__)
 #  define expect(expr,value)    (__builtin_expect ((expr),(value)) )
@@ -192,7 +193,7 @@ typedef struct {
   //
   SArray      *preLineTagKV;
   SArray      *maxTagKVs;
-  SArray      *masColKVs;
+  SArray      *maxColKVs;
 
   SSmlLineInfo preLine;
   STableMeta  *currSTableMeta;

source/client/src/clientSml.c

@@ -1073,6 +1073,7 @@ static int32_t smlModifyDBSchemas(SSmlHandle *info) {
   return 0;
 
 end:
+  taosHashCancelIterate(info->superTables, tmp);
   taosHashCleanup(hashTmp);
   taosMemoryFreeClear(pTableMeta);
   catalogRefreshTableMeta(info->pCatalog, &conn, &pName, 1);
@@ -1191,6 +1192,7 @@ void freeSSmlKv(void *data) {
   SSmlKv *kv = (SSmlKv *)data;
   if (kv->keyEscaped) taosMemoryFree((void *)(kv->key));
   if (kv->valueEscaped) taosMemoryFree((void *)(kv->value));
+  if (kv->type == TSDB_DATA_TYPE_GEOMETRY) geosFreeBuffer((void *)(kv->value));
 }
 
 void smlDestroyInfo(SSmlHandle *info) {
@@ -1433,6 +1435,7 @@ static int32_t smlInsertData(SSmlHandle *info) {
     code = smlCheckAuth(info, &conn, pName.tname, AUTH_TYPE_WRITE);
     if(code != TSDB_CODE_SUCCESS){
       taosMemoryFree(measure);
+      taosHashCancelIterate(info->childTables, oneTable);
       return code;
     }
 
@@ -1441,6 +1444,7 @@ static int32_t smlInsertData(SSmlHandle *info) {
     if (code != TSDB_CODE_SUCCESS) {
       uError("SML:0x%" PRIx64 " catalogGetTableHashVgroup failed. table name: %s", info->id, tableData->childTableName);
       taosMemoryFree(measure);
+      taosHashCancelIterate(info->childTables, oneTable);
       return code;
     }
     taosHashPut(info->pVgHash, (const char *)&vg.vgId, sizeof(vg.vgId), (char *)&vg, sizeof(vg));
@@ -1450,6 +1454,7 @@ static int32_t smlInsertData(SSmlHandle *info) {
     if (unlikely(NULL == pMeta || NULL == (*pMeta)->tableMeta)) {
       uError("SML:0x%" PRIx64 " NULL == pMeta. table name: %s", info->id, tableData->childTableName);
       taosMemoryFree(measure);
+      taosHashCancelIterate(info->childTables, oneTable);
       return TSDB_CODE_SML_INTERNAL_ERROR;
     }
 
@@ -1465,6 +1470,7 @@ static int32_t smlInsertData(SSmlHandle *info) {
     taosMemoryFree(measure);
     if (code != TSDB_CODE_SUCCESS) {
       uError("SML:0x%" PRIx64 " smlBindData failed", info->id);
+      taosHashCancelIterate(info->childTables, oneTable);
       return code;
     }
     oneTable = (SSmlTableInfo **)taosHashIterate(info->childTables, oneTable);

source/client/src/clientSmlLine.c

@@ -102,6 +102,30 @@ int32_t smlParseValue(SSmlKv *pVal, SSmlMsgBuf *msg) {
     return TSDB_CODE_TSC_INVALID_VALUE;
   }
 
+  if (pVal->value[0] == 'g' || pVal->value[0] == 'G') {  // geometry
+    if (pVal->value[1] == '"' && pVal->value[pVal->length - 1] == '"' && pVal->length >= sizeof("POINT")+3) {
+      int32_t code = initCtxGeomFromText();
+      if (code != TSDB_CODE_SUCCESS) {
+        return code;
+      }
+      char* tmp = taosMemoryCalloc(pVal->length, 1);
+      memcpy(tmp, pVal->value + 2, pVal->length - 3);
+      code = doGeomFromText(tmp, (unsigned char **)&pVal->value, &pVal->length);
+      taosMemoryFree(tmp);
+      if (code != TSDB_CODE_SUCCESS) {
+        return code;
+      }
+
+      pVal->type = TSDB_DATA_TYPE_GEOMETRY;
+      if (pVal->length > TSDB_MAX_BINARY_LEN - VARSTR_HEADER_SIZE) {
+        geosFreeBuffer((void*)(pVal->value));
+        return TSDB_CODE_PAR_INVALID_VAR_COLUMN_LEN;
+      }
+      return TSDB_CODE_SUCCESS;
+    }
+    return TSDB_CODE_TSC_INVALID_VALUE;
+  }
+
   if (pVal->value[0] == 't' || pVal->value[0] == 'T') {
     if (pVal->length == 1 ||
         (pVal->length == 4 && (pVal->value[1] == 'r' || pVal->value[1] == 'R') &&
@@ -390,14 +414,14 @@ static int32_t smlParseColKv(SSmlHandle *info, char **sql, char *sqlEnd, SSmlLin
           SSmlKv   kv = {.key = tag->name, .keyLen = strlen(tag->name), .type = tag->type};
           if (tag->type == TSDB_DATA_TYPE_NCHAR) {
             kv.length = (tag->bytes - VARSTR_HEADER_SIZE) / TSDB_NCHAR_SIZE;
-          } else if (tag->type == TSDB_DATA_TYPE_BINARY) {
+          } else if (tag->type == TSDB_DATA_TYPE_BINARY || tag->type == TSDB_DATA_TYPE_GEOMETRY) {
             kv.length = tag->bytes - VARSTR_HEADER_SIZE;
           }
           taosArrayPush((*tmp)->cols, &kv);
         }
       }
       info->currSTableMeta = (*tmp)->tableMeta;
-      info->masColKVs = (*tmp)->cols;
+      info->maxColKVs = (*tmp)->cols;
     }
   }
 
@@ -512,13 +536,13 @@ static int32_t smlParseColKv(SSmlHandle *info, char **sql, char *sqlEnd, SSmlLin
         freeSSmlKv(&kv);
         return TSDB_CODE_SUCCESS;
       }
-      if (cnt >= taosArrayGetSize(info->masColKVs)) {
+      if (cnt >= taosArrayGetSize(info->maxColKVs)) {
         info->dataFormat = false;
         info->reRun = true;
         freeSSmlKv(&kv);
         return TSDB_CODE_SUCCESS;
       }
-      SSmlKv *maxKV = (SSmlKv *)taosArrayGet(info->masColKVs, cnt);
+      SSmlKv *maxKV = (SSmlKv *)taosArrayGet(info->maxColKVs, cnt);
       if (kv.type != maxKV->type) {
         info->dataFormat = false;
         info->reRun = true;
@@ -663,14 +687,15 @@ int32_t smlParseInfluxString(SSmlHandle *info, char *sql, char *sqlEnd, SSmlLine
   if (info->dataFormat) {
     uDebug("SML:0x%" PRIx64 " smlParseInfluxString format true, ts:%" PRId64, info->id, ts);
     ret = smlBuildCol(info->currTableDataCtx, info->currSTableMeta->schema, &kv, 0);
-    if (ret != TSDB_CODE_SUCCESS) {
+    if (ret == TSDB_CODE_SUCCESS) {
-      return ret;
-    }
       ret = smlBuildRow(info->currTableDataCtx);
-    if (ret != TSDB_CODE_SUCCESS) {
+    }
+
+    clearColValArray(info->currTableDataCtx->pValues);
+    if (unlikely(ret != TSDB_CODE_SUCCESS)) {
+      smlBuildInvalidDataMsg(&info->msgBuf, "smlBuildCol error", NULL);
       return ret;
     }
-    clearColValArray(info->currTableDataCtx->pValues);
   } else {
     uDebug("SML:0x%" PRIx64 " smlParseInfluxString format false, ts:%" PRId64, info->id, ts);
     taosArraySet(elements->colArray, 0, &kv);

source/client/src/clientTmq.c

@@ -1863,10 +1863,10 @@ static int32_t tmqHandleNoPollRsp(tmq_t* tmq, SMqRspWrapper* rspWrapper, bool* p
   return 0;
 }
 
-static void updateVgInfo(SMqClientVg* pVg, STqOffsetVal* reqOffset, STqOffsetVal* rspOffset, int64_t sver, int64_t ever, int64_t consumerId){
+static void updateVgInfo(SMqClientVg* pVg, STqOffsetVal* reqOffset, STqOffsetVal* rspOffset, int64_t sver, int64_t ever, int64_t consumerId, bool hasData){
   if (!pVg->seekUpdated) {
     tscDebug("consumer:0x%" PRIx64" local offset is update, since seekupdate not set", consumerId);
-    pVg->offsetInfo.beginOffset = *reqOffset;
+    if(hasData) pVg->offsetInfo.beginOffset = *reqOffset;
     pVg->offsetInfo.endOffset = *rspOffset;
   } else {
     tscDebug("consumer:0x%" PRIx64" local offset is NOT update, since seekupdate is set", consumerId);
@@ -1929,7 +1929,7 @@ static void* tmqHandleAllRsp(tmq_t* tmq, int64_t timeout, bool pollIfReset) {
           pVg->epSet = *pollRspWrapper->pEpset;
         }
 
-        updateVgInfo(pVg, &pDataRsp->reqOffset, &pDataRsp->rspOffset, pDataRsp->head.walsver, pDataRsp->head.walever, tmq->consumerId);
+        updateVgInfo(pVg, &pDataRsp->reqOffset, &pDataRsp->rspOffset, pDataRsp->head.walsver, pDataRsp->head.walever, tmq->consumerId, pDataRsp->blockNum != 0);
 
         char buf[TSDB_OFFSET_LEN] = {0};
         tFormatOffset(buf, TSDB_OFFSET_LEN, &pDataRsp->rspOffset);
@@ -1979,7 +1979,7 @@ static void* tmqHandleAllRsp(tmq_t* tmq, int64_t timeout, bool pollIfReset) {
           return NULL;
         }
 
-        updateVgInfo(pVg, &pollRspWrapper->metaRsp.rspOffset, &pollRspWrapper->metaRsp.rspOffset, pollRspWrapper->metaRsp.head.walsver, pollRspWrapper->metaRsp.head.walever, tmq->consumerId);
+        updateVgInfo(pVg, &pollRspWrapper->metaRsp.rspOffset, &pollRspWrapper->metaRsp.rspOffset, pollRspWrapper->metaRsp.head.walsver, pollRspWrapper->metaRsp.head.walever, tmq->consumerId, true);
         // build rsp
         SMqMetaRspObj* pRsp = tmqBuildMetaRspFromWrapper(pollRspWrapper);
         taosFreeQitem(pollRspWrapper);
@@ -2007,7 +2007,7 @@ static void* tmqHandleAllRsp(tmq_t* tmq, int64_t timeout, bool pollIfReset) {
           return NULL;
         }
 
-        updateVgInfo(pVg, &pollRspWrapper->taosxRsp.reqOffset, &pollRspWrapper->taosxRsp.rspOffset, pollRspWrapper->taosxRsp.head.walsver, pollRspWrapper->taosxRsp.head.walever, tmq->consumerId);
+        updateVgInfo(pVg, &pollRspWrapper->taosxRsp.reqOffset, &pollRspWrapper->taosxRsp.rspOffset, pollRspWrapper->taosxRsp.head.walsver, pollRspWrapper->taosxRsp.head.walever, tmq->consumerId, pollRspWrapper->taosxRsp.blockNum != 0);
 
         if (pollRspWrapper->taosxRsp.blockNum == 0) {
           tscDebug("consumer:0x%" PRIx64 " taosx empty block received, vgId:%d, vg total:%" PRId64 ", reqId:0x%" PRIx64,

source/client/test/CMakeLists.txt

@@ -20,7 +20,7 @@ TARGET_LINK_LIBRARIES(
 ADD_EXECUTABLE(smlTest smlTest.cpp)
 TARGET_LINK_LIBRARIES(
         smlTest
-        PUBLIC os util common transport parser catalog scheduler function gtest taos_static qcom
+        PUBLIC os util common transport parser catalog scheduler function gtest taos_static qcom geometry
 )
 
 TARGET_INCLUDE_DIRECTORIES(

source/client/test/clientTests.cpp

@@ -1442,178 +1442,4 @@ TEST(clientCase, sub_tb_mt_test) {
   }
 }
 
-TEST(clientCase, ts_3756) {
-//  taos_options(TSDB_OPTION_CONFIGDIR, "~/first/cfg");
-
-  TAOS* pConn = taos_connect("localhost", "root", "taosdata", NULL, 0);
-  ASSERT_NE(pConn, nullptr);
-
-  tmq_conf_t* conf = tmq_conf_new();
-
-  tmq_conf_set(conf, "enable.auto.commit", "false");
-  tmq_conf_set(conf, "auto.commit.interval.ms", "2000");
-  tmq_conf_set(conf, "group.id", "group_id_2");
-  tmq_conf_set(conf, "td.connect.user", "root");
-  tmq_conf_set(conf, "td.connect.pass", "taosdata");
-  tmq_conf_set(conf, "auto.offset.reset", "latest");
-  tmq_conf_set(conf, "msg.with.table.name", "false");
-
-  tmq_t* tmq = tmq_consumer_new(conf, NULL, 0);
-  tmq_conf_destroy(conf);
-
-  // 创建订阅 topics 列表
-  tmq_list_t* topicList = tmq_list_new();
-  tmq_list_append(topicList, "tp");
-
-  // 启动订阅
-  tmq_subscribe(tmq, topicList);
-  tmq_list_destroy(topicList);
-
-  TAOS_FIELD* fields = NULL;
-  int32_t     numOfFields = 0;
-  int32_t     precision = 0;
-  int32_t     totalRows = 0;
-  int32_t     msgCnt = 0;
-  int32_t     timeout = 200;
-
-  int32_t count = 0;
-
-  tmq_topic_assignment* pAssign = NULL;
-  int32_t numOfAssign = 0;
-
-  int32_t code = tmq_get_topic_assignment(tmq, "tp", &pAssign, &numOfAssign);
-  if (code != 0) {
-    printf("error occurs:%s\n", tmq_err2str(code));
-    tmq_free_assignment(pAssign);
-    tmq_consumer_close(tmq);
-    taos_close(pConn);
-    fprintf(stderr, "%d msg consumed, include %d rows\n", msgCnt, totalRows);
-    return;
-  }
-
-  for(int i = 0; i < numOfAssign; i++){
-    printf("assign i:%d, vgId:%d, offset:%lld, start:%lld, end:%lld\n", i, pAssign[i].vgId, pAssign[i].currentOffset, pAssign[i].begin, pAssign[i].end);
-  }
-
-//  tmq_offset_seek(tmq, "tp", pAssign[0].vgId, 4);
-  tmq_free_assignment(pAssign);
-
-  code = tmq_get_topic_assignment(tmq, "tp", &pAssign, &numOfAssign);
-  if (code != 0) {
-    printf("error occurs:%s\n", tmq_err2str(code));
-    tmq_free_assignment(pAssign);
-    tmq_consumer_close(tmq);
-    taos_close(pConn);
-    fprintf(stderr, "%d msg consumed, include %d rows\n", msgCnt, totalRows);
-    return;
-  }
-
-  for(int i = 0; i < numOfAssign; i++){
-    printf("assign i:%d, vgId:%d, offset:%lld, start:%lld, end:%lld\n", i, pAssign[i].vgId, pAssign[i].currentOffset, pAssign[i].begin, pAssign[i].end);
-  }
-
-  tmq_free_assignment(pAssign);
-
-  code = tmq_get_topic_assignment(tmq, "tp", &pAssign, &numOfAssign);
-  if (code != 0) {
-    printf("error occurs:%s\n", tmq_err2str(code));
-    tmq_free_assignment(pAssign);
-    tmq_consumer_close(tmq);
-    taos_close(pConn);
-    fprintf(stderr, "%d msg consumed, include %d rows\n", msgCnt, totalRows);
-    return;
-  }
-
-  for(int i = 0; i < numOfAssign; i++){
-    printf("assign i:%d, vgId:%d, offset:%lld, start:%lld, end:%lld\n", i, pAssign[i].vgId, pAssign[i].currentOffset, pAssign[i].begin, pAssign[i].end);
-  }
-
-  while (1) {
-    printf("start to poll\n");
-    TAOS_RES* pRes = tmq_consumer_poll(tmq, timeout);
-    if (pRes) {
-      char buf[128];
-
-      const char* topicName = tmq_get_topic_name(pRes);
-//      const char* dbName = tmq_get_db_name(pRes);
-//      int32_t     vgroupId = tmq_get_vgroup_id(pRes);
-//
-//      printf("topic: %s\n", topicName);
-//      printf("db: %s\n", dbName);
-//      printf("vgroup id: %d\n", vgroupId);
-
-      printSubResults(pRes, &totalRows);
-
-      tmq_topic_assignment* pAssignTmp = NULL;
-      int32_t numOfAssignTmp = 0;
-
-      code = tmq_get_topic_assignment(tmq, "tp", &pAssignTmp, &numOfAssignTmp);
-      if (code != 0) {
-        printf("error occurs:%s\n", tmq_err2str(code));
-        tmq_free_assignment(pAssign);
-        tmq_consumer_close(tmq);
-        taos_close(pConn);
-        fprintf(stderr, "%d msg consumed, include %d rows\n", msgCnt, totalRows);
-        return;
-      }
-
-      for(int i = 0; i < numOfAssign; i++){
-        printf("assign i:%d, vgId:%d, offset:%lld, start:%lld, end:%lld\n", i, pAssignTmp[i].vgId, pAssignTmp[i].currentOffset, pAssignTmp[i].begin, pAssignTmp[i].end);
-      }
-      if(numOfAssign != 0){
-        int i = 0;
-        for(; i < numOfAssign; i++){
-          if(pAssign[i].currentOffset != pAssignTmp[i].currentOffset){
-            break;
-          }
-        }
-        if(i == numOfAssign){
-          printf("all position is same\n");
-          break;
-        }
-        tmq_free_assignment(pAssign);
-      }
-      numOfAssign = numOfAssignTmp;
-      pAssign = pAssignTmp;
-
-    } else {
-//      tmq_offset_seek(tmq, "tp", pAssign[0].vgId, pAssign[0].currentOffset);
-//      tmq_offset_seek(tmq, "tp", pAssign[1].vgId, pAssign[1].currentOffset);
-//      tmq_commit_sync(tmq, pRes);
-      continue;
-    }
-
-//    tmq_commit_sync(tmq, pRes);
-    if (pRes != NULL) {
-      taos_free_result(pRes);
-      //      if ((++count) > 1) {
-      //        break;
-      //      }
-    } else {
-//      break;
-    }
-
-//    tmq_offset_seek(tmq, "tp", pAssign[0].vgId, pAssign[0].begin);
-  }
-
-  tmq_free_assignment(pAssign);
-
-  code = tmq_get_topic_assignment(tmq, "tp", &pAssign, &numOfAssign);
-  if (code != 0) {
-    printf("error occurs:%s\n", tmq_err2str(code));
-    tmq_free_assignment(pAssign);
-    tmq_consumer_close(tmq);
-    taos_close(pConn);
-    fprintf(stderr, "%d msg consumed, include %d rows\n", msgCnt, totalRows);
-    return;
-  }
-
-  for(int i = 0; i < numOfAssign; i++){
-    printf("assign i:%d, vgId:%d, offset:%lld, start:%lld, end:%lld\n", i, pAssign[i].vgId, pAssign[i].currentOffset, pAssign[i].begin, pAssign[i].end);
-  }
-
-  tmq_consumer_close(tmq);
-  taos_close(pConn);
-  fprintf(stderr, "%d msg consumed, include %d rows\n", msgCnt, totalRows);
-}
 #pragma GCC diagnostic pop

source/common/CMakeLists.txt

@@ -16,6 +16,14 @@ ENDIF ()
 IF (TD_STORAGE)
   ADD_DEFINITIONS(-D_STORAGE)
   TARGET_LINK_LIBRARIES(common PRIVATE storage)
+
+  IF(${TD_LINUX})
+    IF(${BUILD_WITH_COS})
+      add_definitions(-DUSE_COS)
+    ENDIF(${BUILD_WITH_COS})
+
+  ENDIF(${TD_LINUX})
+
 ENDIF ()
 
 target_include_directories(

source/common/src/systable.c

@@ -314,7 +314,7 @@ static const SSysDbTableSchema userUserPrivilegesSchema[] = {
 static const SSysTableMeta infosMeta[] = {
     {TSDB_INS_TABLE_DNODES, dnodesSchema, tListLen(dnodesSchema), true},
     {TSDB_INS_TABLE_MNODES, mnodesSchema, tListLen(mnodesSchema), true},
-    {TSDB_INS_TABLE_MODULES, modulesSchema, tListLen(modulesSchema), true},
+    // {TSDB_INS_TABLE_MODULES, modulesSchema, tListLen(modulesSchema), true},
     {TSDB_INS_TABLE_QNODES, qnodesSchema, tListLen(qnodesSchema), true},
     {TSDB_INS_TABLE_SNODES, snodesSchema, tListLen(snodesSchema), true},
     {TSDB_INS_TABLE_CLUSTER, clusterSchema, tListLen(clusterSchema), true},

source/common/src/tglobal.c

@@ -237,6 +237,14 @@ int64_t tsCheckpointInterval = 3 * 60 * 60 * 1000;
 bool    tsFilterScalarMode = false;
 int32_t tsKeepTimeOffset = 0;  // latency of data migration
 
+char   tsS3Endpoint[TSDB_FQDN_LEN] = "<endpoint>";
+char   tsS3AccessKey[TSDB_FQDN_LEN] = "<accesskey>";
+char   tsS3AccessKeyId[TSDB_FQDN_LEN] = "<accesskeyid>";
+char   tsS3AccessKeySecret[TSDB_FQDN_LEN] = "<accesskeysecrect>";
+char   tsS3BucketName[TSDB_FQDN_LEN] = "<bucketname>";
+char   tsS3AppId[TSDB_FQDN_LEN] = "<appid>";
+int8_t tsS3Enabled = false;
+
 #ifndef _STORAGE
 int32_t taosSetTfsCfg(SConfig *pCfg) {
   SConfigItem *pItem = cfgGetItem(pCfg, "dataDir");
@@ -258,7 +266,43 @@ int32_t taosSetTfsCfg(SConfig *pCfg) {
 int32_t taosSetTfsCfg(SConfig *pCfg);
 #endif
 
-struct SConfig *taosGetCfg() { return tsCfg; }
+int32_t taosSetS3Cfg(SConfig *pCfg) {
+  tstrncpy(tsS3AccessKey, cfgGetItem(pCfg, "s3Accesskey")->str, TSDB_FQDN_LEN);
+  if (tsS3AccessKey[0] == '<') {
+    return 0;
+  }
+  char *colon = strchr(tsS3AccessKey, ':');
+  if (!colon) {
+    uError("invalid access key:%s", tsS3AccessKey);
+    return -1;
+  }
+  *colon = '\0';
+  tstrncpy(tsS3AccessKeyId, tsS3AccessKey, TSDB_FQDN_LEN);
+  tstrncpy(tsS3AccessKeySecret, colon + 1, TSDB_FQDN_LEN);
+  tstrncpy(tsS3Endpoint, cfgGetItem(pCfg, "s3Endpoint")->str, TSDB_FQDN_LEN);
+  tstrncpy(tsS3BucketName, cfgGetItem(pCfg, "s3BucketName")->str, TSDB_FQDN_LEN);
+  char *cos = strstr(tsS3Endpoint, "cos.");
+  if (cos) {
+    char *appid = strrchr(tsS3BucketName, '-');
+    if (!appid) {
+      uError("failed to locate appid in bucket:%s", tsS3BucketName);
+      return -1;
+    } else {
+      tstrncpy(tsS3AppId, appid + 1, TSDB_FQDN_LEN);
+    }
+  }
+  if (tsS3BucketName[0] != '<' && tsDiskCfgNum > 1) {
+#ifdef USE_COS
+    tsS3Enabled = true;
+#endif
+  }
+
+  return 0;
+}
+
+struct SConfig *taosGetCfg() {
+  return tsCfg;
+}
 
 static int32_t taosLoadCfg(SConfig *pCfg, const char **envCmd, const char *inputCfgDir, const char *envFile,
                            char *apolloUrl) {
@@ -581,6 +625,10 @@ static int32_t taosAddServerCfg(SConfig *pCfg) {
   if (cfgAddInt32(pCfg, "maxStreamBackendCache", tsMaxStreamBackendCache, 16, 1024, CFG_SCOPE_SERVER) != 0) return -1;
   if (cfgAddInt32(pCfg, "pqSortMemThreshold", tsPQSortMemThreshold, 1, 10240, CFG_SCOPE_SERVER) != 0) return -1;
 
+  if (cfgAddString(pCfg, "s3Accesskey", tsS3AccessKey, CFG_SCOPE_SERVER) != 0) return -1;
+  if (cfgAddString(pCfg, "s3Endpoint", tsS3Endpoint, CFG_SCOPE_SERVER) != 0) return -1;
+  if (cfgAddString(pCfg, "s3BucketName", tsS3BucketName, CFG_SCOPE_SERVER) != 0) return -1;
+
   GRANT_CFG_ADD;
   return 0;
 }
@@ -1502,6 +1550,7 @@ int32_t taosInitCfg(const char *cfgDir, const char **envCmd, const char *envFile
     if (taosSetServerCfg(tsCfg)) return -1;
     if (taosSetReleaseCfg(tsCfg)) return -1;
     if (taosSetTfsCfg(tsCfg) != 0) return -1;
+    if (taosSetS3Cfg(tsCfg) != 0) return -1;
   }
   taosSetSystemCfg(tsCfg);
 

source/dnode/mgmt/mgmt_mnode/src/mmFile.c

@@ -65,7 +65,7 @@ int32_t mmReadFile(const char *path, SMnodeOpt *pOption) {
   char      file[PATH_MAX] = {0};
   snprintf(file, sizeof(file), "%s%smnode.json", path, TD_DIRSEP);
 
-  if (taosStatFile(file, NULL, NULL) < 0) {
+  if (taosStatFile(file, NULL, NULL, NULL) < 0) {
     dInfo("mnode file:%s not exist", file);
     return 0;
   }

source/dnode/mgmt/mgmt_vnode/src/vmFile.c

@@ -97,7 +97,7 @@ int32_t vmGetVnodeListFromFile(SVnodeMgmt *pMgmt, SWrapperCfg **ppCfgs, int32_t
   SWrapperCfg *pCfgs = NULL;
   snprintf(file, sizeof(file), "%s%svnodes.json", pMgmt->path, TD_DIRSEP);
 
-  if (taosStatFile(file, NULL, NULL) < 0) {
+  if (taosStatFile(file, NULL, NULL, NULL) < 0) {
     dInfo("vnode file:%s not exist", file);
     return 0;
   }

source/dnode/mgmt/node_util/src/dmEps.c

@@ -100,7 +100,7 @@ int32_t dmReadEps(SDnodeData *pData) {
     goto _OVER;
   }
 
-  if (taosStatFile(file, NULL, NULL) < 0) {
+  if (taosStatFile(file, NULL, NULL, NULL) < 0) {
     dInfo("dnode file:%s not exist", file);
     code = 0;
     goto _OVER;
@@ -469,7 +469,7 @@ static int32_t dmReadDnodePairs(SDnodeData *pData) {
   char      file[PATH_MAX] = {0};
   snprintf(file, sizeof(file), "%s%sdnode%sep.json", tsDataDir, TD_DIRSEP, TD_DIRSEP);
 
-  if (taosStatFile(file, NULL, NULL) < 0) {
+  if (taosStatFile(file, NULL, NULL, NULL) < 0) {
     dDebug("dnode file:%s not exist", file);
     code = 0;
     goto _OVER;

source/dnode/mgmt/node_util/src/dmFile.c

@@ -38,7 +38,7 @@ int32_t dmReadFile(const char *path, const char *name, bool *pDeployed) {
   char      file[PATH_MAX] = {0};
   snprintf(file, sizeof(file), "%s%s%s.json", path, TD_DIRSEP, name);
 
-  if (taosStatFile(file, NULL, NULL) < 0) {
+  if (taosStatFile(file, NULL, NULL, NULL) < 0) {
     dInfo("file:%s not exist", file);
     code = 0;
     goto _OVER;

source/dnode/mnode/impl/src/mndShow.c

@@ -58,8 +58,10 @@ static int32_t convertToRetrieveType(char *name, int32_t len) {
     type = TSDB_MGMT_TABLE_DNODE;
   } else if (strncasecmp(name, TSDB_INS_TABLE_MNODES, len) == 0) {
     type = TSDB_MGMT_TABLE_MNODE;
+/*
   } else if (strncasecmp(name, TSDB_INS_TABLE_MODULES, len) == 0) {
     type = TSDB_MGMT_TABLE_MODULE;
+*/
   } else if (strncasecmp(name, TSDB_INS_TABLE_QNODES, len) == 0) {
     type = TSDB_MGMT_TABLE_QNODE;
   } else if (strncasecmp(name, TSDB_INS_TABLE_SNODES, len) == 0) {

source/dnode/mnode/impl/src/mndSubscribe.c

@@ -244,7 +244,7 @@ static void doRemoveLostConsumers(SMqRebOutputObj *pOutput, SHashObj *pHash, con
 
         SMqRebOutputVg outputVg = {.oldConsumerId = consumerId, .newConsumerId = -1, .pVgEp = pVgEp};
         taosHashPut(pHash, &pVgEp->vgId, sizeof(int32_t), &outputVg, sizeof(SMqRebOutputVg));
-        mInfo("sub:%s mq re-balance remove vgId:%d from consumer:%" PRIx64, pSubKey, pVgEp->vgId, consumerId);
+        mInfo("sub:%s mq re-balance remove vgId:%d from consumer:0x%" PRIx64, pSubKey, pVgEp->vgId, consumerId);
       }
 
       taosArrayDestroy(pConsumerEp->vgs);
@@ -489,7 +489,7 @@ static int32_t mndDoRebalance(SMnode *pMnode, const SMqRebInputObj *pInput, SMqR
             SMqVgEp *pVgEp = taosArrayGetP(pConsumerEpNew->vgs, i);
             if(pVgEp->vgId == d1->vgId){
               jump = true;
-              mInfo("pSub->offsetRows jump, because consumer id:%"PRIx64 " and vgId:%d not change", pConsumerEp->consumerId, pVgEp->vgId);
+              mInfo("pSub->offsetRows jump, because consumer id:0x%"PRIx64 " and vgId:%d not change", pConsumerEp->consumerId, pVgEp->vgId);
               break;
             }
           }

source/dnode/vnode/CMakeLists.txt

@@ -8,6 +8,7 @@ set(
   "src/vnd/vnodeCommit.c"
   "src/vnd/vnodeQuery.c"
   "src/vnd/vnodeModule.c"
+  "src/vnd/vnodeCos.c"
   "src/vnd/vnodeSvr.c"
   "src/vnd/vnodeSync.c"
   "src/vnd/vnodeSnapshot.c"
@@ -155,6 +156,45 @@ target_link_libraries(
     PUBLIC index
 )
 
+if(${TD_LINUX})
+set(CMAKE_FIND_LIBRARY_SUFFIXES ".a")
+find_library(APR_LIBRARY apr-1 PATHS /usr/local/apr/lib/)
+find_library(APR_UTIL_LIBRARY aprutil-1 PATHS /usr/local/apr/lib/)
+find_library(MINIXML_LIBRARY mxml)
+find_library(CURL_LIBRARY curl)
+target_link_libraries(
+    vnode
+
+    # s3
+    PUBLIC cos_c_sdk_static
+    PUBLIC ${APR_UTIL_LIBRARY}
+    PUBLIC ${APR_LIBRARY}
+    PUBLIC ${MINIXML_LIBRARY}
+    PUBLIC ${CURL_LIBRARY}
+)
+
+# s3
+FIND_PROGRAM(APR_CONFIG_BIN NAMES apr-config apr-1-config PATHS /usr/bin /usr/local/bin /usr/local/apr/bin/)
+IF (APR_CONFIG_BIN)
+  EXECUTE_PROCESS(
+    COMMAND ${APR_CONFIG_BIN} --includedir
+    OUTPUT_VARIABLE APR_INCLUDE_DIR
+    OUTPUT_STRIP_TRAILING_WHITESPACE
+    )
+ENDIF()
+include_directories (${APR_INCLUDE_DIR})
+target_include_directories(
+  vnode
+  PUBLIC "${TD_SOURCE_DIR}/contrib/cos-c-sdk-v5/cos_c_sdk"
+  PUBLIC "$ENV{HOME}/.cos-local.1/include"
+  )
+
+if(${BUILD_WITH_COS})
+  add_definitions(-DUSE_COS)
+endif(${BUILD_WITH_COS})
+
+endif(${TD_LINUX})
+
 IF (TD_GRANT)
   TARGET_LINK_LIBRARIES(vnode PUBLIC grant)
 ENDIF ()
@@ -169,8 +209,6 @@ if(${BUILD_WITH_ROCKSDB})
     add_definitions(-DUSE_ROCKSDB)
 endif(${BUILD_WITH_ROCKSDB})
 
-
-
 if(${BUILD_TEST})
     add_subdirectory(test)
 endif(${BUILD_TEST})

source/dnode/vnode/src/inc/tq.h

@@ -127,7 +127,7 @@ void    tqDestroyTqHandle(void* data);
 // tqRead
 int32_t tqScanTaosx(STQ* pTq, const STqHandle* pHandle, STaosxRsp* pRsp, SMqMetaRsp* pMetaRsp, STqOffsetVal* offset);
 int32_t tqScanData(STQ* pTq, const STqHandle* pHandle, SMqDataRsp* pRsp, STqOffsetVal* pOffset);
-int32_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, SWalCkHead** pHeadWithCkSum, uint64_t reqId);
+int32_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, uint64_t reqId);
 
 // tqExec
 int32_t tqTaosxScanLog(STQ* pTq, STqHandle* pHandle, SPackedData submit, STaosxRsp* pRsp, int32_t* totalRows);
@@ -175,7 +175,7 @@ int32_t extractDelDataBlock(const void* pData, int32_t len, int64_t ver, SStream
 int32_t tqExtractDataForMq(STQ* pTq, STqHandle* pHandle, const SMqPollReq* pRequest, SRpcMsg* pMsg);
 int32_t tqDoSendDataRsp(const SRpcHandleInfo* pRpcHandleInfo, const SMqDataRsp* pRsp, int32_t epoch, int64_t consumerId,
                         int32_t type, int64_t sver, int64_t ever);
-int32_t tqInitDataRsp(SMqDataRsp* pRsp, const SMqPollReq* pReq);
+int32_t tqInitDataRsp(SMqDataRsp* pRsp, STqOffsetVal pOffset);
 #ifdef __cplusplus
 }
 #endif

source/dnode/vnode/src/inc/vndCos.h

@@ -0,0 +1,40 @@
+/*
+ * 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 _TD_VND_COS_H_
+#define _TD_VND_COS_H_
+
+#include "vnd.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+extern int8_t tsS3Enabled;
+
+int32_t s3Init();
+void    s3CleanUp();
+int32_t s3PutObjectFromFile(const char *file, const char *object);
+void    s3DeleteObjects(const char *object_name[], int nobject);
+bool    s3Exists(const char *object_name);
+bool    s3Get(const char *object_name, const char *path);
+void    s3EvictCache(const char *path, long object_size);
+long    s3Size(const char *object_name);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /*_TD_VND_COS_H_*/

source/dnode/vnode/src/inc/vnodeInt.h

@@ -167,7 +167,7 @@ int metaAddIndexToSTable(SMeta* pMeta, int64_t version, SVCreateStbReq* pReq);
 int metaDropIndexFromSTable(SMeta* pMeta, int64_t version, SDropIndexReq* pReq);
 
 int64_t       metaGetTimeSeriesNum(SMeta* pMeta);
-SMCtbCursor*  metaOpenCtbCursor(SMeta* pMeta, tb_uid_t uid, int lock);
+SMCtbCursor*  metaOpenCtbCursor(void* pVnode, tb_uid_t uid, int lock);
 void          metaCloseCtbCursor(SMCtbCursor* pCtbCur, int lock);
 tb_uid_t      metaCtbCursorNext(SMCtbCursor* pCtbCur);
 SMStbCursor*  metaOpenStbCursor(SMeta* pMeta, tb_uid_t uid);

source/dnode/vnode/src/meta/metaQuery.c

@@ -408,17 +408,9 @@ _err:
   return NULL;
 }
 
-struct SMCtbCursor {
-  SMeta   *pMeta;
-  TBC     *pCur;
-  tb_uid_t suid;
-  void    *pKey;
-  void    *pVal;
-  int      kLen;
-  int      vLen;
-};
 
-SMCtbCursor *metaOpenCtbCursor(SMeta *pMeta, tb_uid_t uid, int lock) {
+SMCtbCursor *metaOpenCtbCursor(void* pVnode, tb_uid_t uid, int lock) {
+  SMeta* pMeta = ((SVnode*)pVnode)->pMeta;
   SMCtbCursor *pCtbCur = NULL;
   SCtbIdxKey   ctbIdxKey;
   int          ret = 0;
@@ -435,7 +427,7 @@ SMCtbCursor *metaOpenCtbCursor(SMeta *pMeta, tb_uid_t uid, int lock) {
     metaRLock(pMeta);
   }
 
-  ret = tdbTbcOpen(pMeta->pCtbIdx, &pCtbCur->pCur, NULL);
+  ret = tdbTbcOpen(pMeta->pCtbIdx, (TBC**)&pCtbCur->pCur, NULL);
   if (ret < 0) {
     metaULock(pMeta);
     taosMemoryFree(pCtbCur);
@@ -1373,7 +1365,7 @@ int32_t metaGetTableTagsByUids(void *pVnode, int64_t suid, SArray *uidList) {
 }
 
 int32_t metaGetTableTags(void *pVnode, uint64_t suid, SArray *pUidTagInfo) {
-  SMCtbCursor *pCur = metaOpenCtbCursor(((SVnode *)pVnode)->pMeta, suid, 1);
+  SMCtbCursor *pCur = metaOpenCtbCursor(pVnode, suid, 1);
 
   // If len > 0 means there already have uids, and we only want the
   // tags of the specified tables, of which uid in the uid list. Otherwise, all table tags are retrieved and kept
@@ -1434,36 +1426,37 @@ int32_t metaGetInfo(SMeta *pMeta, int64_t uid, SMetaInfo *pInfo, SMetaReader *pR
   int     nData = 0;
   int     lock = 0;
 
-  metaRLock(pMeta);
+  if (pReader && !(pReader->flags & META_READER_NOLOCK)) {
+    lock = 1;
+  }
+
+  if(!lock) metaRLock(pMeta);
 
   // search cache
   if (metaCacheGet(pMeta, uid, pInfo) == 0) {
-    metaULock(pMeta);
+    if(!lock) metaULock(pMeta);
     goto _exit;
   }
 
   // search TDB
   if (tdbTbGet(pMeta->pUidIdx, &uid, sizeof(uid), &pData, &nData) < 0) {
     // not found
-    metaULock(pMeta);
+    if(!lock) metaULock(pMeta);
     code = TSDB_CODE_NOT_FOUND;
     goto _exit;
   }
 
-  metaULock(pMeta);
+  if(!lock) metaULock(pMeta);
 
   pInfo->uid = uid;
   pInfo->suid = ((SUidIdxVal *)pData)->suid;
   pInfo->version = ((SUidIdxVal *)pData)->version;
   pInfo->skmVer = ((SUidIdxVal *)pData)->skmVer;
 
-  if (pReader != NULL) {
-    lock = !(pReader->flags & META_READER_NOLOCK);
   if (lock) {
     metaULock(pReader->pMeta);
     // metaReaderReleaseLock(pReader);
   }
-  }
   // upsert the cache
   metaWLock(pMeta);
   metaCacheUpsert(pMeta, pInfo);

source/dnode/vnode/src/sma/smaRollup.c

@@ -905,6 +905,7 @@ int32_t tdProcessRSmaSubmit(SSma *pSma, int64_t version, void *pReq, void *pMsg,
         tb_uid_t *pTbSuid = (tb_uid_t *)taosHashGetKey(pIter, NULL);
         if (tdExecuteRSmaAsync(pSma, version, pMsg, len, inputType, *pTbSuid) < 0) {
           smaError("vgId:%d, failed to process rsma submit exec 2 since: %s", SMA_VID(pSma), terrstr());
+          taosHashCancelIterate(uidStore.uidHash, pIter);
           goto _err;
         }
       }

source/dnode/vnode/src/tq/tq.c

@@ -289,9 +289,8 @@ int32_t tqPushEmptyDataRsp(STqHandle* pHandle, int32_t vgId) {
   }
 
   SMqDataRsp dataRsp = {0};
-  tqInitDataRsp(&dataRsp, &req);
+  tqInitDataRsp(&dataRsp, req.reqOffset);
   dataRsp.blockNum = 0;
-  dataRsp.rspOffset = dataRsp.reqOffset;
   char buf[TSDB_OFFSET_LEN] = {0};
   tFormatOffset(buf, TSDB_OFFSET_LEN, &dataRsp.reqOffset);
   tqInfo("tqPushEmptyDataRsp to consumer:0x%"PRIx64 " vgId:%d, offset:%s, reqId:0x%" PRIx64, req.consumerId, vgId, buf, req.reqId);
@@ -391,7 +390,6 @@ int32_t tqProcessSeekReq(STQ* pTq, SRpcMsg* pMsg) {
   }
 
   tqDebug("tmq seek: consumer:0x%" PRIx64 " vgId:%d, subkey %s", req.consumerId, vgId, req.subKey);
-
   STqHandle* pHandle = taosHashGet(pTq->pHandle, req.subKey, strlen(req.subKey));
   if (pHandle == NULL) {
     tqWarn("tmq seek: consumer:0x%" PRIx64 " vgId:%d subkey %s not found", req.consumerId, vgId, req.subKey);
@@ -719,7 +717,7 @@ int32_t tqProcessVgWalInfoReq(STQ* pTq, SRpcMsg* pMsg) {
   walReaderValidVersionRange(pHandle->execHandle.pTqReader->pWalReader, &sver, &ever);
 
   SMqDataRsp dataRsp = {0};
-  tqInitDataRsp(&dataRsp, &req);
+  tqInitDataRsp(&dataRsp, req.reqOffset);
 
   if (req.useSnapshot == true) {
     tqError("consumer:0x%" PRIx64 " vgId:%d subkey:%s snapshot not support wal info", consumerId, vgId, req.subKey);

source/dnode/vnode/src/tq/tqMeta.c

@@ -356,7 +356,7 @@ static int restoreHandle(STQ* pTq, void* pVal, int vLen, STqHandle* handle){
   if(buildHandle(pTq, handle) < 0){
     return -1;
   }
-  tqInfo("tq restore %s consumer %" PRId64 " vgId:%d", handle->subKey, handle->consumerId, vgId);
+  tqInfo("restoreHandle %s consumer 0x%" PRIx64 " vgId:%d", handle->subKey, handle->consumerId, vgId);
   return taosHashPut(pTq->pHandle, handle->subKey, strlen(handle->subKey), handle, sizeof(STqHandle));
 }
 
@@ -384,7 +384,7 @@ int32_t tqCreateHandle(STQ* pTq, SMqRebVgReq* req, STqHandle* handle){
   if(buildHandle(pTq, handle) < 0){
     return -1;
   }
-  tqInfo("tq restore %s consumer %" PRId64 " vgId:%d", handle->subKey, handle->consumerId, vgId);
+  tqInfo("tqCreateHandle %s consumer 0x%" PRIx64 " vgId:%d", handle->subKey, handle->consumerId, vgId);
   return taosHashPut(pTq->pHandle, handle->subKey, strlen(handle->subKey), handle, sizeof(STqHandle));
 }
 

source/dnode/vnode/src/tq/tqOffsetSnapshot.c

@@ -60,7 +60,7 @@ int32_t tqOffsetSnapRead(STqOffsetReader* pReader, uint8_t** ppData) {
   }
 
   int64_t sz = 0;
-  if (taosStatFile(fname, &sz, NULL) < 0) {
+  if (taosStatFile(fname, &sz, NULL, NULL) < 0) {
     taosCloseFile(&pFile);
     taosMemoryFree(fname);
     return -1;

source/dnode/vnode/src/tq/tqPush.c

@@ -39,7 +39,7 @@ int32_t tqPushMsg(STQ* pTq, void* msg, int32_t msgLen, tmsg_t msgType, int64_t v
   int32_t numOfTasks = streamMetaGetNumOfTasks(pTq->pStreamMeta);
   taosRUnLockLatch(&pTq->pStreamMeta->lock);
 
-  tqDebug("handle submit, restore:%d, size:%d", pTq->pVnode->restored, numOfTasks);
+  tqTrace("handle submit, restore:%d, size:%d", pTq->pVnode->restored, numOfTasks);
 
   // push data for stream processing:
   // 1. the vnode has already been restored.

source/dnode/vnode/src/tq/tqRead.c

@@ -184,70 +184,63 @@ end:
   return tbSuid == realTbSuid;
 }
 
-int32_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, SWalCkHead** ppCkHead, uint64_t reqId) {
+int32_t tqFetchLog(STQ* pTq, STqHandle* pHandle, int64_t* fetchOffset, uint64_t reqId) {
-  int32_t code = 0;
+  int32_t code = -1;
   int32_t vgId = TD_VID(pTq->pVnode);
 
-  taosThreadMutexLock(&pHandle->pWalReader->mutex);
   int64_t offset = *fetchOffset;
+  int64_t lastVer = walGetLastVer(pHandle->pWalReader->pWal);
+  int64_t committedVer = walGetCommittedVer(pHandle->pWalReader->pWal);
+  int64_t appliedVer = walGetAppliedVer(pHandle->pWalReader->pWal);
 
-  while (1) {
+  wDebug("vgId:%d, wal start to fetch, index:%" PRId64 ", last index:%" PRId64 " commit index:%" PRId64 ", applied index:%" PRId64,
-    if (walFetchHead(pHandle->pWalReader, offset, *ppCkHead) < 0) {
+         vgId, offset, lastVer, committedVer, appliedVer);
+
+  while (offset <= appliedVer) {
+    if (walFetchHead(pHandle->pWalReader, offset) < 0) {
       tqDebug("tmq poll: consumer:0x%" PRIx64 ", (epoch %d) vgId:%d offset %" PRId64
               ", no more log to return, reqId:0x%" PRIx64,
               pHandle->consumerId, pHandle->epoch, vgId, offset, reqId);
-      *fetchOffset = offset;
-      code = -1;
       goto END;
     }
 
     tqDebug("vgId:%d, consumer:0x%" PRIx64 " taosx get msg ver %" PRId64 ", type: %s, reqId:0x%" PRIx64, vgId,
-            pHandle->consumerId, offset, TMSG_INFO((*ppCkHead)->head.msgType), reqId);
+            pHandle->consumerId, offset, TMSG_INFO(pHandle->pWalReader->pHead->head.msgType), reqId);
 
-    if ((*ppCkHead)->head.msgType == TDMT_VND_SUBMIT) {
+    if (pHandle->pWalReader->pHead->head.msgType == TDMT_VND_SUBMIT) {
-      code = walFetchBody(pHandle->pWalReader, ppCkHead);
+      code = walFetchBody(pHandle->pWalReader);
-
-      if (code < 0) {
-        *fetchOffset = offset;
-        code = -1;
-        goto END;
-      }
-      *fetchOffset = offset;
-      code = 0;
       goto END;
     } else {
       if (pHandle->fetchMeta != WITH_DATA) {
-        SWalCont* pHead = &((*ppCkHead)->head);
+        SWalCont* pHead = &(pHandle->pWalReader->pHead->head);
         if (IS_META_MSG(pHead->msgType) && !(pHead->msgType == TDMT_VND_DELETE && pHandle->fetchMeta == ONLY_META)) {
-          code = walFetchBody(pHandle->pWalReader, ppCkHead);
+          code = walFetchBody(pHandle->pWalReader);
           if (code < 0) {
-            *fetchOffset = offset;
-            code = -1;
             goto END;
           }
 
+          pHead = &(pHandle->pWalReader->pHead->head);
           if (isValValidForTable(pHandle, pHead)) {
-            *fetchOffset = offset;
             code = 0;
             goto END;
           } else {
             offset++;
+            code = -1;
             continue;
           }
         }
       }
-      code = walSkipFetchBody(pHandle->pWalReader, *ppCkHead);
+      code = walSkipFetchBody(pHandle->pWalReader);
       if (code < 0) {
-        *fetchOffset = offset;
-        code = -1;
         goto END;
       }
       offset++;
     }
+    code = -1;
   }
 
 END:
-  taosThreadMutexUnlock(&pHandle->pWalReader->mutex);
+  *fetchOffset = offset;
   return code;
 }
 
@@ -1095,6 +1088,7 @@ int32_t tqUpdateTbUidList(STQ* pTq, const SArray* tbUidList, bool isAdd) {
         if(ret != TDB_CODE_SUCCESS) {
           tqError("qGetTableList in tqUpdateTbUidList error:%d handle %s consumer:0x%" PRIx64, ret, pTqHandle->subKey, pTqHandle->consumerId);
           taosArrayDestroy(list);
+          taosHashCancelIterate(pTq->pHandle, pIter);
           return ret;
         }
         tqReaderSetTbUidList(pTqHandle->execHandle.pTqReader, list, NULL);

source/dnode/vnode/src/tq/tqUtil.c

@@ -20,8 +20,9 @@
 static int32_t tqSendMetaPollRsp(STqHandle* pHandle, const SRpcMsg* pMsg, const SMqPollReq* pReq,
                                  const SMqMetaRsp* pRsp, int32_t vgId);
 
-int32_t tqInitDataRsp(SMqDataRsp* pRsp, const SMqPollReq* pReq) {
+int32_t tqInitDataRsp(SMqDataRsp* pRsp, STqOffsetVal pOffset) {
-  pRsp->reqOffset = pReq->reqOffset;
+  pRsp->reqOffset = pOffset;
+  pRsp->rspOffset = pOffset;
 
   pRsp->blockData = taosArrayInit(0, sizeof(void*));
   pRsp->blockDataLen = taosArrayInit(0, sizeof(int32_t));
@@ -35,8 +36,9 @@ int32_t tqInitDataRsp(SMqDataRsp* pRsp, const SMqPollReq* pReq) {
   return 0;
 }
 
-static int32_t tqInitTaosxRsp(STaosxRsp* pRsp, const SMqPollReq* pReq) {
+static int32_t tqInitTaosxRsp(STaosxRsp* pRsp, STqOffsetVal pOffset) {
-  pRsp->reqOffset = pReq->reqOffset;
+  pRsp->reqOffset = pOffset;
+  pRsp->rspOffset = pOffset;
 
   pRsp->withTbName = 1;
   pRsp->withSchema = 1;
@@ -69,7 +71,6 @@ static int32_t tqInitTaosxRsp(STaosxRsp* pRsp, const SMqPollReq* pReq) {
 
 static int32_t extractResetOffsetVal(STqOffsetVal* pOffsetVal, STQ* pTq, STqHandle* pHandle, const SMqPollReq* pRequest, SRpcMsg* pMsg, bool* pBlockReturned) {
   uint64_t     consumerId = pRequest->consumerId;
-  STqOffsetVal reqOffset = pRequest->reqOffset;
   STqOffset*   pOffset = tqOffsetRead(pTq->pOffsetStore, pRequest->subKey);
   int32_t      vgId = TD_VID(pTq->pVnode);
 
@@ -86,7 +87,7 @@ static int32_t extractResetOffsetVal(STqOffsetVal* pOffsetVal, STQ* pTq, STqHand
     return 0;
   } else {
     // no poll occurs in this vnode for this topic, let's seek to the right offset value.
-    if (reqOffset.type == TMQ_OFFSET__RESET_EARLIEST) {
+    if (pRequest->reqOffset.type == TMQ_OFFSET__RESET_EARLIEST) {
       if (pRequest->useSnapshot) {
         tqDebug("tmq poll: consumer:0x%" PRIx64 ", subkey:%s, vgId:%d, (earliest) set offset to be snapshot",
                 consumerId, pHandle->subKey, vgId);
@@ -100,12 +101,12 @@ static int32_t extractResetOffsetVal(STqOffsetVal* pOffsetVal, STQ* pTq, STqHand
         walRefFirstVer(pTq->pVnode->pWal, pHandle->pRef);
         tqOffsetResetToLog(pOffsetVal, pHandle->pRef->refVer);
       }
-    } else if (reqOffset.type == TMQ_OFFSET__RESET_LATEST) {
+    } else if (pRequest->reqOffset.type == TMQ_OFFSET__RESET_LATEST) {
       walRefLastVer(pTq->pVnode->pWal, pHandle->pRef);
       SMqDataRsp dataRsp = {0};
-      tqInitDataRsp(&dataRsp, pRequest);
+      tqOffsetResetToLog(pOffsetVal, pHandle->pRef->refVer + 1);
 
-      tqOffsetResetToLog(&dataRsp.rspOffset, pHandle->pRef->refVer + 1);
+      tqInitDataRsp(&dataRsp, *pOffsetVal);
       tqDebug("tmq poll: consumer:0x%" PRIx64 ", subkey %s, vgId:%d, (latest) offset reset to %" PRId64, consumerId,
               pHandle->subKey, vgId, dataRsp.rspOffset.version);
       int32_t code = tqSendDataRsp(pHandle, pMsg, pRequest, &dataRsp, TMQ_MSG_TYPE__POLL_DATA_RSP, vgId);
@@ -113,7 +114,7 @@ static int32_t extractResetOffsetVal(STqOffsetVal* pOffsetVal, STQ* pTq, STqHand
 
       *pBlockReturned = true;
       return code;
-    } else if (reqOffset.type == TMQ_OFFSET__RESET_NONE) {
+    } else if (pRequest->reqOffset.type == TMQ_OFFSET__RESET_NONE) {
       tqError("tmq poll: subkey:%s, no offset committed for consumer:0x%" PRIx64
               " in vg %d, subkey %s, reset none failed",
               pHandle->subKey, consumerId, vgId, pRequest->subKey);
@@ -125,11 +126,11 @@ static int32_t extractResetOffsetVal(STqOffsetVal* pOffsetVal, STQ* pTq, STqHand
   return 0;
 }
 
-static void setRequestVersion(STqOffsetVal* offset, int64_t ver){
+//static void setRequestVersion(STqOffsetVal* offset, int64_t ver){
-  if(offset->type == TMQ_OFFSET__LOG){
+//  if(offset->type == TMQ_OFFSET__LOG){
-    offset->version = ver + 1;
+//    offset->version = ver;
-  }
+//  }
-}
+//}
 
 static int32_t extractDataAndRspForNormalSubscribe(STQ* pTq, STqHandle* pHandle, const SMqPollReq* pRequest,
                                                    SRpcMsg* pMsg, STqOffsetVal* pOffset) {
@@ -138,8 +139,8 @@ static int32_t extractDataAndRspForNormalSubscribe(STQ* pTq, STqHandle* pHandle,
   terrno        = 0;
 
   SMqDataRsp dataRsp = {0};
-  tqInitDataRsp(&dataRsp, pRequest);
+  tqInitDataRsp(&dataRsp, *pOffset);
-  dataRsp.reqOffset.type = pOffset->type; // stroe origin type for getting offset in tmq_get_vgroup_offset
+//  dataRsp.reqOffset.type = pOffset->type; // store origin type for getting offset in tmq_get_vgroup_offset
 
   qSetTaskId(pHandle->execHandle.task, consumerId, pRequest->reqId);
   int code = tqScanData(pTq, pHandle, &dataRsp, pOffset);
@@ -152,8 +153,7 @@ static int32_t extractDataAndRspForNormalSubscribe(STQ* pTq, STqHandle* pHandle,
     // lock
     taosWLockLatch(&pTq->lock);
     int64_t ver = walGetCommittedVer(pTq->pVnode->pWal);
-    if (pOffset->version >= ver ||
+    if (dataRsp.rspOffset.version > ver) {  // check if there are data again to avoid lost data
-        dataRsp.rspOffset.version >= ver) {  // check if there are data again to avoid lost data
       code = tqRegisterPushHandle(pTq, pHandle, pMsg);
       taosWUnLockLatch(&pTq->lock);
       goto end;
@@ -161,7 +161,7 @@ static int32_t extractDataAndRspForNormalSubscribe(STQ* pTq, STqHandle* pHandle,
     taosWUnLockLatch(&pTq->lock);
   }
 
-  setRequestVersion(&dataRsp.reqOffset, pOffset->version);
+//  setRequestVersion(&dataRsp.reqOffset, pOffset->version);
   code = tqSendDataRsp(pHandle, pMsg, pRequest, (SMqDataRsp*)&dataRsp, TMQ_MSG_TYPE__POLL_DATA_RSP, vgId);
 
 end : {
@@ -179,11 +179,10 @@ static int32_t extractDataAndRspForDbStbSubscribe(STQ* pTq, STqHandle* pHandle,
                                                   SRpcMsg* pMsg, STqOffsetVal* offset) {
   int         code = 0;
   int32_t     vgId = TD_VID(pTq->pVnode);
-  SWalCkHead* pCkHead = NULL;
   SMqMetaRsp  metaRsp = {0};
   STaosxRsp   taosxRsp = {0};
-  tqInitTaosxRsp(&taosxRsp, pRequest);
+  tqInitTaosxRsp(&taosxRsp, *offset);
-  taosxRsp.reqOffset.type = offset->type;   // store origin type for getting offset in tmq_get_vgroup_offset
+//  taosxRsp.reqOffset.type = offset->type;   // store origin type for getting offset in tmq_get_vgroup_offset
 
   if (offset->type != TMQ_OFFSET__LOG) {
     if (tqScanTaosx(pTq, pHandle, &taosxRsp, &metaRsp, offset) < 0) {
@@ -216,14 +215,7 @@ static int32_t extractDataAndRspForDbStbSubscribe(STQ* pTq, STqHandle* pHandle,
   if (offset->type == TMQ_OFFSET__LOG) {
     walReaderVerifyOffset(pHandle->pWalReader, offset);
     int64_t fetchVer = offset->version;
-    pCkHead = taosMemoryMalloc(sizeof(SWalCkHead) + 2048);
-    if (pCkHead == NULL) {
-      terrno = TSDB_CODE_OUT_OF_MEMORY;
-      code = -1;
-      goto end;
-    }
 
-    walSetReaderCapacity(pHandle->pWalReader, 2048);
     int totalRows = 0;
     while (1) {
       int32_t savedEpoch = atomic_load_32(&pHandle->epoch);
@@ -234,14 +226,14 @@ static int32_t extractDataAndRspForDbStbSubscribe(STQ* pTq, STqHandle* pHandle,
         break;
       }
 
-      if (tqFetchLog(pTq, pHandle, &fetchVer, &pCkHead, pRequest->reqId) < 0) {
+      if (tqFetchLog(pTq, pHandle, &fetchVer, pRequest->reqId) < 0) {
         tqOffsetResetToLog(&taosxRsp.rspOffset, fetchVer);
-        setRequestVersion(&taosxRsp.reqOffset, offset->version);
+//        setRequestVersion(&taosxRsp.reqOffset, offset->version);
         code = tqSendDataRsp(pHandle, pMsg, pRequest, (SMqDataRsp*)&taosxRsp, TMQ_MSG_TYPE__POLL_DATA_RSP, vgId);
         goto end;
       }
 
-      SWalCont* pHead = &pCkHead->head;
+      SWalCont* pHead = &pHandle->pWalReader->pHead->head;
       tqDebug("tmq poll: consumer:0x%" PRIx64 " (epoch %d) iter log, vgId:%d offset %" PRId64 " msgType %d",
               pRequest->consumerId, pRequest->epoch, vgId, fetchVer, pHead->msgType);
 
@@ -249,7 +241,7 @@ static int32_t extractDataAndRspForDbStbSubscribe(STQ* pTq, STqHandle* pHandle,
       if (pHead->msgType != TDMT_VND_SUBMIT) {
         if (totalRows > 0) {
           tqOffsetResetToLog(&taosxRsp.rspOffset, fetchVer);
-          setRequestVersion(&taosxRsp.reqOffset, offset->version);
+//          setRequestVersion(&taosxRsp.reqOffset, offset->version);
           code = tqSendDataRsp(pHandle, pMsg, pRequest, (SMqDataRsp*)&taosxRsp, TMQ_MSG_TYPE__POLL_DATA_RSP, vgId);
           goto end;
         }
@@ -279,7 +271,7 @@ static int32_t extractDataAndRspForDbStbSubscribe(STQ* pTq, STqHandle* pHandle,
 
       if (totalRows >= 4096 || taosxRsp.createTableNum > 0) {
         tqOffsetResetToLog(&taosxRsp.rspOffset, fetchVer + 1);
-        setRequestVersion(&taosxRsp.reqOffset, offset->version);
+//        setRequestVersion(&taosxRsp.reqOffset, offset->version);
         code = tqSendDataRsp(pHandle, pMsg, pRequest, (SMqDataRsp*)&taosxRsp, taosxRsp.createTableNum > 0 ? TMQ_MSG_TYPE__POLL_DATA_META_RSP : TMQ_MSG_TYPE__POLL_DATA_RSP, vgId);
         goto end;
       } else {
@@ -291,20 +283,17 @@ static int32_t extractDataAndRspForDbStbSubscribe(STQ* pTq, STqHandle* pHandle,
 end:
 
   tDeleteSTaosxRsp(&taosxRsp);
-  taosMemoryFreeClear(pCkHead);
   return code;
 }
 
 int32_t tqExtractDataForMq(STQ* pTq, STqHandle* pHandle, const SMqPollReq* pRequest, SRpcMsg* pMsg) {
-  int32_t      code = -1;
-  STqOffsetVal offset = {0};
   STqOffsetVal reqOffset = pRequest->reqOffset;
 
   // 1. reset the offset if needed
-  if (IS_OFFSET_RESET_TYPE(reqOffset.type)) {
+  if (IS_OFFSET_RESET_TYPE(pRequest->reqOffset.type)) {
     // handle the reset offset cases, according to the consumer's choice.
     bool blockReturned = false;
-    code = extractResetOffsetVal(&offset, pTq, pHandle, pRequest, pMsg, &blockReturned);
+    int32_t code = extractResetOffsetVal(&reqOffset, pTq, pHandle, pRequest, pMsg, &blockReturned);
     if (code != 0) {
       return code;
     }
@@ -313,20 +302,17 @@ int32_t tqExtractDataForMq(STQ* pTq, STqHandle* pHandle, const SMqPollReq* pRequ
     if (blockReturned) {
       return 0;
     }
-  } else if(reqOffset.type != 0){  // use the consumer specified offset
+  } else if(reqOffset.type == 0){  // use the consumer specified offset
-    // the offset value can not be monotonious increase??
-    offset = reqOffset;
-  } else {
     uError("req offset type is 0");
     return TSDB_CODE_TMQ_INVALID_MSG;
   }
 
   // this is a normal subscribe requirement
   if (pHandle->execHandle.subType == TOPIC_SUB_TYPE__COLUMN) {
-    return extractDataAndRspForNormalSubscribe(pTq, pHandle, pRequest, pMsg, &offset);
+    return extractDataAndRspForNormalSubscribe(pTq, pHandle, pRequest, pMsg, &reqOffset);
   } else {  // todo handle the case where re-balance occurs.
     // for taosx
-    return extractDataAndRspForDbStbSubscribe(pTq, pHandle, pRequest, pMsg, &offset);
+    return extractDataAndRspForDbStbSubscribe(pTq, pHandle, pRequest, pMsg, &reqOffset);
   }
 }
 

source/dnode/vnode/src/tsdb/tsdbFS.c

@@ -176,7 +176,7 @@ static int32_t tsdbScanAndTryFixFS(STsdb *pTsdb) {
   // SDelFile
   if (pTsdb->fs.pDelFile) {
     tsdbDelFileName(pTsdb, pTsdb->fs.pDelFile, fname);
-    if (taosStatFile(fname, &size, NULL)) {
+    if (taosStatFile(fname, &size, NULL, NULL)) {
       code = TAOS_SYSTEM_ERROR(errno);
       TSDB_CHECK_CODE(code, lino, _exit);
     }
@@ -195,7 +195,7 @@ static int32_t tsdbScanAndTryFixFS(STsdb *pTsdb) {
 
     // head =========
     tsdbHeadFileName(pTsdb, pSet->diskId, pSet->fid, pSet->pHeadF, fname);
-    if (taosStatFile(fname, &size, NULL)) {
+    if (taosStatFile(fname, &size, NULL, NULL)) {
       code = TAOS_SYSTEM_ERROR(errno);
       TSDB_CHECK_CODE(code, lino, _exit);
     }
@@ -206,7 +206,7 @@ static int32_t tsdbScanAndTryFixFS(STsdb *pTsdb) {
 
     // data =========
     tsdbDataFileName(pTsdb, pSet->diskId, pSet->fid, pSet->pDataF, fname);
-    if (taosStatFile(fname, &size, NULL)) {
+    if (taosStatFile(fname, &size, NULL, NULL)) {
       code = TAOS_SYSTEM_ERROR(errno);
       TSDB_CHECK_CODE(code, lino, _exit);
     }
@@ -221,7 +221,7 @@ static int32_t tsdbScanAndTryFixFS(STsdb *pTsdb) {
 
     // sma =============
     tsdbSmaFileName(pTsdb, pSet->diskId, pSet->fid, pSet->pSmaF, fname);
-    if (taosStatFile(fname, &size, NULL)) {
+    if (taosStatFile(fname, &size, NULL, NULL)) {
       code = TAOS_SYSTEM_ERROR(errno);
       TSDB_CHECK_CODE(code, lino, _exit);
     }
@@ -237,7 +237,7 @@ static int32_t tsdbScanAndTryFixFS(STsdb *pTsdb) {
     // stt ===========
     for (int32_t iStt = 0; iStt < pSet->nSttF; iStt++) {
       tsdbSttFileName(pTsdb, pSet->diskId, pSet->fid, pSet->aSttF[iStt], fname);
-      if (taosStatFile(fname, &size, NULL)) {
+      if (taosStatFile(fname, &size, NULL, NULL)) {
         code = TAOS_SYSTEM_ERROR(errno);
         TSDB_CHECK_CODE(code, lino, _exit);
       }

source/dnode/vnode/src/tsdb/tsdbReaderWriter.c

@@ -14,6 +14,7 @@
  */
 
 #include "tsdb.h"
+#include "vndCos.h"
 
 // =============== PAGE-WISE FILE ===============
 int32_t tsdbOpenFile(const char *path, int32_t szPage, int32_t flag, STsdbFD **ppFD) {
@@ -33,11 +34,26 @@ int32_t tsdbOpenFile(const char *path, int32_t szPage, int32_t flag, STsdbFD **p
   pFD->szPage = szPage;
   pFD->flag = flag;
   pFD->pFD = taosOpenFile(path, flag);
+  if (pFD->pFD == NULL) {
+    const char *object_name = taosDirEntryBaseName((char *)path);
+    long        s3_size = s3Size(object_name);
+    if (!strncmp(path + strlen(path) - 5, ".data", 5) && s3_size > 0) {
+      s3EvictCache(path, s3_size);
+      s3Get(object_name, path);
+
+      pFD->pFD = taosOpenFile(path, flag);
+
       if (pFD->pFD == NULL) {
         code = TAOS_SYSTEM_ERROR(errno);
         taosMemoryFree(pFD);
         goto _exit;
       }
+    } else {
+      code = TAOS_SYSTEM_ERROR(errno);
+      taosMemoryFree(pFD);
+      goto _exit;
+    }
+  }
   pFD->szPage = szPage;
   pFD->pgno = 0;
   pFD->pBuf = taosMemoryCalloc(1, szPage);
@@ -50,7 +66,7 @@ int32_t tsdbOpenFile(const char *path, int32_t szPage, int32_t flag, STsdbFD **p
 
   // not check file size when reading data files.
   if (flag != TD_FILE_READ) {
-    if (taosStatFile(path, &pFD->szFile, NULL) < 0) {
+    if (taosStatFile(path, &pFD->szFile, NULL, NULL) < 0) {
       code = TAOS_SYSTEM_ERROR(errno);
       taosMemoryFree(pFD->pBuf);
       taosCloseFile(&pFD->pFD);

source/dnode/vnode/src/tsdb/tsdbRetention.c

@@ -15,6 +15,7 @@
 
 #include "tsdb.h"
 #include "tsdbFS2.h"
+#include "vndCos.h"
 
 typedef struct {
   STsdb  *tsdb;
@@ -41,6 +42,28 @@ static int32_t tsdbDoRemoveFileObject(SRTNer *rtner, const STFileObj *fobj) {
   return TARRAY2_APPEND(rtner->fopArr, op);
 }
 
+static int32_t tsdbRemoveFileObjectS3(SRTNer *rtner, const STFileObj *fobj) {
+  int32_t code = 0, lino = 0;
+
+  STFileOp op = {
+      .optype = TSDB_FOP_REMOVE,
+      .fid = fobj->f->fid,
+      .of = fobj->f[0],
+  };
+
+  code = TARRAY2_APPEND(rtner->fopArr, op);
+  TSDB_CHECK_CODE(code, lino, _exit);
+
+  const char *object_name = taosDirEntryBaseName((char *)fobj->fname);
+  s3DeleteObjects(&object_name, 1);
+
+_exit:
+  if (code) {
+    TSDB_ERROR_LOG(TD_VID(rtner->tsdb->pVnode), lino, code);
+  }
+  return code;
+}
+
 static int32_t tsdbDoCopyFile(SRTNer *rtner, const STFileObj *from, const STFile *to) {
   int32_t code = 0;
   int32_t lino = 0;
@@ -76,6 +99,34 @@ _exit:
   return code;
 }
 
+static int32_t tsdbCopyFileS3(SRTNer *rtner, const STFileObj *from, const STFile *to) {
+  int32_t code = 0;
+  int32_t lino = 0;
+
+  char      fname[TSDB_FILENAME_LEN];
+  TdFilePtr fdFrom = NULL;
+  TdFilePtr fdTo = NULL;
+
+  tsdbTFileName(rtner->tsdb, to, fname);
+
+  fdFrom = taosOpenFile(from->fname, TD_FILE_READ);
+  if (fdFrom == NULL) code = terrno;
+  TSDB_CHECK_CODE(code, lino, _exit);
+
+  char *object_name = taosDirEntryBaseName(fname);
+  code = s3PutObjectFromFile(from->fname, object_name);
+  TSDB_CHECK_CODE(code, lino, _exit);
+
+  taosCloseFile(&fdFrom);
+
+_exit:
+  if (code) {
+    TSDB_ERROR_LOG(TD_VID(rtner->tsdb->pVnode), lino, code);
+    taosCloseFile(&fdFrom);
+  }
+  return code;
+}
+
 static int32_t tsdbDoMigrateFileObj(SRTNer *rtner, const STFileObj *fobj, const SDiskID *did) {
   int32_t  code = 0;
   int32_t  lino = 0;
@@ -123,6 +174,53 @@ _exit:
   return code;
 }
 
+static int32_t tsdbMigrateDataFileS3(SRTNer *rtner, const STFileObj *fobj, const SDiskID *did) {
+  int32_t  code = 0;
+  int32_t  lino = 0;
+  STFileOp op = {0};
+
+  // remove old
+  op = (STFileOp){
+      .optype = TSDB_FOP_REMOVE,
+      .fid = fobj->f->fid,
+      .of = fobj->f[0],
+  };
+
+  code = TARRAY2_APPEND(rtner->fopArr, op);
+  TSDB_CHECK_CODE(code, lino, _exit);
+
+  // create new
+  op = (STFileOp){
+      .optype = TSDB_FOP_CREATE,
+      .fid = fobj->f->fid,
+      .nf =
+          {
+              .type = fobj->f->type,
+              .did = did[0],
+              .fid = fobj->f->fid,
+              .cid = fobj->f->cid,
+              .size = fobj->f->size,
+              .stt[0] =
+                  {
+                      .level = fobj->f->stt[0].level,
+                  },
+          },
+  };
+
+  code = TARRAY2_APPEND(rtner->fopArr, op);
+  TSDB_CHECK_CODE(code, lino, _exit);
+
+  // do copy the file
+  code = tsdbCopyFileS3(rtner, fobj, &op.nf);
+  TSDB_CHECK_CODE(code, lino, _exit);
+
+_exit:
+  if (code) {
+    TSDB_ERROR_LOG(TD_VID(rtner->tsdb->pVnode), lino, code);
+  }
+  return code;
+}
+
 typedef struct {
   STsdb  *tsdb;
   int32_t sync;
@@ -201,9 +299,15 @@ static int32_t tsdbDoRetention2(void *arg) {
       for (int32_t ftype = 0; (ftype < TSDB_FTYPE_MAX) && (fobj = rtner->ctx->fset->farr[ftype], 1); ++ftype) {
         if (fobj == NULL) continue;
 
+        int32_t nlevel = tfsGetLevel(rtner->tsdb->pVnode->pTfs);
+        if (tsS3Enabled && nlevel > 1 && TSDB_FTYPE_DATA == ftype && fobj->f->did.level == nlevel - 1) {
+          code = tsdbRemoveFileObjectS3(rtner, fobj);
+          TSDB_CHECK_CODE(code, lino, _exit);
+        } else {
           code = tsdbDoRemoveFileObject(rtner, fobj);
           TSDB_CHECK_CODE(code, lino, _exit);
         }
+      }
 
       SSttLvl *lvl;
       TARRAY2_FOREACH(rtner->ctx->fset->lvlArr, lvl) {
@@ -228,9 +332,16 @@ static int32_t tsdbDoRetention2(void *arg) {
         if (fobj == NULL) continue;
 
         if (fobj->f->did.level == did.level) continue;
+
+        int32_t nlevel = tfsGetLevel(rtner->tsdb->pVnode->pTfs);
+        if (tsS3Enabled && nlevel > 1 && TSDB_FTYPE_DATA == ftype && did.level == nlevel - 1) {
+          code = tsdbMigrateDataFileS3(rtner, fobj, &did);
+          TSDB_CHECK_CODE(code, lino, _exit);
+        } else {
           code = tsdbDoMigrateFileObj(rtner, fobj, &did);
           TSDB_CHECK_CODE(code, lino, _exit);
         }
+      }
 
       // stt
       SSttLvl *lvl;

source/dnode/vnode/src/tsdb/tsdbWrite.c

@@ -76,7 +76,7 @@ int tsdbScanAndConvertSubmitMsg(STsdb *pTsdb, SSubmitReq2 *pMsg) {
   int32_t       code = 0;
   STsdbKeepCfg *pCfg = &pTsdb->keepCfg;
   TSKEY         now = taosGetTimestamp(pCfg->precision);
-  TSKEY         minKey = now - tsTickPerMin[pCfg->precision] * pCfg->keep2;
+  TSKEY         minKey = now - tsTickPerMin[pCfg->precision] * pCfg->keep1;
   TSKEY         maxKey = tsMaxKeyByPrecision[pCfg->precision];
   int32_t       size = taosArrayGetSize(pMsg->aSubmitTbData);
 

source/dnode/vnode/src/vnd/vnodeCos.c

@@ -0,0 +1,323 @@
+#define ALLOW_FORBID_FUNC
+
+#include "vndCos.h"
+
+extern char tsS3Endpoint[];
+extern char tsS3AccessKeyId[];
+extern char tsS3AccessKeySecret[];
+extern char tsS3BucketName[];
+extern char tsS3AppId[];
+
+#ifdef USE_COS
+#include "cos_api.h"
+#include "cos_http_io.h"
+#include "cos_log.h"
+
+int32_t s3Init() {
+  if (cos_http_io_initialize(NULL, 0) != COSE_OK) {
+    return -1;
+  }
+
+  // set log level, default COS_LOG_WARN
+  cos_log_set_level(COS_LOG_WARN);
+
+  // set log output, default stderr
+  cos_log_set_output(NULL);
+
+  return 0;
+}
+
+void s3CleanUp() { cos_http_io_deinitialize(); }
+
+static void log_status(cos_status_t *s) {
+  cos_warn_log("status->code: %d", s->code);
+  if (s->error_code) cos_warn_log("status->error_code: %s", s->error_code);
+  if (s->error_msg) cos_warn_log("status->error_msg: %s", s->error_msg);
+  if (s->req_id) cos_warn_log("status->req_id: %s", s->req_id);
+}
+
+static void s3InitRequestOptions(cos_request_options_t *options, int is_cname) {
+  options->config = cos_config_create(options->pool);
+
+  cos_config_t *config = options->config;
+
+  cos_str_set(&config->endpoint, tsS3Endpoint);
+  cos_str_set(&config->access_key_id, tsS3AccessKeyId);
+  cos_str_set(&config->access_key_secret, tsS3AccessKeySecret);
+  cos_str_set(&config->appid, tsS3AppId);
+
+  config->is_cname = is_cname;
+
+  options->ctl = cos_http_controller_create(options->pool, 0);
+}
+
+int32_t s3PutObjectFromFile(const char *file_str, const char *object_str) {
+  int32_t                code = 0;
+  cos_pool_t            *p = NULL;
+  int                    is_cname = 0;
+  cos_status_t          *s = NULL;
+  cos_request_options_t *options = NULL;
+  cos_string_t           bucket, object, file;
+  cos_table_t           *resp_headers;
+  int                    traffic_limit = 0;
+
+  cos_pool_create(&p, NULL);
+  options = cos_request_options_create(p);
+  s3InitRequestOptions(options, is_cname);
+  cos_table_t *headers = NULL;
+  if (traffic_limit) {
+    // 限速值设置范围为819200 - 838860800，即100KB/s - 100MB/s，如果超出该范围将返回400错误
+    headers = cos_table_make(p, 1);
+    cos_table_add_int(headers, "x-cos-traffic-limit", 819200);
+  }
+  cos_str_set(&bucket, tsS3BucketName);
+  cos_str_set(&file, file_str);
+  cos_str_set(&object, object_str);
+  s = cos_put_object_from_file(options, &bucket, &object, &file, headers, &resp_headers);
+  log_status(s);
+
+  cos_pool_destroy(p);
+
+  if (s->code != 200) {
+    return code = s->code;
+  }
+
+  return code;
+}
+
+void s3DeleteObjects(const char *object_name[], int nobject) {
+  cos_pool_t            *p = NULL;
+  int                    is_cname = 0;
+  cos_string_t           bucket;
+  cos_table_t           *resp_headers = NULL;
+  cos_request_options_t *options = NULL;
+  cos_list_t             object_list;
+  cos_list_t             deleted_object_list;
+  int                    is_quiet = COS_TRUE;
+
+  cos_pool_create(&p, NULL);
+  options = cos_request_options_create(p);
+  s3InitRequestOptions(options, is_cname);
+  cos_str_set(&bucket, tsS3BucketName);
+
+  cos_list_init(&object_list);
+  cos_list_init(&deleted_object_list);
+
+  for (int i = 0; i < nobject; ++i) {
+    cos_object_key_t *content = cos_create_cos_object_key(p);
+    cos_str_set(&content->key, object_name[i]);
+    cos_list_add_tail(&content->node, &object_list);
+  }
+
+  cos_status_t *s = cos_delete_objects(options, &bucket, &object_list, is_quiet, &resp_headers, &deleted_object_list);
+  log_status(s);
+
+  cos_pool_destroy(p);
+
+  if (cos_status_is_ok(s)) {
+    cos_warn_log("delete objects succeeded\n");
+  } else {
+    cos_warn_log("delete objects failed\n");
+  }
+}
+
+bool s3Exists(const char *object_name) {
+  bool                      ret = false;
+  cos_pool_t               *p = NULL;
+  int                       is_cname = 0;
+  cos_status_t             *s = NULL;
+  cos_request_options_t    *options = NULL;
+  cos_string_t              bucket;
+  cos_string_t              object;
+  cos_table_t              *resp_headers;
+  cos_table_t              *headers = NULL;
+  cos_object_exist_status_e object_exist;
+
+  cos_pool_create(&p, NULL);
+  options = cos_request_options_create(p);
+  s3InitRequestOptions(options, is_cname);
+  cos_str_set(&bucket, tsS3BucketName);
+  cos_str_set(&object, object_name);
+
+  s = cos_check_object_exist(options, &bucket, &object, headers, &object_exist, &resp_headers);
+  if (object_exist == COS_OBJECT_NON_EXIST) {
+    cos_warn_log("object: %.*s non exist.\n", object.len, object.data);
+  } else if (object_exist == COS_OBJECT_EXIST) {
+    ret = true;
+    cos_warn_log("object: %.*s exist.\n", object.len, object.data);
+  } else {
+    cos_warn_log("object: %.*s unknown status.\n", object.len, object.data);
+    log_status(s);
+  }
+
+  cos_pool_destroy(p);
+
+  return ret;
+}
+
+bool s3Get(const char *object_name, const char *path) {
+  bool                   ret = false;
+  cos_pool_t            *p = NULL;
+  int                    is_cname = 0;
+  cos_status_t          *s = NULL;
+  cos_request_options_t *options = NULL;
+  cos_string_t           bucket;
+  cos_string_t           object;
+  cos_string_t           file;
+  cos_table_t           *resp_headers = NULL;
+  cos_table_t           *headers = NULL;
+  int                    traffic_limit = 0;
+
+  //创建内存池
+  cos_pool_create(&p, NULL);
+
+  //初始化请求选项
+  options = cos_request_options_create(p);
+  s3InitRequestOptions(options, is_cname);
+  cos_str_set(&bucket, tsS3BucketName);
+  if (traffic_limit) {
+    //限速值设置范围为819200 - 838860800，即100KB/s - 100MB/s，如果超出该范围将返回400错误
+    headers = cos_table_make(p, 1);
+    cos_table_add_int(headers, "x-cos-traffic-limit", 819200);
+  }
+
+  //下载对象
+  cos_str_set(&file, path);
+  cos_str_set(&object, object_name);
+  s = cos_get_object_to_file(options, &bucket, &object, headers, NULL, &file, &resp_headers);
+  if (cos_status_is_ok(s)) {
+    ret = true;
+    cos_warn_log("get object succeeded\n");
+  } else {
+    cos_warn_log("get object failed\n");
+  }
+
+  //销毁内存池
+  cos_pool_destroy(p);
+
+  return ret;
+}
+
+typedef struct {
+  int64_t size;
+  int32_t atime;
+  char    name[TSDB_FILENAME_LEN];
+} SEvictFile;
+
+static int32_t evictFileCompareAsce(const void *pLeft, const void *pRight) {
+  SEvictFile *lhs = (SEvictFile *)pLeft;
+  SEvictFile *rhs = (SEvictFile *)pRight;
+  return lhs->atime < rhs->atime ? -1 : 1;
+}
+
+void s3EvictCache(const char *path, long object_size) {
+  SDiskSize disk_size = {0};
+  char      dir_name[TSDB_FILENAME_LEN] = "\0";
+
+  tstrncpy(dir_name, path, TSDB_FILENAME_LEN);
+  taosDirName(dir_name);
+
+  if (taosGetDiskSize((char *)dir_name, &disk_size) < 0) {
+    terrno = TAOS_SYSTEM_ERROR(errno);
+    vError("failed to get disk:%s size since %s", path, terrstr());
+    return;
+  }
+
+  if (object_size >= disk_size.avail - (1 << 30)) {
+    // evict too old files
+    // 1, list data files' atime under dir(path)
+    tdbDirPtr pDir = taosOpenDir(dir_name);
+    if (pDir == NULL) {
+      terrno = TAOS_SYSTEM_ERROR(errno);
+      vError("failed to open %s since %s", dir_name, terrstr());
+    }
+    SArray        *evict_files = taosArrayInit(16, sizeof(SEvictFile));
+    tdbDirEntryPtr pDirEntry;
+    while ((pDirEntry = taosReadDir(pDir)) != NULL) {
+      char *name = taosGetDirEntryName(pDirEntry);
+      if (!strncmp(name + strlen(name) - 5, ".data", 5)) {
+        SEvictFile e_file = {0};
+        char       entry_name[TSDB_FILENAME_LEN] = "\0";
+        int        dir_len = strlen(dir_name);
+
+        memcpy(e_file.name, dir_name, dir_len);
+        e_file.name[dir_len] = '/';
+        memcpy(e_file.name + dir_len + 1, name, strlen(name));
+
+        taosStatFile(e_file.name, &e_file.size, NULL, &e_file.atime);
+
+        taosArrayPush(evict_files, &e_file);
+      }
+    }
+    taosCloseDir(&pDir);
+
+    // 2, sort by atime
+    taosArraySort(evict_files, evictFileCompareAsce);
+
+    // 3, remove files ascendingly until we get enough object_size space
+    long   evict_size = 0;
+    size_t ef_size = TARRAY_SIZE(evict_files);
+    for (size_t i = 0; i < ef_size; ++i) {
+      SEvictFile *evict_file = taosArrayGet(evict_files, i);
+      taosRemoveFile(evict_file->name);
+      evict_size += evict_file->size;
+      if (evict_size >= object_size) {
+        break;
+      }
+    }
+
+    taosArrayDestroy(evict_files);
+  }
+}
+
+long s3Size(const char *object_name) {
+  long size = 0;
+
+  cos_pool_t            *p = NULL;
+  int                    is_cname = 0;
+  cos_status_t          *s = NULL;
+  cos_request_options_t *options = NULL;
+  cos_string_t           bucket;
+  cos_string_t           object;
+  cos_table_t           *resp_headers = NULL;
+
+  //创建内存池
+  cos_pool_create(&p, NULL);
+
+  //初始化请求选项
+  options = cos_request_options_create(p);
+  s3InitRequestOptions(options, is_cname);
+  cos_str_set(&bucket, tsS3BucketName);
+
+  //获取对象元数据
+  cos_str_set(&object, object_name);
+  s = cos_head_object(options, &bucket, &object, NULL, &resp_headers);
+  // print_headers(resp_headers);
+  if (cos_status_is_ok(s)) {
+    char *content_length_str = (char *)apr_table_get(resp_headers, COS_CONTENT_LENGTH);
+    if (content_length_str != NULL) {
+      size = atol(content_length_str);
+    }
+    cos_warn_log("head object succeeded: %ld\n", size);
+  } else {
+    cos_warn_log("head object failed\n");
+  }
+
+  //销毁内存池
+  cos_pool_destroy(p);
+
+  return size;
+}
+
+#else
+
+int32_t s3Init() { return 0; }
+void    s3CleanUp() {}
+int32_t s3PutObjectFromFile(const char *file, const char *object) { return 0; }
+void    s3DeleteObjects(const char *object_name[], int nobject) {}
+bool    s3Exists(const char *object_name) { return false; }
+bool    s3Get(const char *object_name, const char *path) { return false; }
+void    s3EvictCache(const char *path, long object_size) {}
+long    s3Size(const char *object_name) { return 0; }
+
+#endif

source/dnode/vnode/src/vnd/vnodeInitApi.c

@@ -96,6 +96,10 @@ void initMetadataAPI(SStoreMeta* pMeta) {
 
   pMeta->metaGetCachedTbGroup = metaGetCachedTbGroup;
   pMeta->metaPutTbGroupToCache = metaPutTbGroupToCache;
+
+  pMeta->openCtbCursor = metaOpenCtbCursor;
+  pMeta->closeCtbCursor = metaCloseCtbCursor;
+  pMeta->ctbCursorNext = metaCtbCursorNext;
 }
 
 void initTqAPI(SStoreTqReader* pTq) {

source/dnode/vnode/src/vnd/vnodeModule.c

@@ -14,6 +14,7 @@
  */
 
 #include "vnd.h"
+#include "vndCos.h"
 
 typedef struct SVnodeTask SVnodeTask;
 struct SVnodeTask {
@@ -81,6 +82,9 @@ int vnodeInit(int nthreads) {
   if (tqInit() < 0) {
     return -1;
   }
+  if (s3Init() < 0) {
+    return -1;
+  }
 
   return 0;
 }
@@ -112,6 +116,7 @@ void vnodeCleanup() {
   walCleanUp();
   tqCleanUp();
   smaCleanUp();
+  s3CleanUp();
 }
 
 int vnodeScheduleTaskEx(int tpid, int (*execute)(void*), void* arg) {

source/dnode/vnode/src/vnd/vnodeQuery.c

@@ -440,7 +440,7 @@ int32_t vnodeGetTableList(void* pVnode, int8_t type, SArray* pList) {
 }
 
 int32_t vnodeGetAllTableList(SVnode *pVnode, uint64_t uid, SArray *list) {
-  SMCtbCursor *pCur = metaOpenCtbCursor(pVnode->pMeta, uid, 1);
+  SMCtbCursor *pCur = metaOpenCtbCursor(pVnode, uid, 1);
 
   while (1) {
     tb_uid_t id = metaCtbCursorNext(pCur);
@@ -462,7 +462,7 @@ int32_t vnodeGetCtbIdListByFilter(SVnode *pVnode, int64_t suid, SArray *list, bo
 
 int32_t vnodeGetCtbIdList(void *pVnode, int64_t suid, SArray *list) {
   SVnode      *pVnodeObj = pVnode;
-  SMCtbCursor *pCur = metaOpenCtbCursor(pVnodeObj->pMeta, suid, 1);
+  SMCtbCursor *pCur = metaOpenCtbCursor(pVnodeObj, suid, 1);
 
   while (1) {
     tb_uid_t id = metaCtbCursorNext(pCur);
@@ -521,7 +521,7 @@ int32_t vnodeGetStbIdListByFilter(SVnode *pVnode, int64_t suid, SArray *list, bo
 }
 
 int32_t vnodeGetCtbNum(SVnode *pVnode, int64_t suid, int64_t *num) {
-  SMCtbCursor *pCur = metaOpenCtbCursor(pVnode->pMeta, suid, 0);
+  SMCtbCursor *pCur = metaOpenCtbCursor(pVnode, suid, 0);
   if (!pCur) {
     return TSDB_CODE_FAILED;
   }

source/dnode/vnode/src/vnd/vnodeSvr.c

@@ -628,6 +628,7 @@ int32_t vnodeProcessFetchMsg(SVnode *pVnode, SRpcMsg *pMsg, SQueueInfo *pInfo) {
       return tqProcessVgCommittedInfoReq(pVnode->pTq, pMsg);
     case TDMT_VND_TMQ_SEEK:
       return tqProcessSeekReq(pVnode->pTq, pMsg);
+
     default:
       vError("unknown msg type:%d in fetch queue", pMsg->msgType);
       return TSDB_CODE_APP_ERROR;

source/libs/command/src/explain.c

@@ -291,17 +291,17 @@ int32_t qExplainResNodeToRowsImpl(SExplainResNode *pResNode, SExplainCtx *ctx, i
   switch (pNode->type) {
     case QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN: {
       STagScanPhysiNode *pTagScanNode = (STagScanPhysiNode *)pNode;
-      EXPLAIN_ROW_NEW(level, EXPLAIN_TAG_SCAN_FORMAT, pTagScanNode->tableName.tname);
+      EXPLAIN_ROW_NEW(level, EXPLAIN_TAG_SCAN_FORMAT, pTagScanNode->scan.tableName.tname);
       EXPLAIN_ROW_APPEND(EXPLAIN_LEFT_PARENTHESIS_FORMAT);
       if (pResNode->pExecInfo) {
         QRY_ERR_RET(qExplainBufAppendExecInfo(pResNode->pExecInfo, tbuf, &tlen));
         EXPLAIN_ROW_APPEND(EXPLAIN_BLANK_FORMAT);
       }
-      if (pTagScanNode->pScanPseudoCols) {
+      if (pTagScanNode->scan.pScanPseudoCols) {
-        EXPLAIN_ROW_APPEND(EXPLAIN_PSEUDO_COLUMNS_FORMAT, pTagScanNode->pScanPseudoCols->length);
+        EXPLAIN_ROW_APPEND(EXPLAIN_PSEUDO_COLUMNS_FORMAT, pTagScanNode->scan.pScanPseudoCols->length);
         EXPLAIN_ROW_APPEND(EXPLAIN_BLANK_FORMAT);
       }
-      EXPLAIN_ROW_APPEND(EXPLAIN_WIDTH_FORMAT, pTagScanNode->node.pOutputDataBlockDesc->totalRowSize);
+      EXPLAIN_ROW_APPEND(EXPLAIN_WIDTH_FORMAT, pTagScanNode->scan.node.pOutputDataBlockDesc->totalRowSize);
       EXPLAIN_ROW_APPEND(EXPLAIN_RIGHT_PARENTHESIS_FORMAT);
       EXPLAIN_ROW_END();
       QRY_ERR_RET(qExplainResAppendRow(ctx, tbuf, tlen, level));
@@ -309,11 +309,11 @@ int32_t qExplainResNodeToRowsImpl(SExplainResNode *pResNode, SExplainCtx *ctx, i
       if (verbose) {
         EXPLAIN_ROW_NEW(level + 1, EXPLAIN_OUTPUT_FORMAT);
         EXPLAIN_ROW_APPEND(EXPLAIN_COLUMNS_FORMAT,
-                           nodesGetOutputNumFromSlotList(pTagScanNode->node.pOutputDataBlockDesc->pSlots));
+                           nodesGetOutputNumFromSlotList(pTagScanNode->scan.node.pOutputDataBlockDesc->pSlots));
         EXPLAIN_ROW_APPEND(EXPLAIN_BLANK_FORMAT);
-        EXPLAIN_ROW_APPEND(EXPLAIN_WIDTH_FORMAT, pTagScanNode->node.pOutputDataBlockDesc->outputRowSize);
+        EXPLAIN_ROW_APPEND(EXPLAIN_WIDTH_FORMAT, pTagScanNode->scan.node.pOutputDataBlockDesc->outputRowSize);
-        EXPLAIN_ROW_APPEND_LIMIT(pTagScanNode->node.pLimit);
+        EXPLAIN_ROW_APPEND_LIMIT(pTagScanNode->scan.node.pLimit);
-        EXPLAIN_ROW_APPEND_SLIMIT(pTagScanNode->node.pSlimit);
+        EXPLAIN_ROW_APPEND_SLIMIT(pTagScanNode->scan.node.pSlimit);
         EXPLAIN_ROW_END();
         QRY_ERR_RET(qExplainResAppendRow(ctx, tbuf, tlen, level + 1));
 

source/libs/executor/inc/executil.h

@@ -190,4 +190,6 @@ void printDataBlock(SSDataBlock* pBlock, const char* flag);
 void getNextTimeWindow(const SInterval* pInterval, STimeWindow* tw, int32_t order);
 void getInitialStartTimeWindow(SInterval* pInterval, TSKEY ts, STimeWindow* w, bool ascQuery);
 
+SSDataBlock* createTagValBlockForFilter(SArray* pColList, int32_t numOfTables, SArray* pUidTagList, void* pVnode,
+                                        SStorageAPI* pStorageAPI);
 #endif  // TDENGINE_EXECUTIL_H

source/libs/executor/inc/executorInt.h

@@ -251,6 +251,12 @@ typedef struct STableMergeScanInfo {
   SSortExecInfo   sortExecInfo;
 } STableMergeScanInfo;
 
+typedef struct STagScanFilterContext {
+  SHashObj* colHash;
+  int32_t   index;
+  SArray*   cInfoList;
+} STagScanFilterContext;
+
 typedef struct STagScanInfo {
   SColumnInfo*    pCols;
   SSDataBlock*    pRes;
@@ -259,6 +265,14 @@ typedef struct STagScanInfo {
   SLimitNode*     pSlimit;
   SReadHandle     readHandle;
   STableListInfo* pTableListInfo;
+  uint64_t        suid;
+  void*           pCtbCursor;
+  SNode*          pTagCond;
+  SNode*          pTagIndexCond;
+  STagScanFilterContext filterCtx;
+  SArray*         aUidTags; // SArray<STUidTagInfo>
+  SArray*         aFilterIdxs; // SArray<int32_t>
+  SStorageAPI*    pStorageAPI;
 } STagScanInfo;
 
 typedef enum EStreamScanMode {

source/libs/executor/inc/operator.h

@@ -81,7 +81,7 @@ SOperatorInfo* createTableScanOperatorInfo(STableScanPhysiNode* pTableScanNode,
 
 SOperatorInfo* createTableMergeScanOperatorInfo(STableScanPhysiNode* pTableScanNode, SReadHandle* readHandle, STableListInfo* pTableListInfo, SExecTaskInfo* pTaskInfo);
 
-SOperatorInfo* createTagScanOperatorInfo(SReadHandle* pReadHandle, STagScanPhysiNode* pPhyNode, STableListInfo* pTableListInfo, SExecTaskInfo* pTaskInfo);
+SOperatorInfo* createTagScanOperatorInfo(SReadHandle* pReadHandle, STagScanPhysiNode* pPhyNode, STableListInfo* pTableListInfo, SNode* pTagCond, SNode*pTagIndexCond, SExecTaskInfo* pTaskInfo);
 
 SOperatorInfo* createSysTableScanOperatorInfo(void* readHandle, SSystemTableScanPhysiNode* pScanPhyNode, const char* pUser, SExecTaskInfo* pTaskInfo);
 

source/libs/executor/inc/querytask.h

@@ -99,7 +99,7 @@ struct SExecTaskInfo {
 void           buildTaskId(uint64_t taskId, uint64_t queryId, char* dst);
 SExecTaskInfo* doCreateTask(uint64_t queryId, uint64_t taskId, int32_t vgId, EOPTR_EXEC_MODEL model, SStorageAPI* pAPI);
 void           doDestroyTask(SExecTaskInfo* pTaskInfo);
-bool           isTaskKilled(SExecTaskInfo* pTaskInfo);
+bool           isTaskKilled(void* pTaskInfo);
 void           setTaskKilled(SExecTaskInfo* pTaskInfo, int32_t rspCode);
 void           setTaskStatus(SExecTaskInfo* pTaskInfo, int8_t status);
 int32_t        createExecTaskInfo(SSubplan* pPlan, SExecTaskInfo** pTaskInfo, SReadHandle* pHandle, uint64_t taskId,

source/libs/executor/inc/tsort.h

@@ -191,7 +191,8 @@ int32_t getProperSortPageSize(size_t rowSize, uint32_t numOfCols);
 bool tsortIsClosed(SSortHandle* pHandle);
 void tsortSetClosed(SSortHandle* pHandle);
 
-void setSingleTableMerge(SSortHandle* pHandle);
+void tsortSetSingleTableMerge(SSortHandle* pHandle);
+void tsortSetAbortCheckFn(SSortHandle* pHandle, bool (*checkFn)(void* param), void* param);
 
 #ifdef __cplusplus
 }

source/libs/executor/src/executil.c

@@ -47,8 +47,6 @@ static int32_t optimizeTbnameInCondImpl(void* metaHandle, SArray* list, SNode* p
 
 static int32_t      getTableList(void* pVnode, SScanPhysiNode* pScanNode, SNode* pTagCond, SNode* pTagIndexCond,
                                  STableListInfo* pListInfo, uint8_t* digest, const char* idstr, SStorageAPI* pStorageAPI);
-static SSDataBlock* createTagValBlockForFilter(SArray* pColList, int32_t numOfTables, SArray* pUidTagList, void* pVnode,
-                                               SStorageAPI* pStorageAPI);
 
 static int64_t getLimit(const SNode* pLimit) { return NULL == pLimit ? -1 : ((SLimitNode*)pLimit)->limit; }
 static int64_t getOffset(const SNode* pLimit) { return NULL == pLimit ? -1 : ((SLimitNode*)pLimit)->offset; }
@@ -846,7 +844,7 @@ static int32_t optimizeTbnameInCondImpl(void* pVnode, SArray* pExistedUidList, S
   return -1;
 }
 
-static SSDataBlock* createTagValBlockForFilter(SArray* pColList, int32_t numOfTables, SArray* pUidTagList, void* pVnode,
+SSDataBlock* createTagValBlockForFilter(SArray* pColList, int32_t numOfTables, SArray* pUidTagList, void* pVnode,
                                                SStorageAPI* pStorageAPI) {
   SSDataBlock* pResBlock = createDataBlock();
   if (pResBlock == NULL) {

source/libs/executor/src/operator.c

@@ -370,17 +370,18 @@ SOperatorInfo* createOperator(SPhysiNode* pPhyNode, SExecTaskInfo* pTaskInfo, SR
       STableCountScanPhysiNode* pTblCountScanNode = (STableCountScanPhysiNode*)pPhyNode;
       pOperator = createTableCountScanOperatorInfo(pHandle, pTblCountScanNode, pTaskInfo);
     } else if (QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN == type) {
-      STagScanPhysiNode* pScanPhyNode = (STagScanPhysiNode*)pPhyNode;
+      STagScanPhysiNode* pTagScanPhyNode = (STagScanPhysiNode*)pPhyNode;
       STableListInfo*    pTableListInfo = tableListCreate();
-      int32_t            code = createScanTableListInfo(pScanPhyNode, NULL, false, pHandle, pTableListInfo, pTagCond,
+      if (!pTagScanPhyNode->onlyMetaCtbIdx) {
+        int32_t code = createScanTableListInfo((SScanPhysiNode*)pTagScanPhyNode, NULL, false, pHandle, pTableListInfo, pTagCond,
                                                pTagIndexCond, pTaskInfo);
         if (code != TSDB_CODE_SUCCESS) {
           pTaskInfo->code = code;
           qError("failed to getTableList, code: %s", tstrerror(code));
           return NULL;
         }
-
+      }
-      pOperator = createTagScanOperatorInfo(pHandle, pScanPhyNode, pTableListInfo, pTaskInfo);
+      pOperator = createTagScanOperatorInfo(pHandle, pTagScanPhyNode, pTableListInfo, pTagCond, pTagIndexCond, pTaskInfo);
     } else if (QUERY_NODE_PHYSICAL_PLAN_BLOCK_DIST_SCAN == type) {
       SBlockDistScanPhysiNode* pBlockNode = (SBlockDistScanPhysiNode*)pPhyNode;
       STableListInfo*          pTableListInfo = tableListCreate();

source/libs/executor/src/querytask.c

@@ -59,7 +59,7 @@ SExecTaskInfo* doCreateTask(uint64_t queryId, uint64_t taskId, int32_t vgId, EOP
   return pTaskInfo;
 }
 
-bool isTaskKilled(SExecTaskInfo* pTaskInfo) { return (0 != pTaskInfo->code); }
+bool isTaskKilled(void* pTaskInfo) { return (0 != ((SExecTaskInfo*)pTaskInfo)->code); }
 
 void setTaskKilled(SExecTaskInfo* pTaskInfo, int32_t rspCode) {
   pTaskInfo->code = rspCode;

source/libs/executor/src/scanoperator.c

@@ -504,7 +504,7 @@ int32_t addTagPseudoColumnData(SReadHandle* pHandle, const SExprInfo* pExpr, int
 
   // 1. check if it is existed in meta cache
   if (pCache == NULL) {
-    pHandle->api.metaReaderFn.initReader(&mr, pHandle->vnode, META_READER_NOLOCK, &pHandle->api.metaFn);
+    pHandle->api.metaReaderFn.initReader(&mr, pHandle->vnode, 0, &pHandle->api.metaFn);
     code = pHandle->api.metaReaderFn.getEntryGetUidCache(&mr, pBlock->info.id.uid);
     if (code != TSDB_CODE_SUCCESS) {
       // when encounter the TSDB_CODE_PAR_TABLE_NOT_EXIST error, we proceed.
@@ -2688,7 +2688,236 @@ static void doTagScanOneTable(SOperatorInfo* pOperator, const SSDataBlock* pRes,
   }
 }
 
-static SSDataBlock* doTagScan(SOperatorInfo* pOperator) {
+static void tagScanFreeUidTag(void* p) {
+  STUidTagInfo* pInfo = p;
+  if (pInfo->pTagVal != NULL) {
+    taosMemoryFree(pInfo->pTagVal);
+  }
+}
+
+static int32_t tagScanCreateResultData(SDataType* pType, int32_t numOfRows, SScalarParam* pParam) {
+  SColumnInfoData* pColumnData = taosMemoryCalloc(1, sizeof(SColumnInfoData));
+  if (pColumnData == NULL) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    return terrno;
+  }
+
+  pColumnData->info.type = pType->type;
+  pColumnData->info.bytes = pType->bytes;
+  pColumnData->info.scale = pType->scale;
+  pColumnData->info.precision = pType->precision;
+
+  int32_t code = colInfoDataEnsureCapacity(pColumnData, numOfRows, true);
+  if (code != TSDB_CODE_SUCCESS) {
+    terrno = code;
+    taosMemoryFree(pColumnData);
+    return terrno;
+  }
+
+  pParam->columnData = pColumnData;
+  pParam->colAlloced = true;
+  return TSDB_CODE_SUCCESS;
+}
+
+static EDealRes tagScanRewriteTagColumn(SNode** pNode, void* pContext) {
+  SColumnNode* pSColumnNode = NULL;
+  if (QUERY_NODE_COLUMN == nodeType((*pNode))) {
+    pSColumnNode = *(SColumnNode**)pNode;
+  } else if (QUERY_NODE_FUNCTION == nodeType((*pNode))) {
+    SFunctionNode* pFuncNode = *(SFunctionNode**)(pNode);
+    if (pFuncNode->funcType == FUNCTION_TYPE_TBNAME) {
+      pSColumnNode = (SColumnNode*)nodesMakeNode(QUERY_NODE_COLUMN);
+      if (NULL == pSColumnNode) {
+        return DEAL_RES_ERROR;
+      }
+      pSColumnNode->colId = -1;
+      pSColumnNode->colType = COLUMN_TYPE_TBNAME;
+      pSColumnNode->node.resType.type = TSDB_DATA_TYPE_VARCHAR;
+      pSColumnNode->node.resType.bytes = TSDB_TABLE_FNAME_LEN - 1 + VARSTR_HEADER_SIZE;
+      nodesDestroyNode(*pNode);
+      *pNode = (SNode*)pSColumnNode;
+    } else {
+      return DEAL_RES_CONTINUE;
+    }
+  } else {
+    return DEAL_RES_CONTINUE;
+  }
+
+  STagScanFilterContext* pCtx = (STagScanFilterContext*)pContext;
+  void*            data = taosHashGet(pCtx->colHash, &pSColumnNode->colId, sizeof(pSColumnNode->colId));
+  if (!data) {
+    taosHashPut(pCtx->colHash, &pSColumnNode->colId, sizeof(pSColumnNode->colId), pNode, sizeof((*pNode)));
+    pSColumnNode->slotId = pCtx->index++;
+    SColumnInfo cInfo = {.colId = pSColumnNode->colId,
+                         .type = pSColumnNode->node.resType.type,
+                         .bytes = pSColumnNode->node.resType.bytes};
+    taosArrayPush(pCtx->cInfoList, &cInfo);
+  } else {
+    SColumnNode* col = *(SColumnNode**)data;
+    pSColumnNode->slotId = col->slotId;
+  }
+
+  return DEAL_RES_CONTINUE;
+}
+
+
+static void tagScanFilterByTagCond(SArray* aUidTags, SNode* pTagCond, SArray* aFilterIdxs, void* pVnode, SStorageAPI* pAPI, STagScanInfo* pInfo) {
+  int32_t code = 0;
+  int32_t numOfTables = taosArrayGetSize(aUidTags);
+
+  SSDataBlock* pResBlock = createTagValBlockForFilter(pInfo->filterCtx.cInfoList, numOfTables, aUidTags, pVnode, pAPI);
+
+  SArray* pBlockList = taosArrayInit(1, POINTER_BYTES);
+  taosArrayPush(pBlockList, &pResBlock);
+  SDataType type = {.type = TSDB_DATA_TYPE_BOOL, .bytes = sizeof(bool)};
+
+  SScalarParam output = {0};
+  tagScanCreateResultData(&type, numOfTables, &output);
+
+  scalarCalculate(pTagCond, pBlockList, &output);
+
+  bool* result = (bool*)output.columnData->pData;
+  for (int32_t i = 0 ; i < numOfTables; ++i) {
+    if (result[i]) {
+      taosArrayPush(aFilterIdxs, &i);
+    }
+  }
+
+  colDataDestroy(output.columnData);
+  taosMemoryFreeClear(output.columnData);
+
+  blockDataDestroy(pResBlock);
+  taosArrayDestroy(pBlockList);
+
+
+}
+
+static void tagScanFillOneCellWithTag(const STUidTagInfo* pUidTagInfo, SExprInfo* pExprInfo, SColumnInfoData* pColInfo, int rowIndex, const SStorageAPI* pAPI, void* pVnode) {
+  if (fmIsScanPseudoColumnFunc(pExprInfo->pExpr->_function.functionId)) {  // tbname
+    char str[TSDB_TABLE_FNAME_LEN + VARSTR_HEADER_SIZE] = {0};
+//    if (pUidTagInfo->name != NULL) {
+//      STR_TO_VARSTR(str, pUidTagInfo->name);
+//    } else {  // name is not retrieved during filter
+//      pAPI->metaFn.getTableNameByUid(pVnode, pUidTagInfo->uid, str);
+//    }
+    STR_TO_VARSTR(str, "ctbidx");
+
+    colDataSetVal(pColInfo, rowIndex, str, false);
+  } else {
+    STagVal tagVal = {0};
+    tagVal.cid = pExprInfo->base.pParam[0].pCol->colId;
+    if (pUidTagInfo->pTagVal == NULL) {
+      colDataSetNULL(pColInfo, rowIndex);
+    } else {
+      const char* p = pAPI->metaFn.extractTagVal(pUidTagInfo->pTagVal, pColInfo->info.type, &tagVal);
+
+      if (p == NULL || (pColInfo->info.type == TSDB_DATA_TYPE_JSON && ((STag*)p)->nTag == 0)) {
+        colDataSetNULL(pColInfo, rowIndex);
+      } else if (pColInfo->info.type == TSDB_DATA_TYPE_JSON) {
+        colDataSetVal(pColInfo, rowIndex, p, false);
+      } else if (IS_VAR_DATA_TYPE(pColInfo->info.type)) {
+        char* tmp = taosMemoryMalloc(tagVal.nData + VARSTR_HEADER_SIZE + 1);
+        varDataSetLen(tmp, tagVal.nData);
+        memcpy(tmp + VARSTR_HEADER_SIZE, tagVal.pData, tagVal.nData);
+        colDataSetVal(pColInfo, rowIndex, tmp, false);
+        taosMemoryFree(tmp);
+      } else {
+        colDataSetVal(pColInfo, rowIndex, (const char*)&tagVal.i64, false);
+      }
+    }
+  }
+}
+
+static int32_t tagScanFillResultBlock(SOperatorInfo* pOperator, SSDataBlock* pRes, SArray* aUidTags, SArray* aFilterIdxs, bool ignoreFilterIdx,
+                                      SStorageAPI* pAPI) {
+  STagScanInfo* pInfo = pOperator->info;
+  SExprInfo*    pExprInfo = &pOperator->exprSupp.pExprInfo[0];
+  if (!ignoreFilterIdx) {
+    size_t szTables = taosArrayGetSize(aFilterIdxs);
+    for (int i = 0; i < szTables; ++i) {
+      int32_t idx = *(int32_t*)taosArrayGet(aFilterIdxs, i);
+      STUidTagInfo* pUidTagInfo = taosArrayGet(aUidTags, idx);
+      for (int32_t j = 0; j < pOperator->exprSupp.numOfExprs; ++j) {
+        SColumnInfoData* pDst = taosArrayGet(pRes->pDataBlock, pExprInfo[j].base.resSchema.slotId);
+        tagScanFillOneCellWithTag(pUidTagInfo, &pExprInfo[j], pDst, i, pAPI, pInfo->readHandle.vnode);
+      }
+    }
+  } else {
+    size_t szTables = taosArrayGetSize(aUidTags);
+    for (int i = 0; i < szTables; ++i) {
+      STUidTagInfo* pUidTagInfo = taosArrayGet(aUidTags, i);
+      for (int32_t j = 0; j < pOperator->exprSupp.numOfExprs; ++j) {
+        SColumnInfoData* pDst = taosArrayGet(pRes->pDataBlock, pExprInfo[j].base.resSchema.slotId);
+        tagScanFillOneCellWithTag(pUidTagInfo, &pExprInfo[j], pDst, i, pAPI, pInfo->readHandle.vnode);
+      }
+    }
+  }
+  return 0;
+}
+
+static SSDataBlock* doTagScanFromCtbIdx(SOperatorInfo* pOperator) {
+  if (pOperator->status == OP_EXEC_DONE) {
+    return NULL;
+  }
+  SExecTaskInfo* pTaskInfo = pOperator->pTaskInfo;
+  SStorageAPI* pAPI = &pTaskInfo->storageAPI;
+
+  STagScanInfo* pInfo = pOperator->info;
+  SSDataBlock*  pRes = pInfo->pRes;
+  blockDataCleanup(pRes);
+
+  if (pInfo->pCtbCursor == NULL) {
+    pInfo->pCtbCursor = pAPI->metaFn.openCtbCursor(pInfo->readHandle.vnode, pInfo->suid, 1);
+  }
+
+  SArray* aUidTags = pInfo->aUidTags;
+  SArray* aFilterIdxs = pInfo->aFilterIdxs;
+  int32_t count = 0;
+
+  while (1) {
+    taosArrayClearEx(aUidTags, tagScanFreeUidTag);
+    taosArrayClear(aFilterIdxs);
+
+    int32_t numTables = 0;
+    while (numTables < pOperator->resultInfo.capacity) {
+      SMCtbCursor* pCur = pInfo->pCtbCursor;
+      tb_uid_t     uid = pAPI->metaFn.ctbCursorNext(pInfo->pCtbCursor);
+      if (uid == 0) {
+        break;
+      }
+      STUidTagInfo info = {.uid = uid, .pTagVal = pCur->pVal};
+      info.pTagVal = taosMemoryMalloc(pCur->vLen);
+      memcpy(info.pTagVal, pCur->pVal, pCur->vLen);
+      taosArrayPush(aUidTags, &info);
+      ++numTables;
+    }
+
+    if (numTables == 0) {
+      break;
+    }
+    bool ignoreFilterIdx = true;
+    if (pInfo->pTagCond != NULL) {
+      ignoreFilterIdx = false;
+      tagScanFilterByTagCond(aUidTags, pInfo->pTagCond, aFilterIdxs, pInfo->readHandle.vnode, pAPI, pInfo);
+    } else {
+      ignoreFilterIdx = true;
+    }
+
+    tagScanFillResultBlock(pOperator, pRes, aUidTags, aFilterIdxs, ignoreFilterIdx, pAPI);
+    
+    count = ignoreFilterIdx ? taosArrayGetSize(aUidTags): taosArrayGetSize(aFilterIdxs);
+
+    if (count != 0) {
+      break;
+    }
+  }
+  
+  pRes->info.rows = count;
+  pOperator->resultInfo.totalRows += count;
+  return (pRes->info.rows == 0) ? NULL : pInfo->pRes;
+}
+
+static SSDataBlock* doTagScanFromMetaEntry(SOperatorInfo* pOperator) {
   if (pOperator->status == OP_EXEC_DONE) {
     return NULL;
   }
@@ -2746,14 +2975,23 @@ static SSDataBlock* doTagScan(SOperatorInfo* pOperator) {
 
 static void destroyTagScanOperatorInfo(void* param) {
   STagScanInfo* pInfo = (STagScanInfo*)param;
+  if (pInfo->pCtbCursor != NULL) {
+    pInfo->pStorageAPI->metaFn.closeCtbCursor(pInfo->pCtbCursor, 1);
+  }
+  taosHashCleanup(pInfo->filterCtx.colHash);
+  taosArrayDestroy(pInfo->filterCtx.cInfoList);
+  taosArrayDestroy(pInfo->aFilterIdxs);
+  taosArrayDestroyEx(pInfo->aUidTags, tagScanFreeUidTag);
+
   pInfo->pRes = blockDataDestroy(pInfo->pRes);
   taosArrayDestroy(pInfo->matchInfo.pList);
   pInfo->pTableListInfo = tableListDestroy(pInfo->pTableListInfo);
   taosMemoryFreeClear(param);
 }
 
-SOperatorInfo* createTagScanOperatorInfo(SReadHandle* pReadHandle, STagScanPhysiNode* pPhyNode,
+SOperatorInfo* createTagScanOperatorInfo(SReadHandle* pReadHandle, STagScanPhysiNode* pTagScanNode,
-                                         STableListInfo* pTableListInfo, SExecTaskInfo* pTaskInfo) {
+                                         STableListInfo* pTableListInfo, SNode* pTagCond, SNode* pTagIndexCond, SExecTaskInfo* pTaskInfo) {
+  SScanPhysiNode* pPhyNode = (SScanPhysiNode*)pTagScanNode;
   STagScanInfo*  pInfo = taosMemoryCalloc(1, sizeof(STagScanInfo));
   SOperatorInfo* pOperator = taosMemoryCalloc(1, sizeof(SOperatorInfo));
   if (pInfo == NULL || pOperator == NULL) {
@@ -2775,6 +3013,11 @@ SOperatorInfo* createTagScanOperatorInfo(SReadHandle* pReadHandle, STagScanPhysi
     goto _error;
   }
 
+  pInfo->pTagCond = pTagCond;
+  pInfo->pTagIndexCond = pTagIndexCond;
+  pInfo->suid = pPhyNode->suid;
+  pInfo->pStorageAPI = &pTaskInfo->storageAPI;
+
   pInfo->pTableListInfo = pTableListInfo;
   pInfo->pRes = createDataBlockFromDescNode(pDescNode);
   pInfo->readHandle = *pReadHandle;
@@ -2786,8 +3029,18 @@ SOperatorInfo* createTagScanOperatorInfo(SReadHandle* pReadHandle, STagScanPhysi
   initResultSizeInfo(&pOperator->resultInfo, 4096);
   blockDataEnsureCapacity(pInfo->pRes, pOperator->resultInfo.capacity);
 
+  if (pTagScanNode->onlyMetaCtbIdx) {
+    pInfo->aUidTags = taosArrayInit(pOperator->resultInfo.capacity, sizeof(STUidTagInfo));
+    pInfo->aFilterIdxs = taosArrayInit(pOperator->resultInfo.capacity, sizeof(int32_t));
+    pInfo->filterCtx.colHash = taosHashInit(4, taosGetDefaultHashFunction(TSDB_DATA_TYPE_SMALLINT), false, HASH_NO_LOCK);
+    pInfo->filterCtx.cInfoList = taosArrayInit(4, sizeof(SColumnInfo));
+    if (pInfo->pTagCond != NULL) {
+      nodesRewriteExprPostOrder(&pTagCond, tagScanRewriteTagColumn, (void*)&pInfo->filterCtx);
+    }
+  }
+  __optr_fn_t tagScanNextFn = (pTagScanNode->onlyMetaCtbIdx) ? doTagScanFromCtbIdx : doTagScanFromMetaEntry;
   pOperator->fpSet =
-      createOperatorFpSet(optrDummyOpenFn, doTagScan, NULL, destroyTagScanOperatorInfo, optrDefaultBufFn, NULL);
+      createOperatorFpSet(optrDummyOpenFn, tagScanNextFn, NULL, destroyTagScanOperatorInfo, optrDefaultBufFn, NULL);
 
   return pOperator;
 
@@ -2930,6 +3183,7 @@ int32_t startGroupTableMergeScan(SOperatorInfo* pOperator) {
                                               pInfo->pSortInputBlock, pTaskInfo->id.str, 0, 0, 0);
 
     tsortSetMergeLimit(pInfo->pSortHandle, mergeLimit);
+    tsortSetAbortCheckFn(pInfo->pSortHandle, isTaskKilled, pOperator->pTaskInfo);
   }
 
   tsortSetFetchRawDataFp(pInfo->pSortHandle, getBlockForTableMergeScan, NULL, NULL);
@@ -2949,7 +3203,7 @@ int32_t startGroupTableMergeScan(SOperatorInfo* pOperator) {
 
   int32_t code = TSDB_CODE_SUCCESS;
   if (numOfTable == 1) {
-    setSingleTableMerge(pInfo->pSortHandle);
+    tsortSetSingleTableMerge(pInfo->pSortHandle);
   } else {
     code = tsortOpen(pInfo->pSortHandle);
   }

source/libs/executor/src/sysscanoperator.c

@@ -1601,6 +1601,11 @@ static SSDataBlock* doSysTableScan(SOperatorInfo* pOperator) {
   SSysTableScanInfo* pInfo = pOperator->info;
   char               dbName[TSDB_DB_NAME_LEN] = {0};
 
+  if (isTaskKilled(pOperator->pTaskInfo)) {
+    setOperatorCompleted(pOperator);
+    return NULL;
+  }
+
   blockDataCleanup(pInfo->pRes);
 
   const char* name = tNameGetTableName(&pInfo->name);

source/libs/executor/src/tsort.c

@@ -71,12 +71,20 @@ struct SSortHandle {
   SMultiwayMergeTreeInfo* pMergeTree;
 
   bool singleTableMerge;
+
+  bool (*abortCheckFn)(void* param);
+  void* abortCheckParam;
 };
 
-void setSingleTableMerge(SSortHandle* pHandle) {
+void tsortSetSingleTableMerge(SSortHandle* pHandle) {
   pHandle->singleTableMerge = true;
 }
 
+void tsortSetAbortCheckFn(SSortHandle *pHandle, bool (*checkFn)(void *), void* param) {
+  pHandle->abortCheckFn = checkFn;
+  pHandle->abortCheckParam = param;
+}
+
 static int32_t msortComparFn(const void* pLeft, const void* pRight, void* param);
 
 // | offset[0] | offset[1] |....| nullbitmap | data |...|
@@ -726,11 +734,10 @@ static int32_t doInternalMergeSort(SSortHandle* pHandle) {
 
       SArray* pPageIdList = taosArrayInit(4, sizeof(int32_t));
       while (1) {
-        if (tsortIsClosed(pHandle)) {
+        if (tsortIsClosed(pHandle) || (pHandle->abortCheckFn && pHandle->abortCheckFn(pHandle->abortCheckParam))) {
           code = terrno = TSDB_CODE_TSC_QUERY_CANCELLED;
           return code;
         }
-
         SSDataBlock* pDataBlock = getSortedBlockDataInner(pHandle, &pHandle->cmpParam, numOfRows);
         if (pDataBlock == NULL) {
           break;

source/libs/function/src/udfd.c

@@ -848,7 +848,7 @@ int32_t udfdSaveFuncBodyToFile(SFuncInfo *pFuncInfo, SUdf *udf) {
 
   char path[PATH_MAX] = {0};
   udfdGetFuncBodyPath(udf, path);
-  bool fileExist = !(taosStatFile(path, NULL, NULL) < 0);
+  bool fileExist = !(taosStatFile(path, NULL, NULL, NULL) < 0);
   if (fileExist) {
     strncpy(udf->path, path, PATH_MAX);
     fnInfo("udfd func body file. reuse existing file %s", path);

source/libs/index/src/indexFstFile.c

@@ -162,7 +162,7 @@ static FORCE_INLINE int idxFileCtxGetSize(IFileCtx* ctx) {
       return ctx->offset;
     } else {
       int64_t file_size = 0;
-      taosStatFile(ctx->file.buf, &file_size, NULL);
+      taosStatFile(ctx->file.buf, &file_size, NULL, NULL);
       return (int)file_size;
     }
   }
@@ -199,7 +199,7 @@ IFileCtx* idxFileCtxCreate(WriterType type, const char* path, bool readOnly, int
       code = taosFtruncateFile(ctx->file.pFile, 0);
       UNUSED(code);
 
-      code = taosStatFile(path, &ctx->file.size, NULL);
+      code = taosStatFile(path, &ctx->file.size, NULL, NULL);
       UNUSED(code);
 
       ctx->file.wBufOffset = 0;

source/libs/nodes/src/nodesCloneFuncs.c

@@ -564,7 +564,9 @@ static int32_t physiScanCopy(const SScanPhysiNode* pSrc, SScanPhysiNode* pDst) {
 }
 
 static int32_t physiTagScanCopy(const STagScanPhysiNode* pSrc, STagScanPhysiNode* pDst) {
-  return physiScanCopy(pSrc, pDst);
+  COPY_BASE_OBJECT_FIELD(scan, physiScanCopy);
+  COPY_SCALAR_FIELD(onlyMetaCtbIdx);
+  return TSDB_CODE_SUCCESS;
 }
 
 static int32_t physiTableScanCopy(const STableScanPhysiNode* pSrc, STableScanPhysiNode* pDst) {

source/libs/nodes/src/nodesCodeFuncs.c

@@ -197,8 +197,10 @@ const char* nodesNodeName(ENodeType type) {
       return "ShowDnodesStmt";
     case QUERY_NODE_SHOW_MNODES_STMT:
       return "ShowMnodesStmt";
+/*
     case QUERY_NODE_SHOW_MODULES_STMT:
       return "ShowModulesStmt";
+*/
     case QUERY_NODE_SHOW_QNODES_STMT:
       return "ShowQnodesStmt";
     case QUERY_NODE_SHOW_SNODES_STMT:
@@ -660,6 +662,7 @@ static const char* jkScanLogicPlanDynamicScanFuncs = "DynamicScanFuncs";
 static const char* jkScanLogicPlanDataRequired = "DataRequired";
 static const char* jkScanLogicPlanTagCond = "TagCond";
 static const char* jkScanLogicPlanGroupTags = "GroupTags";
+static const char* jkScanLogicPlanOnlyMetaCtbIdx = "OnlyMetaCtbIdx";
 
 static int32_t logicScanNodeToJson(const void* pObj, SJson* pJson) {
   const SScanLogicNode* pNode = (const SScanLogicNode*)pObj;
@@ -701,7 +704,9 @@ static int32_t logicScanNodeToJson(const void* pObj, SJson* pJson) {
   if (TSDB_CODE_SUCCESS == code) {
     code = nodeListToJson(pJson, jkScanLogicPlanGroupTags, pNode->pGroupTags);
   }
-
+  if (TSDB_CODE_SUCCESS == code) {
+    code = tjsonAddBoolToObject(pJson, jkScanLogicPlanOnlyMetaCtbIdx, pNode->onlyMetaCtbIdx);
+  }
   return code;
 }
 
@@ -746,6 +751,9 @@ static int32_t jsonToLogicScanNode(const SJson* pJson, void* pObj) {
   if (TSDB_CODE_SUCCESS == code) {
     code = jsonToNodeList(pJson, jkScanLogicPlanGroupTags, &pNode->pGroupTags);
   }
+  if (TSDB_CODE_SUCCESS == code) {
+    code = tjsonGetBoolValue(pJson, jkScanLogicPlanOnlyMetaCtbIdx, &pNode->onlyMetaCtbIdx);
+  }
   
   return code;
 }
@@ -1562,7 +1570,7 @@ static const char* jkScanPhysiPlanTableName = "TableName";
 static const char* jkScanPhysiPlanGroupOrderScan = "GroupOrderScan";
 
 static int32_t physiScanNodeToJson(const void* pObj, SJson* pJson) {
-  const STagScanPhysiNode* pNode = (const STagScanPhysiNode*)pObj;
+  const SScanPhysiNode* pNode = (const SScanPhysiNode*)pObj;
 
   int32_t code = physicPlanNodeToJson(pObj, pJson);
   if (TSDB_CODE_SUCCESS == code) {
@@ -1591,7 +1599,7 @@ static int32_t physiScanNodeToJson(const void* pObj, SJson* pJson) {
 }
 
 static int32_t jsonToPhysiScanNode(const SJson* pJson, void* pObj) {
-  STagScanPhysiNode* pNode = (STagScanPhysiNode*)pObj;
+  SScanPhysiNode* pNode = (SScanPhysiNode*)pObj;
 
   int32_t code = jsonToPhysicPlanNode(pJson, pObj);
   if (TSDB_CODE_SUCCESS == code) {
@@ -1619,6 +1627,30 @@ static int32_t jsonToPhysiScanNode(const SJson* pJson, void* pObj) {
   return code;
 }
 
+static const char* jkTagScanPhysiOnlyMetaCtbIdx = "OnlyMetaCtbIdx";
+
+static int32_t physiTagScanNodeToJson(const void* pObj, SJson* pJson) {
+  const STagScanPhysiNode* pNode = (const STagScanPhysiNode*)pObj;
+
+  int32_t code = physiScanNodeToJson(pObj, pJson);
+
+  if (TSDB_CODE_SUCCESS == code) {
+    code = tjsonAddBoolToObject(pJson, jkTagScanPhysiOnlyMetaCtbIdx, pNode->onlyMetaCtbIdx);
+  }
+  return code;
+}
+
+static int32_t jsonToPhysiTagScanNode(const SJson* pJson, void* pObj) {
+  STagScanPhysiNode* pNode = (STagScanPhysiNode*)pObj;
+
+  int32_t code = jsonToPhysiScanNode(pJson, pObj);
+
+  if (TSDB_CODE_SUCCESS == code) {
+    code = tjsonGetBoolValue(pJson, jkTagScanPhysiOnlyMetaCtbIdx, &pNode->onlyMetaCtbIdx);
+  }
+  return code;
+}
+
 static const char* jkLastRowScanPhysiPlanGroupTags = "GroupTags";
 static const char* jkLastRowScanPhysiPlanGroupSort = "GroupSort";
 
@@ -6590,6 +6622,7 @@ static int32_t specificNodeToJson(const void* pObj, SJson* pJson) {
     case QUERY_NODE_LOGIC_PLAN:
       return logicPlanToJson(pObj, pJson);
     case QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN:
+      return physiTagScanNodeToJson(pObj, pJson);
     case QUERY_NODE_PHYSICAL_PLAN_BLOCK_DIST_SCAN:
       return physiScanNodeToJson(pObj, pJson);
     case QUERY_NODE_PHYSICAL_PLAN_LAST_ROW_SCAN:
@@ -6908,6 +6941,7 @@ static int32_t jsonToSpecificNode(const SJson* pJson, void* pObj) {
     case QUERY_NODE_LOGIC_PLAN:
       return jsonToLogicPlan(pJson, pObj);
     case QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN:
+      return jsonToPhysiTagScanNode(pJson, pObj);
     case QUERY_NODE_PHYSICAL_PLAN_BLOCK_DIST_SCAN:
     case QUERY_NODE_PHYSICAL_PLAN_TABLE_COUNT_SCAN:
       return jsonToPhysiScanNode(pJson, pObj);

source/libs/nodes/src/nodesMsgFuncs.c

@@ -2003,6 +2003,43 @@ static int32_t msgToPhysiScanNode(STlvDecoder* pDecoder, void* pObj) {
   return code;
 }
 
+enum {
+  PHY_TAG_SCAN_CODE_SCAN = 1,
+  PHY_TAG_SCAN_CODE_ONLY_META_CTB_IDX
+};
+
+static int32_t physiTagScanNodeToMsg(const void* pObj, STlvEncoder* pEncoder) {
+  const STagScanPhysiNode* pNode = (const STagScanPhysiNode*)pObj;
+
+  int32_t code = tlvEncodeObj(pEncoder, PHY_TAG_SCAN_CODE_SCAN, physiScanNodeToMsg, &pNode->scan);
+
+  if (TSDB_CODE_SUCCESS == code) {
+    code = tlvEncodeBool(pEncoder, PHY_TAG_SCAN_CODE_ONLY_META_CTB_IDX, pNode->onlyMetaCtbIdx);
+  }
+  return code;
+}
+
+static int32_t msgToPhysiTagScanNode(STlvDecoder* pDecoder, void* pObj) {
+  STagScanPhysiNode* pNode = (STagScanPhysiNode*)pObj;
+
+  int32_t code = TSDB_CODE_SUCCESS;
+  STlv*   pTlv = NULL;
+  tlvForEach(pDecoder, pTlv, code) {
+    switch (pTlv->type) {
+      case PHY_TAG_SCAN_CODE_SCAN:
+        code = tlvDecodeObjFromTlv(pTlv, msgToPhysiScanNode, &pNode->scan);
+        break;
+      case PHY_TAG_SCAN_CODE_ONLY_META_CTB_IDX:
+        code = tlvDecodeBool(pTlv, &pNode->onlyMetaCtbIdx);
+        break;
+      default:
+        break;
+    }
+  }
+
+  return code;
+}
+
 enum {
   PHY_LAST_ROW_SCAN_CODE_SCAN = 1,
   PHY_LAST_ROW_SCAN_CODE_GROUP_TAGS,
@@ -3726,6 +3763,8 @@ static int32_t specificNodeToMsg(const void* pObj, STlvEncoder* pEncoder) {
       code = caseWhenNodeToMsg(pObj, pEncoder);
       break;
     case QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN:
+      code = physiTagScanNodeToMsg(pObj, pEncoder);
+      break;
     case QUERY_NODE_PHYSICAL_PLAN_BLOCK_DIST_SCAN:
       code = physiScanNodeToMsg(pObj, pEncoder);
       break;
@@ -3869,6 +3908,8 @@ static int32_t msgToSpecificNode(STlvDecoder* pDecoder, void* pObj) {
       code = msgToCaseWhenNode(pDecoder, pObj);
       break;
     case QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN:
+      code = msgToPhysiTagScanNode(pDecoder, pObj);
+      break;
     case QUERY_NODE_PHYSICAL_PLAN_BLOCK_DIST_SCAN:
       code = msgToPhysiScanNode(pDecoder, pObj);
       break;

source/libs/nodes/src/nodesUtilFuncs.c

@@ -406,7 +406,7 @@ SNode* nodesMakeNode(ENodeType type) {
       return makeNode(type, sizeof(SRevokeStmt));
     case QUERY_NODE_SHOW_DNODES_STMT:
     case QUERY_NODE_SHOW_MNODES_STMT:
-    case QUERY_NODE_SHOW_MODULES_STMT:
+//    case QUERY_NODE_SHOW_MODULES_STMT:
     case QUERY_NODE_SHOW_QNODES_STMT:
     case QUERY_NODE_SHOW_SNODES_STMT:
     case QUERY_NODE_SHOW_BNODES_STMT:
@@ -982,7 +982,7 @@ void nodesDestroyNode(SNode* pNode) {
       break;
     case QUERY_NODE_SHOW_DNODES_STMT:
     case QUERY_NODE_SHOW_MNODES_STMT:
-    case QUERY_NODE_SHOW_MODULES_STMT:
+//    case QUERY_NODE_SHOW_MODULES_STMT:
     case QUERY_NODE_SHOW_QNODES_STMT:
     case QUERY_NODE_SHOW_SNODES_STMT:
     case QUERY_NODE_SHOW_BNODES_STMT:

source/libs/nodes/test/nodesCloneTest.cpp

@@ -199,9 +199,10 @@ TEST_F(NodesCloneTest, physiScan) {
     ASSERT_EQ(nodeType(pSrc), nodeType(pDst));
     STagScanPhysiNode* pSrcNode = (STagScanPhysiNode*)pSrc;
     STagScanPhysiNode* pDstNode = (STagScanPhysiNode*)pDst;
-    ASSERT_EQ(pSrcNode->uid, pDstNode->uid);
+    ASSERT_EQ(pSrcNode->scan.uid, pDstNode->scan.uid);
-    ASSERT_EQ(pSrcNode->suid, pDstNode->suid);
+    ASSERT_EQ(pSrcNode->scan.suid, pDstNode->scan.suid);
-    ASSERT_EQ(pSrcNode->tableType, pDstNode->tableType);
+    ASSERT_EQ(pSrcNode->scan.tableType, pDstNode->scan.tableType);
+    ASSERT_EQ(pSrcNode->onlyMetaCtbIdx, pDstNode->onlyMetaCtbIdx);
   });
 
   std::unique_ptr<SNode, void (*)(SNode*)> srcNode(nullptr, nodesDestroyNode);

source/libs/parser/src/parAstParser.c

@@ -690,8 +690,10 @@ static int32_t collectMetaKeyFromQuery(SCollectMetaKeyCxt* pCxt, SNode* pStmt) {
       return collectMetaKeyFromShowDnodes(pCxt, (SShowStmt*)pStmt);
     case QUERY_NODE_SHOW_MNODES_STMT:
       return collectMetaKeyFromShowMnodes(pCxt, (SShowStmt*)pStmt);
+/*
     case QUERY_NODE_SHOW_MODULES_STMT:
       return collectMetaKeyFromShowModules(pCxt, (SShowStmt*)pStmt);
+*/
     case QUERY_NODE_SHOW_QNODES_STMT:
       return collectMetaKeyFromShowQnodes(pCxt, (SShowStmt*)pStmt);
     case QUERY_NODE_SHOW_SNODES_STMT:

source/libs/parser/src/parAuthenticator.c

@@ -263,7 +263,7 @@ static int32_t authQuery(SAuthCxt* pCxt, SNode* pStmt) {
       return authAlterTable(pCxt, (SAlterTableStmt*)pStmt);
     case QUERY_NODE_SHOW_DNODES_STMT:
     case QUERY_NODE_SHOW_MNODES_STMT:
-    case QUERY_NODE_SHOW_MODULES_STMT:
+//    case QUERY_NODE_SHOW_MODULES_STMT:
     case QUERY_NODE_SHOW_QNODES_STMT:
     case QUERY_NODE_SHOW_SNODES_STMT:
     case QUERY_NODE_SHOW_BNODES_STMT:

source/libs/parser/src/parInsertSml.c

@@ -22,7 +22,7 @@ static void clearColValArray(SArray* pCols) {
   int32_t num = taosArrayGetSize(pCols);
   for (int32_t i = 0; i < num; ++i) {
     SColVal* pCol = taosArrayGet(pCols, i);
-    if (TSDB_DATA_TYPE_NCHAR == pCol->type) {
+    if (TSDB_DATA_TYPE_NCHAR == pCol->type || TSDB_DATA_TYPE_GEOMETRY == pCol->type) {
       taosMemoryFreeClear(pCol->value.pData);
     }
     pCol->flag = CV_FLAG_NONE;
@@ -237,9 +237,13 @@ int32_t smlBuildCol(STableDataCxt* pTableCxt, SSchema* schema, void* data, int32
     }
     pVal->value.pData = pUcs4;
     pVal->value.nData = len;
-  } else if (kv->type == TSDB_DATA_TYPE_BINARY || kv->type == TSDB_DATA_TYPE_GEOMETRY) {
+  } else if (kv->type == TSDB_DATA_TYPE_BINARY) {
     pVal->value.nData = kv->length;
     pVal->value.pData = (uint8_t*)kv->value;
+  } else if (kv->type == TSDB_DATA_TYPE_GEOMETRY) {
+    pVal->value.nData = kv->length;
+    pVal->value.pData = taosMemoryMalloc(kv->length);
+    memcpy(pVal->value.pData, (uint8_t*)kv->value, kv->length);
   } else {
     memcpy(&pVal->value.val, &(kv->value), kv->length);
   }
@@ -364,9 +368,13 @@ int32_t smlBindData(SQuery* query, bool dataFormat, SArray* tags, SArray* colsSc
         }
         pVal->value.pData = pUcs4;
         pVal->value.nData = len;
-      } else if (kv->type == TSDB_DATA_TYPE_BINARY || kv->type == TSDB_DATA_TYPE_GEOMETRY) {
+      } else if (kv->type == TSDB_DATA_TYPE_BINARY) {
         pVal->value.nData = kv->length;
         pVal->value.pData = (uint8_t*)kv->value;
+      } else if (kv->type == TSDB_DATA_TYPE_GEOMETRY) {
+        pVal->value.nData = kv->length;
+        pVal->value.pData = taosMemoryMalloc(kv->length);
+        memcpy(pVal->value.pData, (uint8_t*)kv->value, kv->length);
       } else {
         memcpy(&pVal->value.val, &(kv->value), kv->length);
       }

source/libs/parser/src/parInsertUtil.c

@@ -333,7 +333,7 @@ int32_t insGetTableDataCxt(SHashObj* pHash, void* id, int32_t idLen, STableMeta*
 
 static void destroyColVal(void* p) {
   SColVal* pVal = p;
-  if (TSDB_DATA_TYPE_NCHAR == pVal->type) {
+  if (TSDB_DATA_TYPE_NCHAR == pVal->type || TSDB_DATA_TYPE_GEOMETRY == pVal->type) {
     taosMemoryFree(pVal->value.pData);
   }
 }

source/libs/parser/src/parTranslater.c

@@ -92,6 +92,7 @@ static const SSysTableShowAdapter sysTableShowAdapter[] = {
     .numOfShowCols = 1,
     .pShowCols = {"*"}
   },
+/*  
   {
     .showType = QUERY_NODE_SHOW_MODULES_STMT,
     .pDbName = TSDB_INFORMATION_SCHEMA_DB,
@@ -99,6 +100,7 @@ static const SSysTableShowAdapter sysTableShowAdapter[] = {
     .numOfShowCols = 1,
     .pShowCols = {"*"}
   },
+*/  
   {
     .showType = QUERY_NODE_SHOW_QNODES_STMT,
     .pDbName = TSDB_INFORMATION_SCHEMA_DB,
@@ -278,7 +280,8 @@ static const SSysTableShowAdapter sysTableShowAdapter[] = {
 static int32_t  translateSubquery(STranslateContext* pCxt, SNode* pNode);
 static int32_t  translateQuery(STranslateContext* pCxt, SNode* pNode);
 static EDealRes translateValue(STranslateContext* pCxt, SValueNode* pVal);
-static int32_t createSimpleSelectStmtFromProjList(const char* pDb, const char* pTable, SNodeList* pProjectionList, SSelectStmt** pStmt);
+static int32_t  createSimpleSelectStmtFromProjList(const char* pDb, const char* pTable, SNodeList* pProjectionList,
+                                                   SSelectStmt** pStmt);
 static int32_t  createLastTsSelectStmt(char* pDb, char* pTable, STableMeta* pMeta, SNode** pQuery);
 static int32_t  setQuery(STranslateContext* pCxt, SQuery* pQuery);
 static int32_t  setRefreshMate(STranslateContext* pCxt, SQuery* pQuery);
@@ -772,7 +775,8 @@ static SNodeList* getProjectList(const SNode* pNode) {
 
 static bool isTimeLineQuery(SNode* pStmt) {
   if (QUERY_NODE_SELECT_STMT == nodeType(pStmt)) {
-    return (TIME_LINE_MULTI == ((SSelectStmt*)pStmt)->timeLineResMode) || (TIME_LINE_GLOBAL == ((SSelectStmt*)pStmt)->timeLineResMode);
+    return (TIME_LINE_MULTI == ((SSelectStmt*)pStmt)->timeLineResMode) ||
+           (TIME_LINE_GLOBAL == ((SSelectStmt*)pStmt)->timeLineResMode);
   } else if (QUERY_NODE_SET_OPERATOR == nodeType(pStmt)) {
     return TIME_LINE_GLOBAL == ((SSetOperator*)pStmt)->timeLineResMode;
   } else {
@@ -1608,9 +1612,11 @@ static int32_t translateInterpFunc(STranslateContext* pCxt, SFunctionNode* pFunc
     return generateSyntaxErrMsg(&pCxt->msgBuf, TSDB_CODE_PAR_NOT_ALLOWED_FUNC);
   }
 
-  if (pSelect->hasInterpFunc && (FUNC_RETURN_ROWS_INDEFINITE == pSelect->returnRows || pSelect->returnRows != fmGetFuncReturnRows(pFunc))) {
+  if (pSelect->hasInterpFunc &&
+      (FUNC_RETURN_ROWS_INDEFINITE == pSelect->returnRows || pSelect->returnRows != fmGetFuncReturnRows(pFunc))) {
     return generateSyntaxErrMsgExt(&pCxt->msgBuf, TSDB_CODE_PAR_NOT_ALLOWED_FUNC,
-                                "%s ignoring null value options cannot be used when applying to multiple columns", pFunc->functionName);
+                                   "%s ignoring null value options cannot be used when applying to multiple columns",
+                                   pFunc->functionName);
   }
 
   if (NULL != pSelect->pWindow || NULL != pSelect->pGroupByList) {
@@ -1648,7 +1654,8 @@ static int32_t translateTimelineFunc(STranslateContext* pCxt, SFunctionNode* pFu
   }
   SSelectStmt* pSelect = (SSelectStmt*)pCxt->pCurrStmt;
   if (NULL != pSelect->pFromTable && QUERY_NODE_TEMP_TABLE == nodeType(pSelect->pFromTable) &&
-      !isGlobalTimeLineQuery(((STempTableNode*)pSelect->pFromTable)->pSubquery) && !isTimeLineAlignedQuery(pCxt->pCurrStmt)) {
+      !isGlobalTimeLineQuery(((STempTableNode*)pSelect->pFromTable)->pSubquery) &&
+      !isTimeLineAlignedQuery(pCxt->pCurrStmt)) {
     return generateSyntaxErrMsgExt(&pCxt->msgBuf, TSDB_CODE_PAR_NOT_ALLOWED_FUNC,
                                    "%s function requires valid time series input", pFunc->functionName);
   }
@@ -2308,7 +2315,8 @@ static EDealRes doCheckExprForGroupBy(SNode** pNode, void* pContext) {
     }
   }
   if (isScanPseudoColumnFunc(*pNode) || QUERY_NODE_COLUMN == nodeType(*pNode)) {
-    if (pSelect->selectFuncNum > 1 || pSelect->hasOtherVectorFunc || !pSelect->hasSelectFunc || (isDistinctOrderBy(pCxt) && pCxt->currClause == SQL_CLAUSE_ORDER_BY)) {
+    if (pSelect->selectFuncNum > 1 || pSelect->hasOtherVectorFunc || !pSelect->hasSelectFunc ||
+        (isDistinctOrderBy(pCxt) && pCxt->currClause == SQL_CLAUSE_ORDER_BY)) {
       return generateDealNodeErrMsg(pCxt, getGroupByErrorCode(pCxt), ((SExprNode*)(*pNode))->userAlias);
     } else {
       return rewriteColToSelectValFunc(pCxt, pNode);
@@ -2711,7 +2719,8 @@ static int32_t replaceTbName(STranslateContext* pCxt, SSelectStmt* pSelect) {
   }
 
   SRealTableNode* pTable = (SRealTableNode*)pSelect->pFromTable;
-  if (TSDB_CHILD_TABLE != pTable->pMeta->tableType && TSDB_NORMAL_TABLE != pTable->pMeta->tableType && TSDB_SYSTEM_TABLE != pTable->pMeta->tableType) {
+  if (TSDB_CHILD_TABLE != pTable->pMeta->tableType && TSDB_NORMAL_TABLE != pTable->pMeta->tableType &&
+      TSDB_SYSTEM_TABLE != pTable->pMeta->tableType) {
     return TSDB_CODE_SUCCESS;
   }
 
@@ -3122,7 +3131,8 @@ static int32_t convertFillValue(STranslateContext* pCxt, SDataType dt, SNodeList
     code = scalarCalculateConstants(pCastFunc, &pCell->pNode);
   }
   if (TSDB_CODE_SUCCESS == code && QUERY_NODE_VALUE != nodeType(pCell->pNode)) {
-    code = generateSyntaxErrMsgExt(&pCxt->msgBuf, TSDB_CODE_PAR_WRONG_VALUE_TYPE, "Fill value can only accept constant");
+    code =
+        generateSyntaxErrMsgExt(&pCxt->msgBuf, TSDB_CODE_PAR_WRONG_VALUE_TYPE, "Fill value can only accept constant");
   } else if (TSDB_CODE_SUCCESS != code) {
     code = generateSyntaxErrMsgExt(&pCxt->msgBuf, TSDB_CODE_PAR_WRONG_VALUE_TYPE, "Filled data type mismatch");
   }
@@ -3588,7 +3598,6 @@ static int32_t createDefaultEveryNode(STranslateContext* pCxt, SNode** pOutput)
   pEvery->isDuration = true;
   pEvery->literal = taosStrdup("1s");
 
-
   *pOutput = (SNode*)pEvery;
   return TSDB_CODE_SUCCESS;
 }
@@ -3687,8 +3696,8 @@ static int32_t translatePartitionBy(STranslateContext* pCxt, SSelectStmt* pSelec
   if (pSelect->pPartitionByList) {
     int8_t typeType = getTableTypeFromTableNode(pSelect->pFromTable);
     SNode* pPar = nodesListGetNode(pSelect->pPartitionByList, 0);
-    if (!((TSDB_NORMAL_TABLE == typeType || TSDB_CHILD_TABLE == typeType) && 
+    if (!((TSDB_NORMAL_TABLE == typeType || TSDB_CHILD_TABLE == typeType) && 1 == pSelect->pPartitionByList->length &&
-      1 == pSelect->pPartitionByList->length && (QUERY_NODE_FUNCTION == nodeType(pPar) && FUNCTION_TYPE_TBNAME == ((SFunctionNode*)pPar)->funcType))) {
+          (QUERY_NODE_FUNCTION == nodeType(pPar) && FUNCTION_TYPE_TBNAME == ((SFunctionNode*)pPar)->funcType))) {
       pSelect->timeLineResMode = TIME_LINE_MULTI;
     }
 
@@ -3955,9 +3964,9 @@ static int32_t translateSetOperProject(STranslateContext* pCxt, SSetOperator* pS
     }
     snprintf(pRightExpr->aliasName, sizeof(pRightExpr->aliasName), "%s", pLeftExpr->aliasName);
     SNode* pProj = createSetOperProject(pSetOperator->stmtName, pLeft);
-    if (QUERY_NODE_COLUMN == nodeType(pLeft) && QUERY_NODE_COLUMN == nodeType(pRight) 
+    if (QUERY_NODE_COLUMN == nodeType(pLeft) && QUERY_NODE_COLUMN == nodeType(pRight) &&
-      && ((SColumnNode*)pLeft)->colId == PRIMARYKEY_TIMESTAMP_COL_ID
+        ((SColumnNode*)pLeft)->colId == PRIMARYKEY_TIMESTAMP_COL_ID &&
-      && ((SColumnNode*)pRight)->colId == PRIMARYKEY_TIMESTAMP_COL_ID) {
+        ((SColumnNode*)pRight)->colId == PRIMARYKEY_TIMESTAMP_COL_ID) {
       ((SColumnNode*)pProj)->colId = PRIMARYKEY_TIMESTAMP_COL_ID;
     }
     if (TSDB_CODE_SUCCESS != nodesListMakeStrictAppend(&pSetOperator->pProjectionList, pProj)) {
@@ -5737,7 +5746,6 @@ static int32_t translateRestoreDnode(STranslateContext* pCxt, SRestoreComponentN
   return buildCmdMsg(pCxt, TDMT_MND_RESTORE_DNODE, (FSerializeFunc)tSerializeSRestoreDnodeReq, &restoreReq);
 }
 
-
 static int32_t getSmaIndexDstVgId(STranslateContext* pCxt, const char* pDbName, const char* pTableName,
                                   int32_t* pVgId) {
   SVgroupInfo vg = {0};
@@ -5879,8 +5887,10 @@ int32_t createIntervalFromCreateSmaIndexStmt(SCreateIndexStmt* pStmt, SInterval*
   pInterval->interval = ((SValueNode*)pStmt->pOptions->pInterval)->datum.i;
   pInterval->intervalUnit = ((SValueNode*)pStmt->pOptions->pInterval)->unit;
   pInterval->offset = NULL != pStmt->pOptions->pOffset ? ((SValueNode*)pStmt->pOptions->pOffset)->datum.i : 0;
-  pInterval->sliding = NULL != pStmt->pOptions->pSliding ? ((SValueNode*)pStmt->pOptions->pSliding)->datum.i : pInterval->interval;
+  pInterval->sliding =
-  pInterval->slidingUnit = NULL != pStmt->pOptions->pSliding ? ((SValueNode*)pStmt->pOptions->pSliding)->unit : pInterval->intervalUnit;
+      NULL != pStmt->pOptions->pSliding ? ((SValueNode*)pStmt->pOptions->pSliding)->datum.i : pInterval->interval;
+  pInterval->slidingUnit =
+      NULL != pStmt->pOptions->pSliding ? ((SValueNode*)pStmt->pOptions->pSliding)->unit : pInterval->intervalUnit;
   pInterval->precision = pStmt->pOptions->tsPrecision;
   return TSDB_CODE_SUCCESS;
 }
@@ -6112,7 +6122,8 @@ static EDealRes checkColumnTagsInCond(SNode* pNode, void* pContext) {
   return DEAL_RES_CONTINUE;
 }
 
-static int32_t checkCollectTopicTags(STranslateContext* pCxt, SCreateTopicStmt* pStmt, STableMeta* pMeta, SNodeList** ppProjection) {
+static int32_t checkCollectTopicTags(STranslateContext* pCxt, SCreateTopicStmt* pStmt, STableMeta* pMeta,
+                                     SNodeList** ppProjection) {
   SBuildTopicContext colCxt = {.colExists = false, .colNotFound = false, .pMeta = pMeta, .pTags = NULL};
   nodesWalkExprPostOrder(pStmt->pWhere, checkColumnTagsInCond, &colCxt);
   if (colCxt.colNotFound) {
@@ -6554,7 +6565,8 @@ static int32_t checkStreamQuery(STranslateContext* pCxt, SCreateStreamStmt* pStm
     return generateSyntaxErrMsgExt(&pCxt->msgBuf, TSDB_CODE_PAR_INVALID_STREAM_QUERY,
                                    "SUBTABLE expression must be of VARCHAR type");
   }
-  if (NULL != pSelect->pSubtable && 0 == LIST_LENGTH(pSelect->pPartitionByList) && subtableExprHasColumnOrPseudoColumn(pSelect->pSubtable)) {
+  if (NULL != pSelect->pSubtable && 0 == LIST_LENGTH(pSelect->pPartitionByList) &&
+      subtableExprHasColumnOrPseudoColumn(pSelect->pSubtable)) {
     return generateSyntaxErrMsgExt(&pCxt->msgBuf, TSDB_CODE_PAR_INVALID_STREAM_QUERY,
                                    "SUBTABLE expression must not has column when no partition by clause");
   }
@@ -7136,7 +7148,6 @@ int32_t translatePostCreateStream(SParseContext* pParseCxt, SQuery* pQuery, void
   return code;
 }
 
-
 static int32_t translateDropStream(STranslateContext* pCxt, SDropStreamStmt* pStmt) {
   SMDropStreamReq dropReq = {0};
   SName           name;
@@ -7167,7 +7178,7 @@ static int32_t translateResumeStream(STranslateContext* pCxt, SResumeStreamStmt*
 
 static int32_t readFromFile(char* pName, int32_t* len, char** buf) {
   int64_t filesize = 0;
-  if (taosStatFile(pName, &filesize, NULL) < 0) {
+  if (taosStatFile(pName, &filesize, NULL, NULL) < 0) {
     return TAOS_SYSTEM_ERROR(errno);
   }
 
@@ -7821,7 +7832,8 @@ static SNodeList* createProjectCols(int32_t ncols, const char* const pCols[]) {
   return pProjections;
 }
 
-static int32_t createSimpleSelectStmtImpl(const char* pDb, const char* pTable, SNodeList* pProjectionList, SSelectStmt** pStmt) {
+static int32_t createSimpleSelectStmtImpl(const char* pDb, const char* pTable, SNodeList* pProjectionList,
+                                          SSelectStmt** pStmt) {
   SSelectStmt* pSelect = (SSelectStmt*)nodesMakeNode(QUERY_NODE_SELECT_STMT);
   if (NULL == pSelect) {
     return TSDB_CODE_OUT_OF_MEMORY;
@@ -7844,7 +7856,6 @@ static int32_t createSimpleSelectStmtImpl(const char* pDb, const char* pTable, S
   return TSDB_CODE_SUCCESS;
 }
 
-
 static int32_t createSimpleSelectStmtFromCols(const char* pDb, const char* pTable, int32_t numOfProjs,
                                               const char* const pProjCol[], SSelectStmt** pStmt) {
   SNodeList* pProjectionList = NULL;
@@ -7858,13 +7869,15 @@ static int32_t createSimpleSelectStmtFromCols(const char* pDb, const char* pTabl
   return createSimpleSelectStmtImpl(pDb, pTable, pProjectionList, pStmt);
 }
 
-static int32_t createSimpleSelectStmtFromProjList(const char* pDb, const char* pTable, SNodeList* pProjectionList, SSelectStmt** pStmt) {
+static int32_t createSimpleSelectStmtFromProjList(const char* pDb, const char* pTable, SNodeList* pProjectionList,
+                                                  SSelectStmt** pStmt) {
   return createSimpleSelectStmtImpl(pDb, pTable, pProjectionList, pStmt);
 }
 
 static int32_t createSelectStmtForShow(ENodeType showType, SSelectStmt** pStmt) {
   const SSysTableShowAdapter* pShow = &sysTableShowAdapter[showType - SYSTABLE_SHOW_TYPE_OFFSET];
-  return createSimpleSelectStmtFromCols(pShow->pDbName, pShow->pTableName, pShow->numOfShowCols, pShow->pShowCols, pStmt);
+  return createSimpleSelectStmtFromCols(pShow->pDbName, pShow->pTableName, pShow->numOfShowCols, pShow->pShowCols,
+                                        pStmt);
 }
 
 static int32_t createSelectStmtForShowTableDist(SShowTableDistributedStmt* pStmt, SSelectStmt** pOutput) {
@@ -8002,8 +8015,8 @@ static int32_t createShowTableTagsProjections(SNodeList** pProjections, SNodeLis
 static int32_t rewriteShowStableTags(STranslateContext* pCxt, SQuery* pQuery) {
   SShowTableTagsStmt* pShow = (SShowTableTagsStmt*)pQuery->pRoot;
   SSelectStmt*        pSelect = NULL;
-  int32_t code = createSimpleSelectStmtFromCols(((SValueNode*)pShow->pDbName)->literal, ((SValueNode*)pShow->pTbName)->literal,
+  int32_t             code = createSimpleSelectStmtFromCols(((SValueNode*)pShow->pDbName)->literal,
-                                        -1, NULL, &pSelect);
+                                                            ((SValueNode*)pShow->pTbName)->literal, -1, NULL, &pSelect);
   if (TSDB_CODE_SUCCESS == code) {
     code = createShowTableTagsProjections(&pSelect->pProjectionList, &pShow->pTags);
   }
@@ -9164,7 +9177,7 @@ static int32_t rewriteQuery(STranslateContext* pCxt, SQuery* pQuery) {
     case QUERY_NODE_SHOW_USERS_STMT:
     case QUERY_NODE_SHOW_DNODES_STMT:
     case QUERY_NODE_SHOW_MNODES_STMT:
-    case QUERY_NODE_SHOW_MODULES_STMT:
+//    case QUERY_NODE_SHOW_MODULES_STMT:
     case QUERY_NODE_SHOW_QNODES_STMT:
     case QUERY_NODE_SHOW_FUNCTIONS_STMT:
     case QUERY_NODE_SHOW_INDEXES_STMT:

source/libs/planner/src/planOptimizer.c

@@ -2679,6 +2679,9 @@ static int32_t tagScanOptimize(SOptimizeContext* pCxt, SLogicSubplan* pLogicSubp
   }
   nodesDestroyNode((SNode*)pAgg);
   tagScanOptCloneAncestorSlimit((SLogicNode*)pScanNode);
+
+  pScanNode->onlyMetaCtbIdx = false;
+  
   pCxt->optimized = true;
   return TSDB_CODE_SUCCESS;
 }

source/libs/planner/src/planPhysiCreater.c

@@ -511,6 +511,20 @@ static int32_t createSimpleScanPhysiNode(SPhysiPlanContext* pCxt, SSubplan* pSub
   return createScanPhysiNodeFinalize(pCxt, pSubplan, pScanLogicNode, pScan, pPhyNode);
 }
 
+static int32_t createTagScanPhysiNode(SPhysiPlanContext* pCxt, SSubplan* pSubplan, SScanLogicNode* pScanLogicNode,
+                                         SPhysiNode** pPhyNode) {
+  STagScanPhysiNode* pScan =
+      (STagScanPhysiNode*)makePhysiNode(pCxt, (SLogicNode*)pScanLogicNode, QUERY_NODE_PHYSICAL_PLAN_TAG_SCAN);
+  if (NULL == pScan) {
+    return TSDB_CODE_OUT_OF_MEMORY;
+  }
+  vgroupInfoToNodeAddr(pScanLogicNode->pVgroupList->vgroups, &pSubplan->execNode);
+  
+  pScan->onlyMetaCtbIdx = pScanLogicNode->onlyMetaCtbIdx;
+
+  return createScanPhysiNodeFinalize(pCxt, pSubplan, pScanLogicNode, (SScanPhysiNode*)pScan, pPhyNode);
+}
+
 static int32_t createLastRowScanPhysiNode(SPhysiPlanContext* pCxt, SSubplan* pSubplan, SScanLogicNode* pScanLogicNode,
                                           SPhysiNode** pPhyNode) {
   SLastRowScanPhysiNode* pScan =
@@ -646,6 +660,7 @@ static int32_t createScanPhysiNode(SPhysiPlanContext* pCxt, SSubplan* pSubplan,
   pCxt->hasScan = true;
   switch (pScanLogicNode->scanType) {
     case SCAN_TYPE_TAG:
+      return createTagScanPhysiNode(pCxt, pSubplan, pScanLogicNode, pPhyNode);
     case SCAN_TYPE_BLOCK_INFO:
       return createSimpleScanPhysiNode(pCxt, pSubplan, pScanLogicNode, pPhyNode);
     case SCAN_TYPE_TABLE_COUNT:

source/libs/sync/src/syncRaftStore.c

@@ -42,7 +42,7 @@ int32_t raftStoreReadFile(SSyncNode *pNode) {
   const char *file = pNode->raftStorePath;
   SRaftStore *pStore = &pNode->raftStore;
 
-  if (taosStatFile(file, NULL, NULL) < 0) {
+  if (taosStatFile(file, NULL, NULL, NULL) < 0) {
     sInfo("vgId:%d, raft store file:%s not exist, use default value", pNode->vgId, file);
     pStore->currentTerm = 0;
     pStore->voteFor.addr = 0;

source/libs/wal/src/walMeta.c

@@ -53,7 +53,7 @@ static FORCE_INLINE int64_t walScanLogGetLastVer(SWal* pWal, int32_t fileIdx) {
   walBuildLogName(pWal, pFileInfo->firstVer, fnameStr);
 
   int64_t fileSize = 0;
-  taosStatFile(fnameStr, &fileSize, NULL);
+  taosStatFile(fnameStr, &fileSize, NULL, NULL);
 
   TdFilePtr pFile = taosOpenFile(fnameStr, TD_FILE_READ | TD_FILE_WRITE);
   if (pFile == NULL) {
@@ -304,7 +304,7 @@ int walRepairLogFileTs(SWal* pWal, bool* updateMeta) {
 
     walBuildLogName(pWal, pFileInfo->firstVer, fnameStr);
     int32_t mtime = 0;
-    if (taosStatFile(fnameStr, NULL, &mtime) < 0) {
+    if (taosStatFile(fnameStr, NULL, &mtime, NULL) < 0) {
       terrno = TAOS_SYSTEM_ERROR(errno);
       wError("vgId:%d, failed to stat file due to %s, file:%s", pWal->cfg.vgId, strerror(errno), fnameStr);
       return -1;
@@ -353,7 +353,7 @@ int walTrimIdxFile(SWal* pWal, int32_t fileIdx) {
   walBuildIdxName(pWal, pFileInfo->firstVer, fnameStr);
 
   int64_t fileSize = 0;
-  taosStatFile(fnameStr, &fileSize, NULL);
+  taosStatFile(fnameStr, &fileSize, NULL, NULL);
   int64_t records = TMAX(0, pFileInfo->lastVer - pFileInfo->firstVer + 1);
   int64_t lastEndOffset = records * sizeof(SWalIdxEntry);
 
@@ -436,7 +436,7 @@ int walCheckAndRepairMeta(SWal* pWal) {
     SWalFileInfo* pFileInfo = taosArrayGet(pWal->fileInfoSet, fileIdx);
 
     walBuildLogName(pWal, pFileInfo->firstVer, fnameStr);
-    int32_t code = taosStatFile(fnameStr, &fileSize, NULL);
+    int32_t code = taosStatFile(fnameStr, &fileSize, NULL, NULL);
     if (code < 0) {
       terrno = TAOS_SYSTEM_ERROR(errno);
       wError("failed to stat file since %s. file:%s", terrstr(), fnameStr);
@@ -522,7 +522,7 @@ int walCheckAndRepairIdxFile(SWal* pWal, int32_t fileIdx) {
   walBuildLogName(pWal, pFileInfo->firstVer, fLogNameStr);
   int64_t fileSize = 0;
 
-  if (taosStatFile(fnameStr, &fileSize, NULL) < 0 && errno != ENOENT) {
+  if (taosStatFile(fnameStr, &fileSize, NULL, NULL) < 0 && errno != ENOENT) {
     wError("vgId:%d, failed to stat file due to %s. file:%s", pWal->cfg.vgId, strerror(errno), fnameStr);
     terrno = TAOS_SYSTEM_ERROR(errno);
     return -1;
@@ -935,7 +935,7 @@ int walLoadMeta(SWal* pWal) {
   walBuildMetaName(pWal, metaVer, fnameStr);
   // read metafile
   int64_t fileSize = 0;
-  taosStatFile(fnameStr, &fileSize, NULL);
+  taosStatFile(fnameStr, &fileSize, NULL, NULL);
   if (fileSize == 0) {
     (void)taosRemoveFile(fnameStr);
     wDebug("vgId:%d, wal find empty meta ver %d", pWal->cfg.vgId, metaVer);

source/libs/wal/src/walRead.c

@@ -16,10 +16,6 @@
 #include "taoserror.h"
 #include "walInt.h"
 
-static int32_t walFetchHeadNew(SWalReader *pRead, int64_t fetchVer);
-static int32_t walFetchBodyNew(SWalReader *pRead);
-static int32_t walSkipFetchBodyNew(SWalReader *pRead);
-
 SWalReader *walOpenReader(SWal *pWal, SWalFilterCond *cond) {
   SWalReader *pReader = taosMemoryCalloc(1, sizeof(SWalReader));
   if (pReader == NULL) {
@@ -70,38 +66,29 @@ int32_t walNextValidMsg(SWalReader *pReader) {
   int64_t fetchVer = pReader->curVersion;
   int64_t lastVer = walGetLastVer(pReader->pWal);
   int64_t committedVer = walGetCommittedVer(pReader->pWal);
-//  int64_t appliedVer = walGetAppliedVer(pReader->pWal);
+  int64_t appliedVer = walGetAppliedVer(pReader->pWal);
-
-//  if(appliedVer < committedVer){   // wait apply ver equal to commit ver, otherwise may lost data when consume data [TD-24010]
-//    wDebug("vgId:%d, wal apply ver:%"PRId64" smaller than commit ver:%"PRId64, pReader->pWal->cfg.vgId, appliedVer, committedVer);
-//  }
-
-//  int64_t endVer = TMIN(appliedVer, committedVer);
-  int64_t endVer = committedVer;
 
   wDebug("vgId:%d, wal start to fetch, index:%" PRId64 ", last index:%" PRId64 " commit index:%" PRId64
-         ", end index:%" PRId64,
+         ", applied index:%" PRId64,
-         pReader->pWal->cfg.vgId, fetchVer, lastVer, committedVer, endVer);
+         pReader->pWal->cfg.vgId, fetchVer, lastVer, committedVer, appliedVer);
-
+  if (fetchVer > appliedVer){
-  if (fetchVer > endVer){
     terrno = TSDB_CODE_WAL_LOG_NOT_EXIST;
     return -1;
   }
-
+  while (fetchVer <= appliedVer) {
-  while (fetchVer <= endVer) {
+    if (walFetchHead(pReader, fetchVer) < 0) {
-    if (walFetchHeadNew(pReader, fetchVer) < 0) {
       return -1;
     }
 
     int32_t type = pReader->pHead->head.msgType;
     if (type == TDMT_VND_SUBMIT || ((type == TDMT_VND_DELETE) && (pReader->cond.deleteMsg == 1)) ||
         (IS_META_MSG(type) && pReader->cond.scanMeta)) {
-      if (walFetchBodyNew(pReader) < 0) {
+      if (walFetchBody(pReader) < 0) {
         return -1;
       }
       return 0;
     } else {
-      if (walSkipFetchBodyNew(pReader) < 0) {
+      if (walSkipFetchBody(pReader) < 0) {
         return -1;
       }
 
@@ -263,104 +250,8 @@ int32_t walReaderSeekVer(SWalReader *pReader, int64_t ver) {
   return 0;
 }
 
-void walSetReaderCapacity(SWalReader *pRead, int32_t capacity) { pRead->capacity = capacity; }
 
-static int32_t walFetchHeadNew(SWalReader *pRead, int64_t fetchVer) {
+int32_t walFetchHead(SWalReader *pRead, int64_t ver) {
-  int64_t contLen;
-  bool    seeked = false;
-
-  wDebug("vgId:%d, wal starts to fetch head, index:%" PRId64, pRead->pWal->cfg.vgId, fetchVer);
-
-  if (pRead->curVersion != fetchVer) {
-    if (walReaderSeekVer(pRead, fetchVer) < 0) {
-      return -1;
-    }
-    seeked = true;
-  }
-
-  while (1) {
-    contLen = taosReadFile(pRead->pLogFile, pRead->pHead, sizeof(SWalCkHead));
-    if (contLen == sizeof(SWalCkHead)) {
-      break;
-    } else if (contLen == 0 && !seeked) {
-      if(walReadSeekVerImpl(pRead, fetchVer) < 0){
-        return -1;
-      }
-      seeked = true;
-      continue;
-    } else {
-      if (contLen < 0) {
-        terrno = TAOS_SYSTEM_ERROR(errno);
-      } else {
-        terrno = TSDB_CODE_WAL_FILE_CORRUPTED;
-      }
-      return -1;
-    }
-  }
-//  pRead->curInvalid = 0;
-  return 0;
-}
-
-static int32_t walFetchBodyNew(SWalReader *pReader) {
-  SWalCont *pReadHead = &pReader->pHead->head;
-  int64_t   ver = pReadHead->version;
-
-  wDebug("vgId:%d, wal starts to fetch body, ver:%" PRId64 " ,len:%d, total", pReader->pWal->cfg.vgId, ver,
-         pReadHead->bodyLen);
-
-  if (pReader->capacity < pReadHead->bodyLen) {
-    SWalCkHead *ptr = (SWalCkHead *)taosMemoryRealloc(pReader->pHead, sizeof(SWalCkHead) + pReadHead->bodyLen);
-    if (ptr == NULL) {
-      terrno = TSDB_CODE_OUT_OF_MEMORY;
-      return -1;
-    }
-
-    pReader->pHead = ptr;
-    pReadHead = &pReader->pHead->head;
-    pReader->capacity = pReadHead->bodyLen;
-  }
-
-  if (pReadHead->bodyLen != taosReadFile(pReader->pLogFile, pReadHead->body, pReadHead->bodyLen)) {
-    if (pReadHead->bodyLen < 0) {
-      terrno = TAOS_SYSTEM_ERROR(errno);
-      wError("vgId:%d, wal fetch body error:%" PRId64 ", read request index:%" PRId64 ", since %s",
-             pReader->pWal->cfg.vgId, pReader->pHead->head.version, ver, tstrerror(terrno));
-    } else {
-      wError("vgId:%d, wal fetch body error:%" PRId64 ", read request index:%" PRId64 ", since file corrupted",
-             pReader->pWal->cfg.vgId, pReader->pHead->head.version, ver);
-      terrno = TSDB_CODE_WAL_FILE_CORRUPTED;
-    }
-    return -1;
-  }
-
-  if (walValidBodyCksum(pReader->pHead) != 0) {
-    wError("vgId:%d, wal fetch body error:%" PRId64 ", since body checksum not passed", pReader->pWal->cfg.vgId, ver);
-    terrno = TSDB_CODE_WAL_FILE_CORRUPTED;
-    return -1;
-  }
-
-  wDebug("vgId:%d, index:%" PRId64 " is fetched, type:%d, cursor advance", pReader->pWal->cfg.vgId, ver, pReader->pHead->head.msgType);
-  pReader->curVersion = ver + 1;
-  return 0;
-}
-
-static int32_t walSkipFetchBodyNew(SWalReader *pRead) {
-  int64_t code;
-
-  code = taosLSeekFile(pRead->pLogFile, pRead->pHead->head.bodyLen, SEEK_CUR);
-  if (code < 0) {
-    terrno = TAOS_SYSTEM_ERROR(errno);
-//    pRead->curInvalid = 1;
-    return -1;
-  }
-
-  pRead->curVersion++;
-  wDebug("vgId:%d, version advance to %" PRId64 ", skip fetch", pRead->pWal->cfg.vgId, pRead->curVersion);
-
-  return 0;
-}
-
-int32_t walFetchHead(SWalReader *pRead, int64_t ver, SWalCkHead *pHead) {
   int64_t code;
   int64_t contLen;
   bool    seeked = false;
@@ -378,15 +269,13 @@ int32_t walFetchHead(SWalReader *pRead, int64_t ver, SWalCkHead *pHead) {
   if (pRead->curVersion != ver) {
     code = walReaderSeekVer(pRead, ver);
     if (code < 0) {
-//      pRead->curVersion = ver;
-//      pRead->curInvalid = 1;
       return -1;
     }
     seeked = true;
   }
 
   while (1) {
-    contLen = taosReadFile(pRead->pLogFile, pHead, sizeof(SWalCkHead));
+    contLen = taosReadFile(pRead->pLogFile, pRead->pHead, sizeof(SWalCkHead));
     if (contLen == sizeof(SWalCkHead)) {
       break;
     } else if (contLen == 0 && !seeked) {
@@ -401,12 +290,11 @@ int32_t walFetchHead(SWalReader *pRead, int64_t ver, SWalCkHead *pHead) {
       } else {
         terrno = TSDB_CODE_WAL_FILE_CORRUPTED;
       }
-//      pRead->curInvalid = 1;
       return -1;
     }
   }
 
-  code = walValidHeadCksum(pHead);
+  code = walValidHeadCksum(pRead->pHead);
 
   if (code != 0) {
     wError("vgId:%d, unexpected wal log index:%" PRId64 ", since head checksum not passed", pRead->pWal->cfg.vgId, ver);
@@ -414,32 +302,27 @@ int32_t walFetchHead(SWalReader *pRead, int64_t ver, SWalCkHead *pHead) {
     return -1;
   }
 
-//  pRead->curInvalid = 0;
   return 0;
 }
 
-int32_t walSkipFetchBody(SWalReader *pRead, const SWalCkHead *pHead) {
+int32_t walSkipFetchBody(SWalReader *pRead) {
-  int64_t code;
-
   wDebug("vgId:%d, skip fetch body %" PRId64 ", first ver:%" PRId64 ", commit ver:%" PRId64 ", last ver:%" PRId64
          ", applied ver:%" PRId64,
-         pRead->pWal->cfg.vgId, pHead->head.version, pRead->pWal->vers.firstVer, pRead->pWal->vers.commitVer,
+         pRead->pWal->cfg.vgId, pRead->pHead->head.version, pRead->pWal->vers.firstVer, pRead->pWal->vers.commitVer,
          pRead->pWal->vers.lastVer, pRead->pWal->vers.appliedVer);
 
-  code = taosLSeekFile(pRead->pLogFile, pHead->head.bodyLen, SEEK_CUR);
+  int64_t code = taosLSeekFile(pRead->pLogFile, pRead->pHead->head.bodyLen, SEEK_CUR);
   if (code < 0) {
     terrno = TAOS_SYSTEM_ERROR(errno);
-//    pRead->curInvalid = 1;
     return -1;
   }
 
   pRead->curVersion++;
-
   return 0;
 }
 
-int32_t walFetchBody(SWalReader *pRead, SWalCkHead **ppHead) {
+int32_t walFetchBody(SWalReader *pRead) {
-  SWalCont *pReadHead = &((*ppHead)->head);
+  SWalCont *pReadHead = &pRead->pHead->head;
   int64_t   ver = pReadHead->version;
 
   wDebug("vgId:%d, fetch body %" PRId64 ", first ver:%" PRId64 ", commit ver:%" PRId64 ", last ver:%" PRId64
@@ -448,13 +331,13 @@ int32_t walFetchBody(SWalReader *pRead, SWalCkHead **ppHead) {
          pRead->pWal->vers.appliedVer);
 
   if (pRead->capacity < pReadHead->bodyLen) {
-    SWalCkHead *ptr = (SWalCkHead *)taosMemoryRealloc(*ppHead, sizeof(SWalCkHead) + pReadHead->bodyLen);
+    SWalCkHead *ptr = (SWalCkHead *)taosMemoryRealloc(pRead->pHead, sizeof(SWalCkHead) + pReadHead->bodyLen);
     if (ptr == NULL) {
       terrno = TSDB_CODE_OUT_OF_MEMORY;
       return -1;
     }
-    *ppHead = ptr;
+    pRead->pHead = ptr;
-    pReadHead = &((*ppHead)->head);
+    pReadHead = &pRead->pHead->head;
     pRead->capacity = pReadHead->bodyLen;
   }
 
@@ -468,27 +351,24 @@ int32_t walFetchBody(SWalReader *pRead, SWalCkHead **ppHead) {
              pRead->pWal->cfg.vgId, pReadHead->version, ver);
       terrno = TSDB_CODE_WAL_FILE_CORRUPTED;
     }
-//    pRead->curInvalid = 1;
     return -1;
   }
 
   if (pReadHead->version != ver) {
     wError("vgId:%d, wal fetch body error, index:%" PRId64 ", read request index:%" PRId64, pRead->pWal->cfg.vgId,
            pReadHead->version, ver);
-//    pRead->curInvalid = 1;
     terrno = TSDB_CODE_WAL_FILE_CORRUPTED;
     return -1;
   }
 
-  if (walValidBodyCksum(*ppHead) != 0) {
+  if (walValidBodyCksum(pRead->pHead) != 0) {
     wError("vgId:%d, wal fetch body error, index:%" PRId64 ", since body checksum not passed", pRead->pWal->cfg.vgId,
            ver);
-//    pRead->curInvalid = 1;
     terrno = TSDB_CODE_WAL_FILE_CORRUPTED;
     return -1;
   }
 
-  pRead->curVersion = ver + 1;
+  pRead->curVersion++;
   return 0;
 }
 

source/os/src/osFile.c

@@ -191,7 +191,7 @@ int32_t taosRenameFile(const char *oldName, const char *newName) {
 #endif
 }
 
-int32_t taosStatFile(const char *path, int64_t *size, int32_t *mtime) {
+int32_t taosStatFile(const char *path, int64_t *size, int32_t *mtime, int32_t *atime) {
 #ifdef WINDOWS
   struct _stati64 fileStat;
   int32_t         code = _stati64(path, &fileStat);
@@ -211,6 +211,10 @@ int32_t taosStatFile(const char *path, int64_t *size, int32_t *mtime) {
     *mtime = fileStat.st_mtime;
   }
 
+  if (atime != NULL) {
+    *atime = fileStat.st_mtime;
+  }
+
   return 0;
 }
 int32_t taosDevInoFile(TdFilePtr pFile, int64_t *stDev, int64_t *stIno) {

source/util/src/tlog.c

@@ -17,8 +17,8 @@
 #include "tlog.h"
 #include "os.h"
 #include "tconfig.h"
-#include "tjson.h"
 #include "tglobal.h"
+#include "tjson.h"
 
 #define LOG_MAX_LINE_SIZE             (10024)
 #define LOG_MAX_LINE_BUFFER_SIZE      (LOG_MAX_LINE_SIZE + 3)
@@ -403,13 +403,13 @@ static int32_t taosOpenLogFile(char *fn, int32_t maxLines, int32_t maxFileNum) {
     strcpy(name, fn);
     strcat(name, ".0");
   }
-  bool log0Exist = taosStatFile(name, NULL, &logstat0_mtime) >= 0;
+  bool log0Exist = taosStatFile(name, NULL, &logstat0_mtime, NULL) >= 0;
 
   if (strlen(fn) < LOG_FILE_NAME_LEN + 50 - 2) {
     strcpy(name, fn);
     strcat(name, ".1");
   }
-  bool log1Exist = taosStatFile(name, NULL, &logstat1_mtime) >= 0;
+  bool log1Exist = taosStatFile(name, NULL, &logstat1_mtime, NULL) >= 0;
 
   // if none of the log files exist, open 0, if both exists, open the old one
   if (!log0Exist && !log1Exist) {
@@ -902,7 +902,7 @@ void taosReadCrashInfo(char* filepath, char** pMsg, int64_t* pMsgLen, TdFilePtr*
 
   if (NULL == *pFd) {
     int64_t filesize = 0;
-    if (taosStatFile(filepath, &filesize, NULL) < 0) {
+    if (taosStatFile(filepath, &filesize, NULL, NULL) < 0) {
       if (ENOENT == errno) {
         return;
       }
@@ -952,8 +952,8 @@ void taosReadCrashInfo(char* filepath, char** pMsg, int64_t* pMsgLen, TdFilePtr*
   readSize = taosReadFile(pFile, buf, msgLen);
   if (msgLen != readSize) {
     truncateFile = true;
-    taosPrintLog(flags, level, dflag, "failed to read file:%s,%p wlen:%" PRId64 " tlen:%" PRId64 " since %s", 
+    taosPrintLog(flags, level, dflag, "failed to read file:%s,%p wlen:%" PRId64 " tlen:%" PRId64 " since %s", filepath,
-        filepath, pFile, readSize, msgLen, terrstr());
+                 pFile, readSize, msgLen, terrstr());
     goto _return;
   }
 

tests/develop-test/2-query/table_count_scan.py

@@ -65,7 +65,7 @@ class TDTestCase:
 
         tdSql.query('select count(*),db_name, stable_name from information_schema.ins_tables group by db_name, stable_name;')
         tdSql.checkRows(3)
-        tdSql.checkData(0, 0, 24)
+        tdSql.checkData(0, 0, 23)
         tdSql.checkData(0, 1, 'information_schema')
         tdSql.checkData(0, 2, None)
         tdSql.checkData(1, 0, 3)
@@ -77,7 +77,7 @@ class TDTestCase:
 
         tdSql.query('select count(1) v,db_name, stable_name from information_schema.ins_tables group by db_name, stable_name order by v desc;')
         tdSql.checkRows(3)
-        tdSql.checkData(0, 0, 24)
+        tdSql.checkData(0, 0, 23)
         tdSql.checkData(0, 1, 'information_schema')
         tdSql.checkData(0, 2, None)
         tdSql.checkData(1, 0, 5)
@@ -93,7 +93,7 @@ class TDTestCase:
         tdSql.checkData(1, 1, 'performance_schema')
         tdSql.checkData(0, 0, 3)
         tdSql.checkData(0, 1, 'tbl_count')
-        tdSql.checkData(2, 0, 24)
+        tdSql.checkData(2, 0, 23)
         tdSql.checkData(2, 1, 'information_schema')
 
         tdSql.query("select count(*) from information_schema.ins_tables where db_name='tbl_count'")
@@ -106,7 +106,7 @@ class TDTestCase:
 
         tdSql.query('select count(*) from information_schema.ins_tables')
         tdSql.checkRows(1)
-        tdSql.checkData(0, 0, 32)
+        tdSql.checkData(0, 0, 31)
 
 
         tdSql.execute('create table stba (ts timestamp, c1 bool, c2 tinyint, c3 smallint, c4 int, c5 bigint, c6 float, c7 double, c8 binary(10), c9 nchar(10), c10 tinyint unsigned, c11 smallint unsigned, c12 int unsigned, c13 bigint unsigned) TAGS(t1 int, t2 binary(10), t3 double);')
@@ -189,7 +189,7 @@ class TDTestCase:
         tdSql.checkData(2, 0, 5)
         tdSql.checkData(2, 1, 'performance_schema')
         tdSql.checkData(2, 2, None)
-        tdSql.checkData(3, 0, 24)
+        tdSql.checkData(3, 0, 23)
         tdSql.checkData(3, 1, 'information_schema')
         tdSql.checkData(3, 2, None)
 
@@ -204,7 +204,7 @@ class TDTestCase:
         tdSql.checkData(2, 0, 5)
         tdSql.checkData(2, 1, 'performance_schema')
         tdSql.checkData(2, 2, None)
-        tdSql.checkData(3, 0, 24)
+        tdSql.checkData(3, 0, 23)
         tdSql.checkData(3, 1, 'information_schema')
         tdSql.checkData(3, 2, None)
 
@@ -215,7 +215,7 @@ class TDTestCase:
         tdSql.checkData(0, 1, 'tbl_count')
         tdSql.checkData(1, 0, 5)
         tdSql.checkData(1, 1, 'performance_schema')
-        tdSql.checkData(2, 0, 24)
+        tdSql.checkData(2, 0, 23)
         tdSql.checkData(2, 1, 'information_schema')
 
         tdSql.query("select count(*) from information_schema.ins_tables where db_name='tbl_count'")
@@ -228,7 +228,7 @@ class TDTestCase:
 
         tdSql.query('select count(*) from information_schema.ins_tables')
         tdSql.checkRows(1)
-        tdSql.checkData(0, 0, 33)
+        tdSql.checkData(0, 0, 32)
 
 
         tdSql.execute('drop database tbl_count')

tests/parallel_test/cases.task

@@ -127,6 +127,7 @@
 ,,y,system-test,./pytest.sh python3 ./test.py -f 7-tmq/dataFromTsdbNWal.py
 ,,y,system-test,./pytest.sh python3 ./test.py -f 7-tmq/dataFromTsdbNWal-multiCtb.py
 ,,y,system-test,./pytest.sh python3 ./test.py -f 7-tmq/tmq_taosx.py
+,,n,system-test,python3 ./test.py -f 7-tmq/tmq_offset.py
 ,,y,system-test,./pytest.sh python3 ./test.py -f 7-tmq/raw_block_interface_test.py
 ,,y,system-test,./pytest.sh python3 ./test.py -f 7-tmq/stbTagFilter-multiCtb.py
 ,,y,system-test,./pytest.sh python3 ./test.py -f 7-tmq/tmqSubscribeStb-r3.py -N 5
@@ -455,7 +456,7 @@
 ,,y,system-test,./pytest.sh python3 ./test.py -f 6-cluster/5dnode3mnodeRestartDnodeInsertDataAsync.py -N 6 -M 3
 #,,y,system-test,./pytest.sh python3 ./test.py -f 6-cluster/5dnode3mnodeRestartDnodeInsertDataAsync.py -N 6 -M 3 -n 3
 ,,n,system-test,python3 ./test.py -f 6-cluster/manually-test/6dnode3mnodeInsertLessDataAlterRep3to1to3.py -N 6 -M 3
-,,n,system-test,python ./test.py  -f 6-cluster/5dnode3mnodeRoll.py  -N 3  -C 1
+,,n,system-test,python3 ./test.py  -f 6-cluster/5dnode3mnodeRoll.py  -N 3  -C 1
 ,,y,system-test,./pytest.sh python3 ./test.py -f 6-cluster/5dnode3mnodeAdd1Ddnoe.py -N 7 -M 3 -C 6
 ,,y,system-test,./pytest.sh python3 ./test.py -f 6-cluster/5dnode3mnodeAdd1Ddnoe.py -N 7 -M 3 -C 6 -n 3
 #,,y,system-test,./pytest.sh python3 ./test.py -f 6-cluster/5dnode3mnodeDrop.py -N 5

tests/parallel_test/container_build.sh

@@ -60,6 +60,7 @@ docker run \
     -v /root/.cargo/git:/root/.cargo/git \
     -v /root/go/pkg/mod:/root/go/pkg/mod \
     -v /root/.cache/go-build:/root/.cache/go-build \
+    -v /root/.cos-local.1:/root/.cos-local.1 \
     -v ${REP_REAL_PATH}/enterprise/src/plugins/taosx/target:${REP_DIR}/enterprise/src/plugins/taosx/target \
     -v ${REP_REAL_PATH}/community/tools/taosws-rs/target:${REP_DIR}/community/tools/taosws-rs/target \
     -v ${REP_REAL_PATH}/community/contrib/cJson/:${REP_DIR}/community/contrib/cJson \
@@ -88,6 +89,7 @@ docker run \
     -v /root/.cargo/git:/root/.cargo/git \
     -v /root/go/pkg/mod:/root/go/pkg/mod \
     -v /root/.cache/go-build:/root/.cache/go-build \
+    -v /root/.cos-local.1:/root/.cos-local.1 \
     -v ${REP_REAL_PATH}/enterprise/src/plugins/taosx/target:${REP_DIR}/enterprise/src/plugins/taosx/target \
     -v ${REP_REAL_PATH}/community/tools/taosws-rs/target:${REP_DIR}/community/tools/taosws-rs/target \
     -v ${REP_REAL_PATH}/community/contrib/cJson/:${REP_DIR}/community/contrib/cJson \

tests/script/tsim/query/sys_tbname.sim

@@ -58,7 +58,7 @@ endi
 
 sql select tbname from information_schema.ins_tables;
 print $rows $data00
-if $rows != 33 then
+if $rows != 32 then
 	return -1
 endi
 if $data00 != @ins_tables@ then

tests/script/tsim/query/tableCount.sim

@@ -53,7 +53,7 @@ sql select stable_name,count(table_name) from information_schema.ins_tables grou
 if $rows != 3 then
   return -1
 endi
-if $data01 != 30 then
+if $data01 != 29 then
   return -1
 endi
 if $data11 != 10 then
@@ -72,7 +72,7 @@ endi
 if $data11 != 5 then
   return -1
 endi
-if $data21 != 24 then
+if $data21 != 23 then
   return -1
 endi
 if $data31 != 5 then
@@ -97,7 +97,7 @@ endi
 if $data42 != 3 then
   return -1
 endi
-if $data52 != 24 then
+if $data52 != 23 then
   return -1
 endi
 if $data62 != 5 then

tests/script/wtest.bat

@@ -17,6 +17,9 @@ rem echo SIM_DIR:    %SIM_DIR%
 set "TSIM_DIR=%SIM_DIR%tsim\"
 rem echo TSIM_DIR:   %TSIM_DIR%
 
+set "DATA_DIR=%TSIM_DIR%data\"
+rem echo DATA_DIR:   %DATA_DIR%
+
 set "CFG_DIR=%TSIM_DIR%cfg\"
 rem echo CFG_DIR:    %CFG_DIR%
 
@@ -30,25 +33,30 @@ if not exist %SIM_DIR%  mkdir %SIM_DIR%
 if not exist %TSIM_DIR% mkdir %TSIM_DIR%
 if exist %CFG_DIR% rmdir /s/q %CFG_DIR%
 if exist %LOG_DIR% rmdir /s/q %LOG_DIR%
+if exist %DATA_DIR% rmdir /s/q %DATA_DIR%
 if not exist %CFG_DIR% mkdir %CFG_DIR%
 if not exist %LOG_DIR% mkdir %LOG_DIR%
+if not exist %DATA_DIR% mkdir %DATA_DIR%
 
 set "fqdn=localhost"
 for /f "skip=1" %%A in (
   'wmic computersystem get caption'
 ) do if not defined fqdn set "fqdn=%%A"
 
-echo firstEp        %fqdn%        > %TAOS_CFG%
+echo firstEp        %fqdn%:7100   > %TAOS_CFG%
+echo secondEp       %fqdn%:7200   >> %TAOS_CFG%
 echo fqdn           %fqdn%        >> %TAOS_CFG%
 echo serverPort     7100          >> %TAOS_CFG%
+echo dataDir        %DATA_DIR%    >> %TAOS_CFG%
 echo logDir         %LOG_DIR%     >> %TAOS_CFG%
 echo scriptDir      %SCRIPT_DIR%  >> %TAOS_CFG%
 echo numOfLogLines  100000000     >> %TAOS_CFG%
 echo rpcDebugFlag   143           >> %TAOS_CFG%
 echo tmrDebugFlag   131           >> %TAOS_CFG%
-echo cDebugFlag     135           >> %TAOS_CFG%
+echo cDebugFlag     143           >> %TAOS_CFG%
 echo qDebugFlag     143           >> %TAOS_CFG%
-echo udebugFlag     135           >> %TAOS_CFG%
+echo uDebugFlag     143           >> %TAOS_CFG%
+echo debugFlag      143           >> %TAOS_CFG%
 echo wal            0             >> %TAOS_CFG%
 echo asyncLog       0             >> %TAOS_CFG%
 echo locale         en_US.UTF-8   >> %TAOS_CFG%

tests/system-test/0-others/information_schema.py

@@ -55,7 +55,7 @@ class TDTestCase:
             ]
         self.binary_str = 'taosdata'
         self.nchar_str = '涛思数据'
-        self.ins_list = ['ins_dnodes','ins_mnodes','ins_modules','ins_qnodes','ins_snodes','ins_cluster','ins_databases','ins_functions',\
+        self.ins_list = ['ins_dnodes','ins_mnodes','ins_qnodes','ins_snodes','ins_cluster','ins_databases','ins_functions',\
             'ins_indexes','ins_stables','ins_tables','ins_tags','ins_columns','ins_users','ins_grants','ins_vgroups','ins_configs','ins_dnode_variables',\
                 'ins_topics','ins_subscriptions','ins_streams','ins_stream_tasks','ins_vnodes','ins_user_privileges']
         self.perf_list = ['perf_connections','perf_queries','perf_consumers','perf_trans','perf_apps']

tests/system-test/2-query/sml.py

@@ -110,6 +110,11 @@ class TDTestCase:
 
         tdSql.query(f"select * from ts3724.`stb2.`")
         tdSql.checkRows(1)
+
+        # tdSql.query(f"select * from td24559.stb order by _ts")
+        # tdSql.checkRows(4)
+        # tdSql.checkData(0, 2, "POINT (4.343000 89.342000)")
+        # tdSql.checkData(3, 2, "GEOMETRYCOLLECTION (MULTIPOINT ((0.000000 0.000000), (1.000000 1.000000)), POINT (3.000000 4.000000), LINESTRING (2.000000 3.000000, 3.000000 4.000000))")
         return
 
     def run(self):

tests/system-test/6-cluster/5dnode3mnodeRoll.py

@@ -27,7 +27,7 @@ import threading
 import time
 import json
 
-BASEVERSION = "3.1.0.0"
+BASEVERSION = "3.1.1.0"
 
 class TDTestCase: