Merge branch '3.0' into fix/TD-18617

cmake/cmake.define

@@ -103,6 +103,9 @@ IF (TD_WINDOWS)
     SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${COMMON_FLAGS}")
 
 ELSE ()
+    IF (${TD_DARWIN})
+        set(CMAKE_MACOSX_RPATH 0)
+    ENDIF ()
     IF (${COVER} MATCHES "true")
         MESSAGE(STATUS "Test coverage mode, add extra flags")
         SET(GCC_COVERAGE_COMPILE_FLAGS "-fprofile-arcs -ftest-coverage")

cmake/taostools_CMakeLists.txt.in

@@ -2,7 +2,7 @@
 # taos-tools
 ExternalProject_Add(taos-tools
         GIT_REPOSITORY https://github.com/taosdata/taos-tools.git
-        GIT_TAG 2af2222
+        GIT_TAG 833b721
         SOURCE_DIR "${TD_SOURCE_DIR}/tools/taos-tools"
         BINARY_DIR ""
         #BUILD_IN_SOURCE TRUE

docs/en/05-get-started/01-docker.md

@@ -42,7 +42,7 @@ To do so, run the following command:
    
    ```
 
-   This command creates the `meters` supertable in the `test` database. In the `meters` supertable, it then creates 10,000 subtables named `d0` to `d9999`. Each table has 10,000 rows and each row has four columns: `ts`, `current`, `voltage`, and `phase`. The timestamps of the data in these columns range from 2017-07-14 10:40:00 000 to 2017-07-14 10:40:09 999. Each table is randomly assigned a `groupId` tag from 1 to ten and a `location` tag of either `California.SanFrancisco` or `California.SanDiego`.
+This command creates the `meters` supertable in the `test` database. In the `meters` supertable, it then creates 10,000 subtables named `d0` to `d9999`. Each table has 10,000 rows and each row has four columns: `ts`, `current`, `voltage`, and `phase`. The timestamps of the data in these columns range from 2017-07-14 10:40:00 000 to 2017-07-14 10:40:09 999. Each table is randomly assigned a `groupId` tag from 1 to 10 and a `location` tag of either `Campbell`, `Cupertino`, `Los Angeles`, `Mountain View`, `Palo Alto`, `San Diego`, `San Francisco`, `San Jose`, `Santa Clara` or `Sunnyvale`.
 
    The `taosBenchmark` command creates a deployment with 100 million data points that you can use for testing purposes. The time required depends on the hardware specifications of the local system.
 

docs/en/12-taos-sql/24-show.md

@@ -3,7 +3,7 @@ sidebar_label: SHOW Statement
 title: SHOW Statement for Metadata
 ---
 
-In addition to running SELECT statements on INFORMATION_SCHEMA, you can also use SHOW to obtain system metadata, information, and status.
+`SHOW` command can be used to get brief system information. To get details about metatadata, information, and status in the system, please use `select` to query the tables in database `INFORMATION_SCHEMA`. 
 
 ## SHOW ACCOUNTS
 

docs/en/14-reference/02-rest-api/02-rest-api.mdx

@@ -10,7 +10,7 @@ One difference from the native connector is that the REST interface is stateless
 
 ## Installation
 
-The REST interface does not rely on any TDengine native library, so the client application does not need to install any TDengine libraries. The client application's development language only needs to support the HTTP protocol.
+The REST interface does not rely on any TDengine native library, so the client application does not need to install any TDengine libraries. The client application's development language only needs to support the HTTP protocol. The REST interface is provided by [taosAdapter](../taosadapter), to use REST interface you need to make sure `taosAdapter` is running properly.
 
 ## Verification
 

docs/en/14-reference/03-connector/cpp.mdx → docs/en/14-reference/03-connector/03-cpp.mdx

 ---
-sidebar_position: 1
 sidebar_label: C/C++
 title: C/C++ Connector
 ---

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

 ---
 toc_max_heading_level: 4
-sidebar_position: 2
 sidebar_label: Java
 title: TDengine Java Connector
 description: The TDengine Java Connector is implemented on the standard JDBC API and provides native and REST connectors.

docs/en/14-reference/03-connector/go.mdx → docs/en/14-reference/03-connector/05-go.mdx

 ---
 toc_max_heading_level: 4
-sidebar_position: 4
 sidebar_label: Go
 title: TDengine Go Connector
 ---

docs/en/14-reference/03-connector/rust.mdx → docs/en/14-reference/03-connector/06-rust.mdx

 ---
 toc_max_heading_level: 4
-sidebar_position: 5
 sidebar_label: Rust
 title: TDengine Rust Connector
 ---

docs/en/14-reference/03-connector/python.mdx → docs/en/14-reference/03-connector/07-python.mdx

 ---
-sidebar_position: 3
 sidebar_label: Python
 title: TDengine Python Connector
 description: "taospy is the official Python connector for TDengine. taospy provides a rich API that makes it easy for Python applications to use TDengine. tasopy wraps both the native and REST interfaces of TDengine, corresponding to the two submodules of tasopy: taos and taosrest. In addition to wrapping the native and REST interfaces, taospy also provides a programming interface that conforms to the Python Data Access Specification (PEP 249), making it easy to integrate taospy with many third-party tools, such as SQLAlchemy and pandas."

docs/en/14-reference/03-connector/node.mdx → docs/en/14-reference/03-connector/08-node.mdx

 ---
 toc_max_heading_level: 4
-sidebar_position: 6
 sidebar_label: Node.js
 title: TDengine Node.js Connector
 ---

docs/en/14-reference/03-connector/csharp.mdx → docs/en/14-reference/03-connector/09-csharp.mdx

 ---
 toc_max_heading_level: 4
-sidebar_position: 7
 sidebar_label: C#
 title: C# Connector
 ---

docs/en/14-reference/03-connector/php.mdx → docs/en/14-reference/03-connector/10-php.mdx

 ---
-sidebar_position: 1
-sidebar_label: PHP (community contribution)
+sidebar_label: PHP
 title: PHP Connector
 ---
 

docs/zh/02-intro.md

@@ -22,8 +22,8 @@ TDengine的主要功能如下：
 9. 提供[命令行程序](../reference/taos-shell)，便于管理集群，检查系统状态，做即席查询
 10. 提供多种数据的[导入](../operation/import)、[导出](../operation/export)
 11. 支持对[TDengine 集群本身的监控](../operation/monitor)
-12. 提供 [C/C++](../reference/connector/cpp), [Java](../reference/connector/java), [Python](../reference/connector/python), [Go](../reference/connector/go), [Rust](../reference/connector/rust), [Node.js](../reference/connector/node) 等多种编程语言的[连接器](../reference/connector/)
-13. 支持 [REST 接口](../reference/rest-api/)
+12. 提供各种语言的[连接器](../connector): 如 C/C++, Java, Go, Node.JS, Rust, Python, C# 等
+13. 支持 [REST 接口](../connector/rest-api/)
 14. 支持与[ Grafana 无缝集成](../third-party/grafana)
 15. 支持与 Google Data Studio 无缝集成
 16. 支持 [Kubernetes 部署](../deployment/k8s)

docs/zh/05-get-started/03-package.md

@@ -9,7 +9,7 @@ import PkgListV3 from "/components/PkgListV3";
 
 您可以[用 Docker 立即体验](../../get-started/docker/) TDengine。如果您希望对 TDengine 贡献代码或对内部实现感兴趣，请参考我们的 [TDengine GitHub 主页](https://github.com/taosdata/TDengine) 下载源码构建和安装.
 
-TDengine 完整的软件包包括服务端（taosd）、用于与第三方系统对接并提供 RESTful 接口的 taosAdapter、应用驱动（taosc）、命令行程序 (CLI，taos) 和一些工具软件。目前 taosAdapter 仅在 Linux 系统上安装和运行，后续将支持 Windows、macOS 等系统。TDengine 除了提供多种语言的连接器之外，还通过 [taosAdapter](../../reference/taosadapter/) 提供 [RESTful 接口](../../reference/rest-api/)。
+TDengine 完整的软件包包括服务端（taosd）、用于与第三方系统对接并提供 RESTful 接口的 taosAdapter、应用驱动（taosc）、命令行程序 (CLI，taos) 和一些工具软件。目前 taosAdapter 仅在 Linux 系统上安装和运行，后续将支持 Windows、macOS 等系统。TDengine 除了提供多种语言的连接器之外，还通过 [taosAdapter](../../reference/taosadapter/) 提供 [RESTful 接口](../../connector/rest-api/)。
 
 为方便使用，标准的服务端安装包包含了 taosd、taosAdapter、taosc、taos、taosdump、taosBenchmark、TDinsight 安装脚本和示例代码；如果您只需要用到服务端程序和客户端连接的 C/C++ 语言支持，也可以仅下载 lite 版本的安装包。
 

docs/zh/05-get-started/index.md

@@ -3,7 +3,7 @@ title: 立即开始
 description: '快速设置 TDengine 环境并体验其高效写入和查询'
 ---
 
-TDengine 完整的软件包包括服务端（taosd）、用于与第三方系统对接并提供 RESTful 接口的 taosAdapter、应用驱动（taosc）、命令行程序 (CLI，taos) 和一些工具软件。TDengine 除了提供多种语言的连接器之外，还通过 [taosAdapter](/reference/taosadapter) 提供 [RESTful 接口](/reference/rest-api)。
+TDengine 完整的软件包包括服务端（taosd）、用于与第三方系统对接并提供 RESTful 接口的 taosAdapter、应用驱动（taosc）、命令行程序 (CLI，taos) 和一些工具软件。TDengine 除了提供多种语言的连接器之外，还通过 [taosAdapter](../reference/taosadapter) 提供 [RESTful 接口](../connector/rest-api)。
 
 本章主要介绍如何利用 Docker 或者安装包快速设置 TDengine 环境并体验其高效写入和查询。
 

docs/zh/07-develop/01-connect/index.md

@@ -33,7 +33,7 @@ TDengine 提供了丰富的应用程序开发接口，为了便于用户快速
 关键不同点在于：
 
 1. 使用 REST 连接，用户无需安装客户端驱动程序 taosc，具有跨平台易用的优势，但性能要下降 30%左右。
-2. 使用原生连接可以体验 TDengine 的全部功能，如[参数绑定接口](../../connector/cpp#参数绑定-api)、[订阅](../../connector/cpp#订阅和消费-api)等等。
+2. 使用原生连接可以体验 TDengine 的全部功能，如[参数绑定接口](../../connector/cpp/#参数绑定-api)、[订阅](../../connector/cpp/#订阅和消费-api)等等。
 
 ## 安装客户端驱动 taosc
 

docs/zh/07-develop/index.md

@@ -12,7 +12,7 @@ title: 开发指南
 7. 在很多场景下(如车辆管理)，应用需要获取每个数据采集点的最新状态，那么建议你采用TDengine的cache功能，而不用单独部署Redis等缓存软件。
 8. 如果你发现TDengine的函数无法满足你的要求，那么你可以使用用户自定义函数来解决问题。
 
-本部分内容就是按照上述的顺序组织的。为便于理解，TDengine为每个功能为每个支持的编程语言都提供了示例代码。如果你希望深入了解SQL的使用，需要查看[SQL手册](/taos-sql/)。如果想更深入地了解各连接器的使用，请阅读[连接器参考指南](../reference/connector/)。如果还希望想将TDengine与第三方系统集成起来，比如Grafana, 请参考[第三方工具](/third-party/)。
+本部分内容就是按照上述的顺序组织的。为便于理解，TDengine为每个功能为每个支持的编程语言都提供了示例代码。如果你希望深入了解SQL的使用，需要查看[SQL手册](/taos-sql/)。如果想更深入地了解各连接器的使用，请阅读[连接器参考指南](../connector/)。如果还希望想将TDengine与第三方系统集成起来，比如Grafana, 请参考[第三方工具](../third-party/)。
 
 如果在开发过程中遇到任何问题，请点击每个页面下方的["反馈问题"](https://github.com/taosdata/TDengine/issues/new/choose), 在GitHub上直接递交issue。
 

docs/zh/08-connector/02-rest-api.mdx

 ---
 title: REST API
+sidebar_label: REST API
+description: 详细介绍 TDengine 提供的 RESTful API.
 ---
 
 为支持各种不同类型平台的开发，TDengine 提供符合 REST 设计标准的 API，即 REST API。为最大程度降低学习成本，不同于其他数据库 REST API 的设计方法，TDengine 直接通过 HTTP POST 请求 BODY 中包含的 SQL 语句来操作数据库，仅需要一个 URL。REST 连接器的使用参见 [视频教程](https://www.taosdata.com/blog/2020/11/11/1965.html)。
@@ -10,7 +12,7 @@ title: REST API
 
 ## 安装
 
-RESTful 接口不依赖于任何 TDengine 的库，因此客户端不需要安装任何 TDengine 的库，只要客户端的开发语言支持 HTTP 协议即可。
+RESTful 接口不依赖于任何 TDengine 的库，因此客户端不需要安装任何 TDengine 的库，只要客户端的开发语言支持 HTTP 协议即可。TDengine 的 RESTful API 由 [taosAdapter](../../reference/taosadapter) 提供，在使用 RESTful API 之前需要确保 `taosAdapter` 正常运行。
 
 ## 验证
 

docs/zh/08-connector/cpp.mdx → docs/zh/08-connector/03-cpp.mdx

 ---
-sidebar_position: 1
 sidebar_label: C/C++
 title: C/C++ Connector
 ---
@@ -22,7 +21,7 @@ TDengine 客户端驱动的动态库位于：
 
 ## 支持的平台
 
-请参考[支持的平台列表](../connector#支持的平台)
+请参考[支持的平台列表](../#支持的平台)
 
 ## 支持的版本
 
@@ -30,7 +29,7 @@ TDengine 客户端驱动的版本号与 TDengine 服务端的版本号是一一
 
 ## 安装步骤
 
-TDengine 客户端驱动的安装请参考 [安装指南](../connector#安装步骤)
+TDengine 客户端驱动的安装请参考 [安装指南](../#安装步骤)
 
 ## 建立连接
 

docs/zh/08-connector/java.mdx → docs/zh/08-connector/04-java.mdx

 ---
 toc_max_heading_level: 4
-sidebar_position: 2
 sidebar_label: Java
 title: TDengine Java Connector
 description: TDengine Java 连接器基于标准 JDBC API 实现, 并提供原生连接与 REST连接两种连接器。
@@ -35,7 +34,7 @@ REST 连接支持所有能运行 Java 的平台。
 
 ## 版本支持
 
-请参考[版本支持列表](../connector#版本支持)
+请参考[版本支持列表](../#版本支持)
 
 ## TDengine DataType 和 Java DataType
 
@@ -64,7 +63,7 @@ TDengine 目前支持时间戳、数字、字符、布尔类型，与 Java 对
 使用 Java Connector 连接数据库前，需要具备以下条件：
 
 - 已安装 Java 1.8 或以上版本运行时环境和 Maven 3.6 或以上版本
-- 已安装 TDengine 客户端驱动（使用原生连接必须安装，使用 REST 连接无需安装），具体步骤请参考[安装客户端驱动](../connector#安装客户端驱动)
+- 已安装 TDengine 客户端驱动（使用原生连接必须安装，使用 REST 连接无需安装），具体步骤请参考[安装客户端驱动](../#安装客户端驱动)
 
 ### 安装连接器
 
@@ -630,7 +629,7 @@ public void setNString(int columnIndex, ArrayList<String> list, int size) throws
 
 ### 无模式写入
 
-TDengine 支持无模式写入功能。无模式写入兼容 InfluxDB 的 行协议（Line Protocol）、OpenTSDB 的 telnet 行协议和 OpenTSDB 的 JSON 格式协议。详情请参见[无模式写入](../../schemaless)。
+TDengine 支持无模式写入功能。无模式写入兼容 InfluxDB 的 行协议（Line Protocol）、OpenTSDB 的 telnet 行协议和 OpenTSDB 的 JSON 格式协议。详情请参见[无模式写入](../../reference/schemaless/)。
 
 **注意**：
 

docs/zh/08-connector/go.mdx → docs/zh/08-connector/05-go.mdx

@@ -30,7 +30,7 @@ REST 连接支持所有能运行 Go 的平台。
 
 ## 版本支持
 
-请参考[版本支持列表](../connector#版本支持)
+请参考[版本支持列表](../#版本支持)
 
 ## 支持的功能特性
 
@@ -56,7 +56,7 @@ REST 连接支持所有能运行 Go 的平台。
 ### 安装前准备
 
 * 安装 Go 开发环境（Go 1.14 及以上，GCC 4.8.5 及以上）
-* 如果使用原生连接器，请安装 TDengine 客户端驱动，具体步骤请参考[安装客户端驱动](../connector#安装客户端驱动)
+* 如果使用原生连接器，请安装 TDengine 客户端驱动，具体步骤请参考[安装客户端驱动](../#安装客户端驱动)
 
 配置好环境变量，检查命令：
 

docs/zh/08-connector/rust.mdx → docs/zh/08-connector/06-rust.mdx

 ---
 toc_max_heading_level: 4
-sidebar_position: 5
 sidebar_label: Rust
 title: TDengine Rust Connector
 ---
@@ -28,7 +27,7 @@ Websocket 连接支持所有能运行 Rust 的平台。
 
 ## 版本支持
 
-请参考[版本支持列表](../connector#版本支持)
+请参考[版本支持列表](../#版本支持)
 
 Rust 连接器仍然在快速开发中，1.0 之前无法保证其向后兼容。建议使用 3.0 版本以上的 TDengine，以避免已知问题。
 
@@ -37,7 +36,7 @@ Rust 连接器仍然在快速开发中，1.0 之前无法保证其向后兼容
 ### 安装前准备
 
 * 安装 Rust 开发工具链
-* 如果使用原生连接，请安装 TDengine 客户端驱动，具体步骤请参考[安装客户端驱动](../connector#安装客户端驱动)
+* 如果使用原生连接，请安装 TDengine 客户端驱动，具体步骤请参考[安装客户端驱动](../#安装客户端驱动)
 
 ### 添加 taos 依赖
 

docs/zh/08-connector/python.mdx → docs/zh/08-connector/07-python.mdx

 ---
-sidebar_position: 3
 sidebar_label: Python
 title: TDengine Python Connector
 description: "taospy 是 TDengine 的官方 Python 连接器。taospy 提供了丰富的 API， 使得 Python 应用可以很方便地使用 TDengine。tasopy 对 TDengine 的原生接口和 REST 接口都进行了封装， 分别对应 tasopy 的两个子模块：tasos 和 taosrest。除了对原生接口和 REST 接口的封装，taospy 还提供了符合 Python 数据访问规范(PEP 249)的编程接口。这使得 taospy 和很多第三方工具集成变得简单，比如 SQLAlchemy 和 pandas"
@@ -8,7 +7,7 @@ description: "taospy 是 TDengine 的官方 Python 连接器。taospy 提供了
 import Tabs from "@theme/Tabs";
 import TabItem from "@theme/TabItem";
 
-`taospy` 是 TDengine 的官方 Python 连接器。`taospy` 提供了丰富的 API， 使得 Python 应用可以很方便地使用 TDengine。`taospy` 对 TDengine 的[原生接口](../connector/cpp)和 [REST 接口](/reference/rest-api)都进行了封装， 分别对应 `taospy` 包的 `taos` 模块 和 `taosrest` 模块。
+`taospy` 是 TDengine 的官方 Python 连接器。`taospy` 提供了丰富的 API， 使得 Python 应用可以很方便地使用 TDengine。`taospy` 对 TDengine 的[原生接口](../cpp)和 [REST 接口](../rest-api)都进行了封装， 分别对应 `taospy` 包的 `taos` 模块 和 `taosrest` 模块。
 除了对原生接口和 REST 接口的封装，`taospy` 还提供了符合 [Python 数据访问规范(PEP 249)](https://peps.python.org/pep-0249/) 的编程接口。这使得 `taospy` 和很多第三方工具集成变得简单，比如 [SQLAlchemy](https://www.sqlalchemy.org/) 和 [pandas](https://pandas.pydata.org/)。
 
 使用客户端驱动提供的原生接口直接与服务端建立的连接的方式下文中称为“原生连接”；使用 taosAdapter 提供的 REST 接口与服务端建立的连接的方式下文中称为“REST 连接”。
@@ -17,7 +16,7 @@ Python 连接器的源码托管在 [GitHub](https://github.com/taosdata/taos-con
 
 ## 支持的平台
 
-- 原生连接[支持的平台](../connector/#支持的平台)和 TDengine 客户端支持的平台一致。
+- 原生连接[支持的平台](../#支持的平台)和 TDengine 客户端支持的平台一致。
 - REST 连接支持所有能运行 Python 的平台。
 
 ## 版本选择
@@ -275,7 +274,7 @@ TaosCursor 类使用原生连接进行写入、查询操作。在客户端多线
 
 ##### RestClient 类的使用
 
-`RestClient` 类是对于 [REST API](/reference/rest-api) 的直接封装。它只包含一个 `sql()` 方法用于执行任意 SQL 语句， 并返回执行结果。
+`RestClient` 类是对于 [REST API](../rest-api) 的直接封装。它只包含一个 `sql()` 方法用于执行任意 SQL 语句， 并返回执行结果。
 
 ```python title="RestClient 的使用"
 {{#include docs/examples/python/rest_client_example.py}}

docs/zh/08-connector/node.mdx → docs/zh/08-connector/08-node.mdx

 ---
 toc_max_heading_level: 4
-sidebar_position: 6
 sidebar_label: Node.js
 title: TDengine Node.js Connector
 ---
@@ -28,7 +27,7 @@ REST 连接器支持所有能运行 Node.js 的平台。
 
 ## 版本支持
 
-请参考[版本支持列表](../connector#版本支持)
+请参考[版本支持列表](../#版本支持)
 
 ## 支持的功能特性
 
@@ -52,7 +51,7 @@ REST 连接器支持所有能运行 Node.js 的平台。
 ### 安装前准备
 
 - 安装 Node.js 开发环境
-- 如果使用 REST 连接器，跳过此步。但如果使用原生连接器，请安装 TDengine 客户端驱动，具体步骤请参考[安装客户端驱动](../connector#安装客户端驱动)。我们使用 [node-gyp](https://github.com/nodejs/node-gyp) 和 TDengine 实例进行交互，还需要根据具体操作系统来安装下文提到的一些依赖工具。
+- 如果使用 REST 连接器，跳过此步。但如果使用原生连接器，请安装 TDengine 客户端驱动，具体步骤请参考[安装客户端驱动](../#安装客户端驱动)。我们使用 [node-gyp](https://github.com/nodejs/node-gyp) 和 TDengine 实例进行交互，还需要根据具体操作系统来安装下文提到的一些依赖工具。
 
 <Tabs defaultValue="Linux">
 <TabItem value="Linux" label="Linux 系统安装依赖工具">

docs/zh/08-connector/csharp.mdx → docs/zh/08-connector/09-csharp.mdx

 ---
 toc_max_heading_level: 4
-sidebar_position: 7
 sidebar_label: C#
 title: C# Connector
 ---
@@ -18,7 +17,7 @@ import CSAsyncQuery from "../07-develop/04-query-data/_cs_async.mdx"
 
 `TDengine.Connector` 是 TDengine 提供的 C# 语言连接器。C# 开发人员可以通过它开发存取 TDengine 集群数据的 C# 应用软件。
 
-`TDengine.Connector` 连接器支持通过 TDengine 客户端驱动（taosc）建立与 TDengine 运行实例的连接，提供数据写入、查询、订阅、schemaless 数据写入、参数绑定接口数据写入等功能 `TDengine.Connector` 目前暂未提供 REST 连接方式，用户可以参考 [REST API](/reference/rest-api/) 文档自行编写。
+`TDengine.Connector` 连接器支持通过 TDengine 客户端驱动（taosc）建立与 TDengine 运行实例的连接，提供数据写入、查询、订阅、schemaless 数据写入、参数绑定接口数据写入等功能 `TDengine.Connector` 目前暂未提供 REST 连接方式，用户可以参考 [REST API](../rest-api/) 文档自行编写。
 
 本文介绍如何在 Linux 或 Windows 环境中安装 `TDengine.Connector`，并通过 `TDengine.Connector` 连接 TDengine 集群，进行数据写入、查询等基本操作。
 
@@ -32,7 +31,7 @@ import CSAsyncQuery from "../07-develop/04-query-data/_cs_async.mdx"
 
 ## 版本支持
 
-请参考[版本支持列表](../connector#版本支持)
+请参考[版本支持列表](../#版本支持)
 
 ## 支持的功能特性
 
@@ -49,7 +48,7 @@ import CSAsyncQuery from "../07-develop/04-query-data/_cs_async.mdx"
 
 * 安装 [.NET SDK](https://dotnet.microsoft.com/download)
 * [Nuget 客户端](https://docs.microsoft.com/en-us/nuget/install-nuget-client-tools) （可选安装）
-* 安装 TDengine 客户端驱动，具体步骤请参考[安装客户端驱动](../connector#安装客户端驱动)
+* 安装 TDengine 客户端驱动，具体步骤请参考[安装客户端驱动](../#安装客户端驱动)
 
 ### 使用 dotnet CLI 安装
 

docs/zh/08-connector/php.mdx → docs/zh/08-connector/10-php.mdx

 ---
-sidebar_position: 1
-sidebar_label: PHP（社区贡献）
+sidebar_label: PHP
 title: PHP Connector
 ---
 
@@ -38,7 +37,7 @@ TDengine 客户端驱动的版本号与 TDengine 服务端的版本号是一一
 
 ### 安装 TDengine 客户端驱动
 
-TDengine 客户端驱动的安装请参考 [安装指南](../connector#安装步骤)
+TDengine 客户端驱动的安装请参考 [安装指南](../#安装步骤)
 
 ### 编译安装 php-tdengine
 

docs/zh/08-connector/03-connector.mdx → docs/zh/08-connector/index.md

 ---
+sidebar_label: 连接器
 title: 连接器
+description: 详细介绍各种语言的连接器及 REST API
 ---
 
 TDengine 提供了丰富的应用程序开发接口，为了便于用户快速开发自己的应用，TDengine 支持了多种编程语言的连接器，其中官方连接器包括支持 C/C++、Java、Python、Go、Node.js、C# 和 Rust 的连接器。这些连接器支持使用原生接口（taosc）和 REST 接口（部分语言暂不支持）连接 TDengine 集群。社区开发者也贡献了多个非官方连接器，例如 ADO.NET 连接器、Lua 连接器和 PHP 连接器。

docs/zh/12-taos-sql/14-stream.md

@@ -18,7 +18,7 @@ stream_options: {
 其中 subquery 是 select 普通查询语法的子集:
 
 ```sql
-subquery: SELECT [DISTINCT] select_list
+subquery: SELECT select_list
     from_clause
     [WHERE condition]
     [PARTITION BY tag_list]
@@ -37,7 +37,7 @@ window_clause: {
 
 其中，SESSION 是会话窗口，tol_val 是时间间隔的最大范围。在 tol_val 时间间隔范围内的数据都属于同一个窗口，如果连续的两条数据的时间超过 tol_val，则自动开启下一个窗口。
 
-窗口的定义与时序数据特色查询中的定义完全相同。
+窗口的定义与时序数据特色查询中的定义完全相同，详见 [TDengine 特色查询](../distinguished)
 
 例如，如下语句创建流式计算，同时自动创建名为 avg_vol 的超级表，此流计算以一分钟为时间窗口、30 秒为前向增量统计这些电表的平均电压，并将来自 meters 表的数据的计算结果写入 avg_vol 表，不同 partition 的数据会分别创建子表并写入不同子表。
 

docs/zh/12-taos-sql/24-show.md

@@ -3,7 +3,7 @@ sidebar_label: SHOW 命令
 title: 使用 SHOW 命令查看系统元数据
 ---
 
-除了使用 `select` 语句查询 `INFORMATION_SCHEMA` 数据库中的表获得系统中的各种元数据、系统信息和状态之外，也可以用 `SHOW` 命令来实现同样的目的。
+SHOW 命令可以用来获取简要的系统信息。若想获取系统中详细的各种元数据、系统信息和状态，请使用 select 语句查询 INFORMATION_SCHEMA 数据库中的表。
 
 ## SHOW ACCOUNTS
 

docs/zh/14-reference/04-taosadapter.md

@@ -156,7 +156,7 @@ AllowWebSockets
 ## 功能列表
 
 - RESTful 接口
-  [https://docs.taosdata.com/reference/rest-api/](https://docs.taosdata.com/reference/rest-api/)
+  [RESTful API](../../connector/rest-api)
 - 兼容 InfluxDB v1 写接口
   [https://docs.influxdata.com/influxdb/v2.0/reference/api/influxdb-1x/write/](https://docs.influxdata.com/influxdb/v2.0/reference/api/influxdb-1x/write/)
 - 兼容 OpenTSDB JSON 和 telnet 格式写入
@@ -179,7 +179,7 @@ AllowWebSockets
 
 ### TDengine RESTful 接口
 
-您可以使用任何支持 http 协议的客户端通过访问 RESTful 接口地址 `http://<fqdn>:6041/rest/sql` 来写入数据到 TDengine 或从 TDengine 中查询数据。细节请参考[官方文档](/reference/rest-api/)。
+您可以使用任何支持 http 协议的客户端通过访问 RESTful 接口地址 `http://<fqdn>:6041/rest/sql` 来写入数据到 TDengine 或从 TDengine 中查询数据。细节请参考[官方文档](../../connector/rest-api/)。
 
 ### InfluxDB
 

docs/zh/14-reference/12-config/index.md

@@ -698,122 +698,123 @@ charset 的有效值是 UTF-8。
 | 45  | numOfVnodeFetchThreads | 否 | 是 |
 | 46  | numOfVnodeWriteThreads | 否 | 是 |
 | 47  | numOfVnodeSyncThreads | 否 | 是 |
-| 48  | numOfQnodeQueryThreads | 否 | 是 |
-| 49  | numOfQnodeFetchThreads | 否 | 是 |
-| 50  | numOfSnodeSharedThreads | 否 | 是 |
-| 51  | numOfSnodeUniqueThreads | 否 | 是 |
-| 52  | rpcQueueMemoryAllowed | 否 | 是 |
-| 53  | logDir | 是 | 是 |
-| 54  | minimalLogDirGB | 是 | 是 |
-| 55  | numOfLogLines | 是 | 是 |
-| 56  | asyncLog | 是 | 是 |
-| 57  | logKeepDays | 是 | 是 |
-| 58  | debugFlag | 是 | 是 |
-| 59  | tmrDebugFlag | 是 | 是 |
-| 60  | uDebugFlag | 是 | 是 |
-| 61  | rpcDebugFlag | 是 | 是 |
-| 62  | jniDebugFlag | 是 | 是 |
-| 63  | qDebugFlag | 是 | 是 |
-| 64  | cDebugFlag | 是 | 是 |
-| 65  | dDebugFlag | 是 | 是 |
-| 66  | vDebugFlag | 是 | 是 |
-| 67  | mDebugFlag | 是 | 是 |
-| 68  | wDebugFlag | 是 | 是 |
-| 69  | sDebugFlag | 是 | 是 |
-| 70  | tsdbDebugFlag | 是 | 是 |
-| 71  | tqDebugFlag | 否 | 是 |
-| 72  | fsDebugFlag | 是 | 是 |
-| 73  | udfDebugFlag | 否 | 是 |
-| 74  | smaDebugFlag | 否 | 是 |
-| 75  | idxDebugFlag | 否 | 是 |
-| 76  | tdbDebugFlag | 否 | 是 |
-| 77  | metaDebugFlag | 否 | 是 |
-| 78  | timezone | 是 | 是 |
-| 79  | locale | 是 | 是 |
-| 80  | charset | 是 | 是 |
-| 81  | udf | 是 | 是 |
-| 82  | enableCoreFile | 是 | 是 |
-| 83  | arbitrator | 是 | 否 |
-| 84  | numOfThreadsPerCore | 是 | 否 |
-| 85  | numOfMnodes | 是 | 否 |
-| 86  | vnodeBak | 是 | 否 |
-| 87  | balance | 是 | 否 |
-| 88  | balanceInterval | 是 | 否 |
-| 89  | offlineThreshold | 是 | 否 |
-| 90  | role | 是 | 否 |
-| 91  | dnodeNopLoop | 是 | 否 |
-| 92  | keepTimeOffset | 是 | 否 |
-| 93  | rpcTimer | 是 | 否 |
-| 94  | rpcMaxTime | 是 | 否 |
-| 95  | rpcForceTcp | 是 | 否 |
-| 96  | tcpConnTimeout | 是 | 否 |
-| 97  | syncCheckInterval | 是 | 否 |
-| 98  | maxTmrCtrl | 是 | 否 |
-| 99  | monitorReplica | 是 | 否 |
-| 100 | smlTagNullName | 是 | 否 |
-| 101 | keepColumnName | 是 | 否 |
-| 102 | ratioOfQueryCores | 是 | 否 |
-| 103 | maxStreamCompDelay | 是 | 否 |
-| 104 | maxFirstStreamCompDelay | 是 | 否 |
-| 105 | retryStreamCompDelay | 是 | 否 |
-| 106 | streamCompDelayRatio | 是 | 否 |
-| 107 | maxVgroupsPerDb | 是 | 否 |
-| 108 | maxTablesPerVnode | 是 | 否 |
-| 109 | minTablesPerVnode | 是 | 否 |
-| 110 | tableIncStepPerVnode | 是 | 否 |
-| 111 | cache | 是 | 否 |
-| 112 | blocks | 是 | 否 |
-| 113 | days | 是 | 否 |
-| 114 | keep | 是 | 否 |
-| 115 | minRows | 是 | 否 |
-| 116 | maxRows | 是 | 否 |
-| 117 | quorum | 是 | 否 |
-| 118 | comp | 是 | 否 |
-| 119 | walLevel | 是 | 否 |
-| 120 | fsync | 是 | 否 |
-| 121 | replica | 是 | 否 |
-| 122 | partitions | 是 | 否 |
-| 123 | quorum | 是 | 否 |
-| 124 | update | 是 | 否 |
-| 125 | cachelast | 是 | 否 |
-| 126 | maxSQLLength | 是 | 否 |
-| 127 | maxWildCardsLength | 是 | 否 |
-| 128 | maxRegexStringLen | 是 | 否 |
-| 129 | maxNumOfOrderedRes | 是 | 否 |
-| 130 | maxConnections | 是 | 否 |
-| 131 | mnodeEqualVnodeNum | 是 | 否 |
-| 132 | http | 是 | 否 |
-| 133 | httpEnableRecordSql | 是 | 否 |
-| 134 | httpMaxThreads | 是 | 否 |
-| 135 | restfulRowLimit | 是 | 否 |
-| 136 | httpDbNameMandatory | 是 | 否 |
-| 137 | httpKeepAlive | 是 | 否 |
-| 138 | enableRecordSql | 是 | 否 |
-| 139 | maxBinaryDisplayWidth | 是 | 否 |
-| 140 | stream | 是 | 否 |
-| 141 | retrieveBlockingModel | 是 | 否 |
-| 142 | tsdbMetaCompactRatio | 是 | 否 |
-| 143 | defaultJSONStrType | 是 | 否 |
-| 144 | walFlushSize | 是 | 否 |
-| 145 | keepTimeOffset | 是 | 否 |
-| 146 | flowctrl | 是 | 否 |
-| 147 | slaveQuery | 是 | 否 |
-| 148 | adjustMaster | 是 | 否 |
-| 149 | topicBinaryLen | 是 | 否 |
-| 150 | telegrafUseFieldNum | 是 | 否 |
-| 151 | deadLockKillQuery | 是 | 否 |
-| 152 | clientMerge | 是 | 否 |
-| 153 | sdbDebugFlag | 是 | 否 |
-| 154 | odbcDebugFlag | 是 | 否 |
-| 155 | httpDebugFlag | 是 | 否 |
-| 156 | monDebugFlag | 是 | 否 |
-| 157 | cqDebugFlag | 是 | 否 |
-| 158 | shortcutFlag | 是 | 否 |
-| 159 | probeSeconds | 是 | 否 |
-| 160 | probeKillSeconds | 是 | 否 |
-| 161 | probeInterval | 是 | 否 |
-| 162 | lossyColumns | 是 | 否 |
-| 163 | fPrecision | 是 | 否 |
-| 164 | dPrecision | 是 | 否 |
-| 165 | maxRange | 是 | 否 |
-| 166 | range | 是 | 否 |
+| 48  | numOfVnodeRsmaThreads | 否 | 是 | 
+| 49  | numOfQnodeQueryThreads | 否 | 是 |
+| 50  | numOfQnodeFetchThreads | 否 | 是 |
+| 51  | numOfSnodeSharedThreads | 否 | 是 |
+| 52  | numOfSnodeUniqueThreads | 否 | 是 |
+| 53  | rpcQueueMemoryAllowed | 否 | 是 |
+| 54  | logDir | 是 | 是 |
+| 55  | minimalLogDirGB | 是 | 是 |
+| 56  | numOfLogLines | 是 | 是 |
+| 57  | asyncLog | 是 | 是 |
+| 58  | logKeepDays | 是 | 是 |
+| 59  | debugFlag | 是 | 是 |
+| 60  | tmrDebugFlag | 是 | 是 |
+| 61  | uDebugFlag | 是 | 是 |
+| 62  | rpcDebugFlag | 是 | 是 |
+| 63  | jniDebugFlag | 是 | 是 |
+| 64  | qDebugFlag | 是 | 是 |
+| 65  | cDebugFlag | 是 | 是 |
+| 66  | dDebugFlag | 是 | 是 |
+| 67  | vDebugFlag | 是 | 是 |
+| 68  | mDebugFlag | 是 | 是 |
+| 69  | wDebugFlag | 是 | 是 |
+| 70  | sDebugFlag | 是 | 是 |
+| 71  | tsdbDebugFlag | 是 | 是 |
+| 72  | tqDebugFlag | 否 | 是 |
+| 73  | fsDebugFlag | 是 | 是 |
+| 74  | udfDebugFlag | 否 | 是 |
+| 75  | smaDebugFlag | 否 | 是 |
+| 76  | idxDebugFlag | 否 | 是 |
+| 77  | tdbDebugFlag | 否 | 是 |
+| 78  | metaDebugFlag | 否 | 是 |
+| 79  | timezone | 是 | 是 |
+| 80  | locale | 是 | 是 |
+| 81  | charset | 是 | 是 |
+| 82  | udf | 是 | 是 |
+| 83  | enableCoreFile | 是 | 是 |
+| 84  | arbitrator | 是 | 否 |
+| 85  | numOfThreadsPerCore | 是 | 否 |
+| 86  | numOfMnodes | 是 | 否 |
+| 87  | vnodeBak | 是 | 否 |
+| 88  | balance | 是 | 否 |
+| 89  | balanceInterval | 是 | 否 |
+| 90  | offlineThreshold | 是 | 否 |
+| 91  | role | 是 | 否 |
+| 92  | dnodeNopLoop | 是 | 否 |
+| 93  | keepTimeOffset | 是 | 否 |
+| 94  | rpcTimer | 是 | 否 |
+| 95  | rpcMaxTime | 是 | 否 |
+| 96  | rpcForceTcp | 是 | 否 |
+| 97  | tcpConnTimeout | 是 | 否 |
+| 98  | syncCheckInterval | 是 | 否 |
+| 99  | maxTmrCtrl | 是 | 否 |
+| 100 | monitorReplica | 是 | 否 |
+| 101 | smlTagNullName | 是 | 否 |
+| 102 | keepColumnName | 是 | 否 |
+| 103 | ratioOfQueryCores | 是 | 否 |
+| 104 | maxStreamCompDelay | 是 | 否 |
+| 105 | maxFirstStreamCompDelay | 是 | 否 |
+| 106 | retryStreamCompDelay | 是 | 否 |
+| 107 | streamCompDelayRatio | 是 | 否 |
+| 108 | maxVgroupsPerDb | 是 | 否 |
+| 109 | maxTablesPerVnode | 是 | 否 |
+| 110 | minTablesPerVnode | 是 | 否 |
+| 111 | tableIncStepPerVnode | 是 | 否 |
+| 112 | cache | 是 | 否 |
+| 113 | blocks | 是 | 否 |
+| 114 | days | 是 | 否 |
+| 115 | keep | 是 | 否 |
+| 116 | minRows | 是 | 否 |
+| 117 | maxRows | 是 | 否 |
+| 118 | quorum | 是 | 否 |
+| 119 | comp | 是 | 否 |
+| 120 | walLevel | 是 | 否 |
+| 121 | fsync | 是 | 否 |
+| 122 | replica | 是 | 否 |
+| 123 | partitions | 是 | 否 |
+| 124 | quorum | 是 | 否 |
+| 125 | update | 是 | 否 |
+| 126 | cachelast | 是 | 否 |
+| 127 | maxSQLLength | 是 | 否 |
+| 128 | maxWildCardsLength | 是 | 否 |
+| 129 | maxRegexStringLen | 是 | 否 |
+| 130 | maxNumOfOrderedRes | 是 | 否 |
+| 131 | maxConnections | 是 | 否 |
+| 132 | mnodeEqualVnodeNum | 是 | 否 |
+| 133 | http | 是 | 否 |
+| 134 | httpEnableRecordSql | 是 | 否 |
+| 135 | httpMaxThreads | 是 | 否 |
+| 136 | restfulRowLimit | 是 | 否 |
+| 137 | httpDbNameMandatory | 是 | 否 |
+| 138 | httpKeepAlive | 是 | 否 |
+| 139 | enableRecordSql | 是 | 否 |
+| 140 | maxBinaryDisplayWidth | 是 | 否 |
+| 141 | stream | 是 | 否 |
+| 142 | retrieveBlockingModel | 是 | 否 |
+| 143 | tsdbMetaCompactRatio | 是 | 否 |
+| 144 | defaultJSONStrType | 是 | 否 |
+| 145 | walFlushSize | 是 | 否 |
+| 146 | keepTimeOffset | 是 | 否 |
+| 147 | flowctrl | 是 | 否 |
+| 148 | slaveQuery | 是 | 否 |
+| 149 | adjustMaster | 是 | 否 |
+| 150 | topicBinaryLen | 是 | 否 |
+| 151 | telegrafUseFieldNum | 是 | 否 |
+| 152 | deadLockKillQuery | 是 | 否 |
+| 153 | clientMerge | 是 | 否 |
+| 154 | sdbDebugFlag | 是 | 否 |
+| 155 | odbcDebugFlag | 是 | 否 |
+| 156 | httpDebugFlag | 是 | 否 |
+| 157 | monDebugFlag | 是 | 否 |
+| 158 | cqDebugFlag | 是 | 否 |
+| 159 | shortcutFlag | 是 | 否 |
+| 160 | probeSeconds | 是 | 否 |
+| 161 | probeKillSeconds | 是 | 否 |
+| 162 | probeInterval | 是 | 否 |
+| 163 | lossyColumns | 是 | 否 |
+| 164 | fPrecision | 是 | 否 |
+| 165 | dPrecision | 是 | 否 |
+| 166 | maxRange | 是 | 否 |
+| 167 | range | 是 | 否 |

docs/zh/20-third-party/09-emq-broker.md

@@ -90,7 +90,7 @@ http://127.0.0.1:6041/rest/sql
 ```
 Basic cm9vdDp0YW9zZGF0YQ==
 ```
-相关文档请参考[ TDengine REST API 文档](/reference/rest-api/)。
+相关文档请参考[ TDengine REST API 文档](../../connector/rest-api/)。
 
 在消息体中输入规则引擎替换模板:
 

docs/zh/28-releases/01-tdengine.md

@@ -9,7 +9,7 @@ import Release from "/components/ReleaseV3";
 
 <Release type="tdengine" version="3.0.0.1" />
 
-## 3.0.0.0
+<!-- ## 3.0.0.0
 
-<Release type="tdengine" version="3.0.0.0" />
+<Release type="tdengine" version="3.0.0.0" /> -->
 

examples/c/stream_demo.c

@@ -13,6 +13,7 @@
  * along with this program. If not, see <http://www.gnu.org/licenses/>.
  */
 
+// clang-format off
 #include <assert.h>
 #include <stdio.h>
 #include <string.h>
@@ -94,13 +95,8 @@ int32_t create_stream() {
   }
   taos_free_result(pRes);
 
-  /*const char* sql = "select min(k), max(k), sum(k) from tu1";*/
-  /*const char* sql = "select min(k), max(k), sum(k) as sum_of_k from st1";*/
-  /*const char* sql = "select sum(k) from tu1 interval(10m)";*/
-  /*pRes = tmq_create_stream(pConn, "stream1", "out1", sql);*/
   pRes = taos_query(pConn,
-                    "create stream stream1 trigger max_delay 10s watermark 10s into outstb as select _wstart start, "
-                    "count(k) from st1 partition by tbname interval(20s) ");
+                    "create stream stream1 trigger at_once watermark 10s into outstb as select _wstart start, avg(k) from st1 partition by tbname interval(10s)");
   if (taos_errno(pRes) != 0) {
     printf("failed to create stream stream1, reason:%s\n", taos_errstr(pRes));
     return -1;

include/common/tcommon.h

@@ -44,6 +44,30 @@ enum {
 )
 // clang-format on
 
+typedef struct {
+  TSKEY    ts;
+  uint64_t groupId;
+} SWinKey;
+
+static inline int SWinKeyCmpr(const void* pKey1, int kLen1, const void* pKey2, int kLen2) {
+  SWinKey* pWin1 = (SWinKey*)pKey1;
+  SWinKey* pWin2 = (SWinKey*)pKey2;
+
+  if (pWin1->groupId > pWin2->groupId) {
+    return 1;
+  } else if (pWin1->groupId < pWin2->groupId) {
+    return -1;
+  }
+
+  if (pWin1->ts > pWin2->ts) {
+    return 1;
+  } else if (pWin1->ts < pWin2->ts) {
+    return -1;
+  }
+
+  return 0;
+}
+
 enum {
   TMQ_MSG_TYPE__DUMMY = 0,
   TMQ_MSG_TYPE__POLL_RSP,
@@ -181,7 +205,7 @@ typedef struct SColumn {
   int16_t slotId;
 
   char    name[TSDB_COL_NAME_LEN];
-  int8_t  flag;  // column type: normal column, tag, or user-input column (integer/float/string)
+  int16_t colType;  // column type: normal column, tag, or window column
   int16_t type;
   int32_t bytes;
   uint8_t precision;

include/common/tglobal.h

@@ -66,6 +66,7 @@ extern int32_t tsNumOfVnodeStreamThreads;
 extern int32_t tsNumOfVnodeFetchThreads;
 extern int32_t tsNumOfVnodeWriteThreads;
 extern int32_t tsNumOfVnodeSyncThreads;
+extern int32_t tsNumOfVnodeRsmaThreads;
 extern int32_t tsNumOfQnodeQueryThreads;
 extern int32_t tsNumOfQnodeFetchThreads;
 extern int32_t tsNumOfSnodeSharedThreads;

include/libs/executor/executor.h

@@ -29,7 +29,7 @@ typedef void* DataSinkHandle;
 struct SRpcMsg;
 struct SSubplan;
 
-typedef struct SReadHandle {
+typedef struct {
   void*   tqReader;
   void*   meta;
   void*   config;
@@ -41,6 +41,7 @@ typedef struct SReadHandle {
   bool    initTableReader;
   bool    initTqReader;
   int32_t numOfVgroups;
+  void*   pStateBackend;
 } SReadHandle;
 
 // in queue mode, data streams are seperated by msg
@@ -78,8 +79,8 @@ int32_t qSetMultiStreamInput(qTaskInfo_t tinfo, const void* pBlocks, size_t numO
 
 /**
  * @brief Cleanup SSDataBlock for StreamScanInfo
- * 
- * @param tinfo 
+ *
+ * @param tinfo
  */
 void tdCleanupStreamInputDataBlock(qTaskInfo_t tinfo);
 
@@ -163,7 +164,7 @@ int32_t qGetQualifiedTableIdList(void* pTableList, const char* tagCond, int32_t
 
 void qProcessRspMsg(void* parent, struct SRpcMsg* pMsg, struct SEpSet* pEpSet);
 
-int32_t qGetExplainExecInfo(qTaskInfo_t tinfo, SArray* pExecInfoList/*,int32_t* resNum, SExplainExecInfo** pRes*/);
+int32_t qGetExplainExecInfo(qTaskInfo_t tinfo, SArray* pExecInfoList /*,int32_t* resNum, SExplainExecInfo** pRes*/);
 
 int32_t qSerializeTaskStatus(qTaskInfo_t tinfo, char** pOutput, int32_t* len);
 

include/libs/nodes/querynodes.h

@@ -57,7 +57,9 @@ typedef enum EColumnType {
   COLUMN_TYPE_COLUMN = 1,
   COLUMN_TYPE_TAG,
   COLUMN_TYPE_TBNAME,
-  COLUMN_TYPE_WINDOW_PC,
+  COLUMN_TYPE_WINDOW_START,
+  COLUMN_TYPE_WINDOW_END,
+  COLUMN_TYPE_WINDOW_DURATION,
   COLUMN_TYPE_GROUP_KEY
 } EColumnType;
 

include/libs/stream/tstream.h

@@ -263,6 +263,14 @@ typedef struct {
   SArray* checkpointVer;
 } SStreamRecoveringState;
 
+// incremental state storage
+typedef struct {
+  SStreamTask* pOwner;
+  TDB*         db;
+  TTB*         pStateDb;
+  TXN          txn;
+} SStreamState;
+
 typedef struct SStreamTask {
   int64_t streamId;
   int32_t taskId;
@@ -312,6 +320,10 @@ typedef struct SStreamTask {
 
   // msg handle
   SMsgCb* pMsgCb;
+
+  // state backend
+  SStreamState* pState;
+
 } SStreamTask;
 
 int32_t tEncodeStreamEpInfo(SEncoder* pEncoder, const SStreamChildEpInfo* pInfo);
@@ -507,7 +519,7 @@ typedef struct SStreamMeta {
   char*        path;
   TDB*         db;
   TTB*         pTaskDb;
-  TTB*         pStateDb;
+  TTB*         pCheckpointDb;
   SHashObj*    pTasks;
   SHashObj*    pRecoverStatus;
   void*        ahandle;
@@ -528,6 +540,37 @@ int32_t streamMetaCommit(SStreamMeta* pMeta);
 int32_t streamMetaRollBack(SStreamMeta* pMeta);
 int32_t streamLoadTasks(SStreamMeta* pMeta);
 
+SStreamState* streamStateOpen(char* path, SStreamTask* pTask);
+void          streamStateClose(SStreamState* pState);
+int32_t       streamStateBegin(SStreamState* pState);
+int32_t       streamStateCommit(SStreamState* pState);
+int32_t       streamStateAbort(SStreamState* pState);
+
+typedef struct {
+  TBC* pCur;
+} SStreamStateCur;
+
+#if 1
+int32_t streamStatePut(SStreamState* pState, const SWinKey* key, const void* value, int32_t vLen);
+int32_t streamStateGet(SStreamState* pState, const SWinKey* key, void** pVal, int32_t* pVLen);
+int32_t streamStateDel(SStreamState* pState, const SWinKey* key);
+void    streamFreeVal(void* val);
+
+SStreamStateCur* streamStateGetCur(SStreamState* pState, const SWinKey* key);
+SStreamStateCur* streamStateSeekKeyNext(SStreamState* pState, const SWinKey* key);
+SStreamStateCur* streamStateSeekKeyPrev(SStreamState* pState, const SWinKey* key);
+void             streamStateFreeCur(SStreamStateCur* pCur);
+
+int32_t streamStateGetKVByCur(SStreamStateCur* pCur, SWinKey* pKey, const void** pVal, int32_t* pVLen);
+
+int32_t streamStateSeekFirst(SStreamState* pState, SStreamStateCur* pCur);
+int32_t streamStateSeekLast(SStreamState* pState, SStreamStateCur* pCur);
+
+int32_t streamStateCurNext(SStreamState* pState, SStreamStateCur* pCur);
+int32_t streamStateCurPrev(SStreamState* pState, SStreamStateCur* pCur);
+
+#endif
+
 #ifdef __cplusplus
 }
 #endif

packaging/release.bat

@@ -40,10 +40,12 @@ if not exist %work_dir%\debug\ver-%2-x86 (
 )
 cd %work_dir%\debug\ver-%2-x64
 call vcvarsall.bat x64
-cmake ../../ -G "NMake Makefiles JOM" -DCMAKE_MAKE_PROGRAM=jom -DBUILD_TOOLS=true -DBUILD_HTTP=false -DVERNUMBER=%2 -DCPUTYPE=x64
+cmake ../../ -G "NMake Makefiles JOM" -DCMAKE_MAKE_PROGRAM=jom -DBUILD_TOOLS=true -DBUILD_HTTP=false -DBUILD_TEST=false -DVERNUMBER=%2 -DCPUTYPE=x64
 cmake --build .
 rd /s /Q C:\TDengine
 cmake --install .
+for /r c:\TDengine %%i in (*.dll) do signtool sign /f D:\\123.pfx /p taosdata %%i
+for /r c:\TDengine %%i in (*.exe) do signtool sign /f D:\\123.pfx /p taosdata %%i
 if not %errorlevel% == 0  ( call :RUNFAILED build x64 failed & exit /b 1)
 cd %package_dir%
 iscc /DMyAppInstallName="%packagServerName_x64%" /DMyAppVersion="%2" /DMyAppExcludeSource="" tools\tdengine.iss /O..\release
@@ -51,19 +53,7 @@ if not %errorlevel% == 0  ( call :RUNFAILED package %packagServerName_x64% faile
 iscc /DMyAppInstallName="%packagClientName_x64%" /DMyAppVersion="%2" /DMyAppExcludeSource="taosd.exe" tools\tdengine.iss /O..\release
 if not %errorlevel% == 0  ( call :RUNFAILED package %packagClientName_x64% failed & exit /b 1)
 
-cd %work_dir%\debug\ver-%2-x86
-call vcvarsall.bat x86
-cmake ../../ -G "NMake Makefiles JOM" -DCMAKE_MAKE_PROGRAM=jom -DBUILD_TOOLS=true -DBUILD_HTTP=false -DVERNUMBER=%2 -DCPUTYPE=x86
-cmake --build .
-rd /s /Q C:\TDengine
-cmake --install .
-if not %errorlevel% == 0  ( call :RUNFAILED build x86 failed & exit /b 1)
-cd %package_dir%
-@REM iscc /DMyAppInstallName="%packagServerName_x86%" /DMyAppVersion="%2" /DMyAppExcludeSource="" tools\tdengine.iss /O..\release
-@REM if not %errorlevel% == 0  ( call :RUNFAILED package %packagServerName_x86% failed & exit /b 1)
-iscc /DMyAppInstallName="%packagClientName_x86%" /DMyAppVersion="%2" /DMyAppExcludeSource="taosd.exe" tools\tdengine.iss /O..\release
-if not %errorlevel% == 0  ( call :RUNFAILED package %packagClientName_x86% failed & exit /b 1)
-
+for /r ..\release %%i in (*.exe) do signtool sign /f d:\\123.pfx /p taosdata %%i
 goto EXIT0
 
 :USAGE

source/client/inc/clientInt.h

@@ -96,7 +96,12 @@ typedef struct {
 
 typedef struct SQueryExecMetric {
   int64_t start;   // start timestamp, us
-  int64_t parsed;  // start to parse, us
+  int64_t syntaxStart;  // start to parse, us
+  int64_t syntaxEnd;  // end to parse, us
+  int64_t ctgStart;  // start to parse, us
+  int64_t ctgEnd;  // end to parse, us
+  int64_t semanticEnd;
+  int64_t execEnd;
   int64_t send;    // start to send to server, us
   int64_t rsp;     // receive response from server, us
 } SQueryExecMetric;

source/client/inc/clientLog.h

@@ -29,6 +29,7 @@ extern "C" {
 #define tscDebug(...)  do { if (cDebugFlag & DEBUG_DEBUG) { taosPrintLog("TSC ", DEBUG_DEBUG, cDebugFlag, __VA_ARGS__); }} while(0)
 #define tscTrace(...)  do { if (cDebugFlag & DEBUG_TRACE) { taosPrintLog("TSC ", DEBUG_TRACE, cDebugFlag, __VA_ARGS__); }} while(0)
 #define tscDebugL(...) do { if (cDebugFlag & DEBUG_DEBUG) { taosPrintLongString("TSC ", DEBUG_DEBUG, cDebugFlag, __VA_ARGS__); }} while(0)
+#define tscPerf(...)   do { taosPrintLog("TSC ", 0, cDebugFlag, __VA_ARGS__); } while(0)
 
 #ifdef __cplusplus
 }

source/client/src/clientEnv.c

@@ -69,14 +69,25 @@ static void deregisterRequest(SRequestObj *pRequest) {
   int32_t currentInst = atomic_sub_fetch_64((int64_t *)&pActivity->currentRequests, 1);
   int32_t num = atomic_sub_fetch_32(&pTscObj->numOfReqs, 1);
 
-  int64_t duration = taosGetTimestampUs() - pRequest->metric.start;
+  int64_t nowUs = taosGetTimestampUs();
+  int64_t duration = nowUs - pRequest->metric.start;
   tscDebug("0x%" PRIx64 " free Request from connObj: 0x%" PRIx64 ", reqId:0x%" PRIx64 " elapsed:%" PRIu64
            " ms, current:%d, app current:%d",
            pRequest->self, pTscObj->id, pRequest->requestId, duration / 1000, num, currentInst);
 
   if (QUERY_NODE_VNODE_MODIF_STMT == pRequest->stmtType) {
+    tscPerf("insert duration %" PRId64 "us: syntax:%" PRId64 "us, ctg:%" PRId64 "us, semantic:%" PRId64 "us, exec:%" PRId64 "us", 
+            duration, pRequest->metric.syntaxEnd - pRequest->metric.syntaxStart, 
+            pRequest->metric.ctgEnd - pRequest->metric.ctgStart,
+            pRequest->metric.semanticEnd - pRequest->metric.ctgEnd,
+            pRequest->metric.execEnd - pRequest->metric.semanticEnd);
     atomic_add_fetch_64((int64_t *)&pActivity->insertElapsedTime, duration);
   } else if (QUERY_NODE_SELECT_STMT == pRequest->stmtType) {
+    tscPerf("select duration %" PRId64 "us: syntax:%" PRId64 "us, ctg:%" PRId64 "us, semantic:%" PRId64 "us, exec:%" PRId64 "us", 
+            duration, pRequest->metric.syntaxEnd - pRequest->metric.syntaxStart, 
+            pRequest->metric.ctgEnd - pRequest->metric.ctgStart,
+            pRequest->metric.semanticEnd - pRequest->metric.ctgEnd,
+            pRequest->metric.execEnd - pRequest->metric.semanticEnd);
     atomic_add_fetch_64((int64_t *)&pActivity->queryElapsedTime, duration);
   }
 
@@ -330,7 +341,6 @@ void doDestroyRequest(void *p) {
   schedulerFreeJob(&pRequest->body.queryJob, 0);
 
   taosMemoryFreeClear(pRequest->msgBuf);
-  taosMemoryFreeClear(pRequest->sqlstr);
   taosMemoryFreeClear(pRequest->pDb);
 
   doFreeReqResultInfo(&pRequest->body.resInfo);
@@ -349,6 +359,7 @@ void doDestroyRequest(void *p) {
     taosMemoryFree(pRequest->body.param);
   }
 
+  taosMemoryFreeClear(pRequest->sqlstr);
   taosMemoryFree(pRequest);
   tscTrace("end to destroy request %" PRIx64 " p:%p", reqId, pRequest);
 }
@@ -393,7 +404,9 @@ void taos_init_imp(void) {
   schedulerInit();
   tscDebug("starting to initialize TAOS driver");
 
+#ifndef WINDOWS
   taosSetCoreDump(true);
+#endif
 
   initTaskQueue();
   fmFuncMgtInit();

source/client/src/clientImpl.c

@@ -842,6 +842,8 @@ void schedulerExecCb(SExecResult* pResult, void* param, int32_t code) {
     }
 
     schedulerFreeJob(&pRequest->body.queryJob, 0);
+
+    pRequest->metric.execEnd = taosGetTimestampUs();
   }
 
   taosMemoryFree(pResult);

source/client/src/clientMain.c

@@ -685,6 +685,8 @@ void retrieveMetaCallback(SMetaData *pResultMeta, void *param, int32_t code) {
   SQuery          *pQuery = pWrapper->pQuery;
   SRequestObj     *pRequest = pWrapper->pRequest;
 
+  pRequest->metric.ctgEnd = taosGetTimestampUs();
+
   if (code == TSDB_CODE_SUCCESS) {
     code = qAnalyseSqlSemantic(pWrapper->pCtx, &pWrapper->catalogReq, pResultMeta, pQuery);
     pRequest->stableQuery = pQuery->stableQuery;
@@ -693,6 +695,8 @@ void retrieveMetaCallback(SMetaData *pResultMeta, void *param, int32_t code) {
     }
   }
 
+  pRequest->metric.semanticEnd = taosGetTimestampUs();
+
   if (code == TSDB_CODE_SUCCESS) {
     if (pQuery->haveResultSet) {
       setResSchemaInfo(&pRequest->body.resInfo, pQuery->pResSchema, pQuery->numOfResCols);
@@ -784,12 +788,16 @@ void doAsyncQuery(SRequestObj *pRequest, bool updateMetaForce) {
 
   SQuery *pQuery = NULL;
 
+  pRequest->metric.syntaxStart = taosGetTimestampUs();
+
   SCatalogReq catalogReq = {.forceUpdate = updateMetaForce, .qNodeRequired = qnodeRequired(pRequest)};
   code = qParseSqlSyntax(pCxt, &pQuery, &catalogReq);
   if (code != TSDB_CODE_SUCCESS) {
     goto _error;
   }
 
+  pRequest->metric.syntaxEnd = taosGetTimestampUs();
+
   if (!updateMetaForce) {
     STscObj            *pTscObj = pRequest->pTscObj;
     SAppClusterSummary *pActivity = &pTscObj->pAppInfo->summary;
@@ -816,6 +824,8 @@ void doAsyncQuery(SRequestObj *pRequest, bool updateMetaForce) {
                            .requestObjRefId = pCxt->requestRid,
                            .mgmtEps = pCxt->mgmtEpSet};
 
+  pRequest->metric.ctgStart = taosGetTimestampUs();
+
   code = catalogAsyncGetAllMeta(pCxt->pCatalog, &conn, &catalogReq, retrieveMetaCallback, pWrapper,
                                 &pRequest->body.queryJob);
   pCxt = NULL;

source/common/src/tglobal.c

@@ -61,6 +61,7 @@ int32_t tsNumOfVnodeStreamThreads = 2;
 int32_t tsNumOfVnodeFetchThreads = 4;
 int32_t tsNumOfVnodeWriteThreads = 2;
 int32_t tsNumOfVnodeSyncThreads = 2;
+int32_t tsNumOfVnodeRsmaThreads = 2;
 int32_t tsNumOfQnodeQueryThreads = 4;
 int32_t tsNumOfQnodeFetchThreads = 4;
 int32_t tsNumOfSnodeSharedThreads = 2;
@@ -76,7 +77,7 @@ bool     tsMonitorComp = false;
 
 // telem
 bool     tsEnableTelem = true;
-int32_t  tsTelemInterval = 86400;
+int32_t  tsTelemInterval = 43200;
 char     tsTelemServer[TSDB_FQDN_LEN] = "telemetry.taosdata.com";
 uint16_t tsTelemPort = 80;
 
@@ -378,6 +379,10 @@ static int32_t taosAddServerCfg(SConfig *pCfg) {
   tsNumOfVnodeSyncThreads = TMAX(tsNumOfVnodeSyncThreads, 16);
   if (cfgAddInt32(pCfg, "numOfVnodeSyncThreads", tsNumOfVnodeSyncThreads, 1, 1024, 0) != 0) return -1;
 
+  tsNumOfVnodeRsmaThreads = tsNumOfCores;
+  tsNumOfVnodeRsmaThreads = TMAX(tsNumOfVnodeRsmaThreads, 4);
+  if (cfgAddInt32(pCfg, "numOfVnodeRsmaThreads", tsNumOfVnodeRsmaThreads, 1, 1024, 0) != 0) return -1;
+
   tsNumOfQnodeQueryThreads = tsNumOfCores * 2;
   tsNumOfQnodeQueryThreads = TMAX(tsNumOfQnodeQueryThreads, 4);
   if (cfgAddInt32(pCfg, "numOfQnodeQueryThreads", tsNumOfQnodeQueryThreads, 1, 1024, 0) != 0) return -1;
@@ -539,6 +544,7 @@ static int32_t taosSetServerCfg(SConfig *pCfg) {
   tsNumOfVnodeFetchThreads = cfgGetItem(pCfg, "numOfVnodeFetchThreads")->i32;
   tsNumOfVnodeWriteThreads = cfgGetItem(pCfg, "numOfVnodeWriteThreads")->i32;
   tsNumOfVnodeSyncThreads = cfgGetItem(pCfg, "numOfVnodeSyncThreads")->i32;
+  tsNumOfVnodeRsmaThreads = cfgGetItem(pCfg, "numOfVnodeRsmaThreads")->i32;
   tsNumOfQnodeQueryThreads = cfgGetItem(pCfg, "numOfQnodeQueryThreads")->i32;
   tsNumOfQnodeFetchThreads = cfgGetItem(pCfg, "numOfQnodeFetchThreads")->i32;
   tsNumOfSnodeSharedThreads = cfgGetItem(pCfg, "numOfSnodeSharedThreads")->i32;
@@ -783,6 +789,8 @@ int32_t taosSetCfg(SConfig *pCfg, char *name) {
         tsNumOfVnodeWriteThreads = cfgGetItem(pCfg, "numOfVnodeWriteThreads")->i32;
       } else if (strcasecmp("numOfVnodeSyncThreads", name) == 0) {
         tsNumOfVnodeSyncThreads = cfgGetItem(pCfg, "numOfVnodeSyncThreads")->i32;
+      } else if (strcasecmp("numOfVnodeRsmaThreads", name) == 0) {
+        tsNumOfVnodeRsmaThreads = cfgGetItem(pCfg, "numOfVnodeRsmaThreads")->i32;
       } else if (strcasecmp("numOfQnodeQueryThreads", name) == 0) {
         tsNumOfQnodeQueryThreads = cfgGetItem(pCfg, "numOfQnodeQueryThreads")->i32;
       } else if (strcasecmp("numOfQnodeFetchThreads", name) == 0) {

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

@@ -131,7 +131,9 @@ static int32_t mndProcessTelemTimer(SRpcMsg* pReq) {
   char* pCont = mndBuildTelemetryReport(pMnode);
   if (pCont != NULL) {
     if (taosSendHttpReport(tsTelemServer, tsTelemPort, pCont, strlen(pCont), HTTP_FLAT) != 0) {
-      mError("failed to send telemetry msg");
+      mError("failed to send telemetry report");
+    } else {
+      mTrace("succeed to send telemetry report");
     }
     taosMemoryFree(pCont);
   }

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

@@ -1308,7 +1308,7 @@ static bool mndTransPerformRedoActionStage(SMnode *pMnode, STrans *pTrans) {
     if (pTrans->policy == TRN_POLICY_ROLLBACK) {
       if (pTrans->lastAction != 0) {
         STransAction *pAction = taosArrayGet(pTrans->redoActions, pTrans->lastAction);
-        if (pAction->retryCode != 0 && pAction->retryCode != pAction->errCode) {
+        if (pAction->retryCode != 0 && pAction->retryCode == pAction->errCode) {
           if (pTrans->failedTimes < 6) {
             mError("trans:%d, stage keep on redoAction since action:%d code:0x%x not 0x%x, failedTimes:%d", pTrans->id,
                    pTrans->lastAction, pTrans->code, pAction->retryCode, pTrans->failedTimes);

source/dnode/mnode/impl/test/sma/CMakeLists.txt

@@ -5,7 +5,9 @@ target_link_libraries(
     PUBLIC sut
 )
 
-add_test(
-    NAME smaTest
-    COMMAND smaTest
-)
+if(NOT ${TD_WINDOWS})
+    add_test(
+        NAME smaTest
+        COMMAND smaTest
+    )
+endif(NOT ${TD_WINDOWS})

source/dnode/mnode/impl/test/stb/CMakeLists.txt

@@ -5,7 +5,9 @@ target_link_libraries(
     PUBLIC sut
 )
 
-add_test(
-    NAME stbTest
-    COMMAND stbTest
-)
+if(NOT ${TD_WINDOWS})
+    add_test(
+        NAME stbTest
+        COMMAND stbTest
\ No newline at end of file
+    )
+endif(NOT ${TD_WINDOWS})
\ No newline at end of file

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

@@ -33,7 +33,6 @@ extern "C" {
 // clang-format on
 
 #define RSMA_TASK_INFO_HASH_SLOT (8)
-#define RSMA_EXECUTOR_MAX        (1)
 
 typedef struct SSmaEnv       SSmaEnv;
 typedef struct SSmaStat      SSmaStat;
@@ -49,9 +48,12 @@ typedef struct SQTaskFWriter SQTaskFWriter;
 struct SSmaEnv {
   SRWLatch  lock;
   int8_t    type;
+  int8_t    flag;  // 0x01 inClose
   SSmaStat *pStat;
 };
 
+#define SMA_ENV_FLG_CLOSE ((int8_t)0x1)
+
 typedef struct {
   int8_t  inited;
   int32_t rsetId;
@@ -93,7 +95,6 @@ struct SRSmaStat {
   int64_t          refId;             // shared by fetch tasks
   volatile int64_t nBufItems;         // number of items in queue buffer
   SRWLatch         lock;              // r/w lock for rsma fs(e.g. qtaskinfo)
-  volatile int8_t  nExecutor;         // [1, max(half of query threads, 4)]
   int8_t           triggerStat;       // shared by fetch tasks
   int8_t           commitStat;        // 0 not in committing, 1 in committing
   SArray          *aTaskFile;         // qTaskFiles committed recently(for recovery/snapshot r/w)
@@ -107,6 +108,7 @@ struct SSmaStat {
     SRSmaStat rsmaStat;  // rollup sma
   };
   T_REF_DECLARE()
+  char data[];
 };
 
 #define SMA_STAT_TSMA(s)     (&(s)->tsmaStat)

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

@@ -189,7 +189,6 @@ SSubmitReq* tqBlockToSubmit(SVnode* pVnode, const SArray* pBlocks, const STSchem
 int32_t smaInit();
 void    smaCleanUp();
 int32_t smaOpen(SVnode* pVnode);
-int32_t smaPreClose(SSma* pSma);
 int32_t smaClose(SSma* pSma);
 int32_t smaBegin(SSma* pSma);
 int32_t smaSyncPreCommit(SSma* pSma);
@@ -199,7 +198,6 @@ int32_t smaAsyncPreCommit(SSma* pSma);
 int32_t smaAsyncCommit(SSma* pSma);
 int32_t smaAsyncPostCommit(SSma* pSma);
 int32_t smaDoRetention(SSma* pSma, int64_t now);
-int32_t smaProcessExec(SSma* pSma, void* pMsg);
 
 int32_t tdProcessTSmaCreate(SSma* pSma, int64_t version, const char* msg);
 int32_t tdProcessTSmaInsert(SSma* pSma, int64_t indexUid, const char* msg);
@@ -323,7 +321,6 @@ struct SVnode {
   TdThreadMutex lock;
   bool          blocked;
   bool          restored;
-  bool          inClose;
   tsem_t        syncSem;
   SQHandle*     pQuery;
 };

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

@@ -23,11 +23,13 @@ extern SSmaMgmt smaMgmt;
 
 // declaration of static functions
 
-static int32_t  tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType, const SSma *pSma);
-static SSmaEnv *tdNewSmaEnv(const SSma *pSma, int8_t smaType, const char *path);
-static int32_t  tdInitSmaEnv(SSma *pSma, int8_t smaType, const char *path, SSmaEnv **pEnv);
-static void    *tdFreeTSmaStat(STSmaStat *pStat);
-static void     tdDestroyRSmaStat(void *pRSmaStat);
+static int32_t tdNewSmaEnv(SSma *pSma, int8_t smaType, SSmaEnv **ppEnv);
+static int32_t tdInitSmaEnv(SSma *pSma, int8_t smaType, SSmaEnv **ppEnv);
+static int32_t tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType, const SSma *pSma);
+static int32_t tdRsmaStartExecutor(const SSma *pSma);
+static int32_t tdRsmaStopExecutor(const SSma *pSma);
+static void   *tdFreeTSmaStat(STSmaStat *pStat);
+static void    tdDestroyRSmaStat(void *pRSmaStat);
 
 /**
  * @brief rsma init
@@ -97,35 +99,42 @@ void smaCleanUp() {
   }
 }
 
-static SSmaEnv *tdNewSmaEnv(const SSma *pSma, int8_t smaType, const char *path) {
+static int32_t tdNewSmaEnv(SSma *pSma, int8_t smaType, SSmaEnv **ppEnv) {
   SSmaEnv *pEnv = NULL;
 
   pEnv = (SSmaEnv *)taosMemoryCalloc(1, sizeof(SSmaEnv));
+  *ppEnv = pEnv;
   if (!pEnv) {
     terrno = TSDB_CODE_OUT_OF_MEMORY;
-    return NULL;
+    return TSDB_CODE_FAILED;
   }
 
   SMA_ENV_TYPE(pEnv) = smaType;
 
   taosInitRWLatch(&(pEnv->lock));
 
+  (smaType == TSDB_SMA_TYPE_TIME_RANGE) ? atomic_store_ptr(&SMA_TSMA_ENV(pSma), *ppEnv)
+                                        : atomic_store_ptr(&SMA_RSMA_ENV(pSma), *ppEnv);
+
   if (tdInitSmaStat(&SMA_ENV_STAT(pEnv), smaType, pSma) != TSDB_CODE_SUCCESS) {
     tdFreeSmaEnv(pEnv);
-    return NULL;
+    *ppEnv = NULL;
+    (smaType == TSDB_SMA_TYPE_TIME_RANGE) ? atomic_store_ptr(&SMA_TSMA_ENV(pSma), NULL)
+                                          : atomic_store_ptr(&SMA_RSMA_ENV(pSma), NULL);
+    return TSDB_CODE_FAILED;
   }
 
-  return pEnv;
+  return TSDB_CODE_SUCCESS;
 }
 
-static int32_t tdInitSmaEnv(SSma *pSma, int8_t smaType, const char *path, SSmaEnv **pEnv) {
-  if (!pEnv) {
+static int32_t tdInitSmaEnv(SSma *pSma, int8_t smaType, SSmaEnv **ppEnv) {
+  if (!ppEnv) {
     terrno = TSDB_CODE_INVALID_PTR;
     return TSDB_CODE_FAILED;
   }
 
-  if (!(*pEnv)) {
-    if (!(*pEnv = tdNewSmaEnv(pSma, smaType, path))) {
+  if (!(*ppEnv)) {
+    if (tdNewSmaEnv(pSma, smaType, ppEnv) != TSDB_CODE_SUCCESS) {
       return TSDB_CODE_FAILED;
     }
   }
@@ -199,7 +208,7 @@ static int32_t tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType, const SSma *pS
    * tdInitSmaStat invoked in other multithread environment later.
    */
   if (!(*pSmaStat)) {
-    *pSmaStat = (SSmaStat *)taosMemoryCalloc(1, sizeof(SSmaStat));
+    *pSmaStat = (SSmaStat *)taosMemoryCalloc(1, sizeof(SSmaStat) + sizeof(TdThread) * tsNumOfVnodeRsmaThreads);
     if (!(*pSmaStat)) {
       terrno = TSDB_CODE_OUT_OF_MEMORY;
       return TSDB_CODE_FAILED;
@@ -231,6 +240,10 @@ static int32_t tdInitSmaStat(SSmaStat **pSmaStat, int8_t smaType, const SSma *pS
       if (!RSMA_INFO_HASH(pRSmaStat)) {
         return TSDB_CODE_FAILED;
       }
+
+      if (tdRsmaStartExecutor(pSma) < 0) {
+        return TSDB_CODE_FAILED;
+      }
     } else if (smaType == TSDB_SMA_TYPE_TIME_RANGE) {
       // TODO
     } else {
@@ -291,6 +304,9 @@ static void tdDestroyRSmaStat(void *pRSmaStat) {
       }
     }
 
+    // step 4:
+    tdRsmaStopExecutor(pSma);
+
     // step 5: free pStat
     taosMemoryFreeClear(pStat);
   }
@@ -381,17 +397,70 @@ int32_t tdCheckAndInitSmaEnv(SSma *pSma, int8_t smaType) {
   pEnv = (smaType == TSDB_SMA_TYPE_TIME_RANGE) ? atomic_load_ptr(&SMA_TSMA_ENV(pSma))
                                                : atomic_load_ptr(&SMA_RSMA_ENV(pSma));
   if (!pEnv) {
-    char rname[TSDB_FILENAME_LEN] = {0};
-
-    if (tdInitSmaEnv(pSma, smaType, rname, &pEnv) < 0) {
+    if (tdInitSmaEnv(pSma, smaType, &pEnv) < 0) {
       tdUnLockSma(pSma);
       return TSDB_CODE_FAILED;
     }
-
-    (smaType == TSDB_SMA_TYPE_TIME_RANGE) ? atomic_store_ptr(&SMA_TSMA_ENV(pSma), pEnv)
-                                          : atomic_store_ptr(&SMA_RSMA_ENV(pSma), pEnv);
   }
   tdUnLockSma(pSma);
 
   return TSDB_CODE_SUCCESS;
 };
+
+void *tdRSmaExecutorFunc(void *param) {
+  setThreadName("vnode-rsma");
+
+  tdRSmaProcessExecImpl((SSma *)param, RSMA_EXEC_OVERFLOW);
+  return NULL;
+}
+
+static int32_t tdRsmaStartExecutor(const SSma *pSma) {
+  TdThreadAttr thAttr = {0};
+  taosThreadAttrInit(&thAttr);
+  taosThreadAttrSetDetachState(&thAttr, PTHREAD_CREATE_JOINABLE);
+
+  SSmaEnv  *pEnv = SMA_RSMA_ENV(pSma);
+  SSmaStat *pStat = SMA_ENV_STAT(pEnv);
+  TdThread *pthread = (TdThread *)&pStat->data;
+
+  for (int32_t i = 0; i < tsNumOfVnodeRsmaThreads; ++i) {
+    if (taosThreadCreate(&pthread[i], &thAttr, tdRSmaExecutorFunc, (void *)pSma) != 0) {
+      terrno = TAOS_SYSTEM_ERROR(errno);
+      smaError("vgId:%d, failed to create pthread for rsma since %s", SMA_VID(pSma), terrstr());
+      return -1;
+    }
+    smaDebug("vgId:%d, success to create pthread for rsma", SMA_VID(pSma));
+  }
+
+  taosThreadAttrDestroy(&thAttr);
+  return 0;
+}
+
+static int32_t tdRsmaStopExecutor(const SSma *pSma) {
+  if (pSma && VND_IS_RSMA(pSma->pVnode)) {
+    SSmaEnv   *pEnv = NULL;
+    SSmaStat  *pStat = NULL;
+    SRSmaStat *pRSmaStat = NULL;
+    TdThread  *pthread = NULL;
+
+    if (!(pEnv = SMA_RSMA_ENV(pSma)) || !(pStat = SMA_ENV_STAT(pEnv))) {
+      return 0;
+    }
+
+    pEnv->flag |= SMA_ENV_FLG_CLOSE;
+    pRSmaStat = (SRSmaStat *)pStat;
+    pthread = (TdThread *)&pStat->data;
+
+    for (int32_t i = 0; i < tsNumOfVnodeRsmaThreads; ++i) {
+      tsem_post(&(pRSmaStat->notEmpty));
+    }
+
+    for (int32_t i = 0; i < tsNumOfVnodeRsmaThreads; ++i) {
+      if (taosCheckPthreadValid(pthread[i])) {
+        smaDebug("vgId:%d, start to join pthread for rsma:%" PRId64, SMA_VID(pSma), pthread[i]);
+        taosThreadJoin(pthread[i], NULL);
+      }
+    }
+  }
+  return 0;
+}
\ No newline at end of file

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

@@ -146,20 +146,6 @@ int32_t smaClose(SSma *pSma) {
   return 0;
 }
 
-int32_t smaPreClose(SSma *pSma) {
-  if (pSma && VND_IS_RSMA(pSma->pVnode)) {
-    SSmaEnv   *pEnv = NULL;
-    SRSmaStat *pStat = NULL;
-    if (!(pEnv = SMA_RSMA_ENV(pSma)) || !(pStat = (SRSmaStat *)SMA_ENV_STAT(pEnv))) {
-      return 0;
-    }
-    for (int32_t i = 0; i < RSMA_EXECUTOR_MAX; ++i) {
-      tsem_post(&(pStat->notEmpty));
-    }
-  }
-  return 0;
-}
-
 /**
  * @brief rsma env restore
  * 

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

@@ -621,7 +621,7 @@ static int32_t tdFetchSubmitReqSuids(SSubmitReq *pMsg, STbUidStore *pStore) {
  */
 int32_t smaDoRetention(SSma *pSma, int64_t now) {
   int32_t code = TSDB_CODE_SUCCESS;
-  if (VND_IS_RSMA(pSma->pVnode)) {
+  if (!VND_IS_RSMA(pSma->pVnode)) {
     return code;
   }
 
@@ -734,10 +734,12 @@ static int32_t tdExecuteRSmaImplAsync(SSma *pSma, const void *pMsg, int32_t inpu
 
   SRSmaStat *pRSmaStat = SMA_RSMA_STAT(pSma);
 
-  tsem_post(&(pRSmaStat->notEmpty));
-
   int64_t nItems = atomic_fetch_add_64(&pRSmaStat->nBufItems, 1);
 
+  if (atomic_load_8(&pInfo->assigned) == 0) {
+    tsem_post(&(pRSmaStat->notEmpty));
+  }
+
   // smoothing consume
   int32_t n = nItems / RSMA_QTASKEXEC_SMOOTH_SIZE;
   if (n > 1) {
@@ -911,39 +913,6 @@ static int32_t tdExecuteRSmaAsync(SSma *pSma, const void *pMsg, int32_t inputTyp
   return TSDB_CODE_SUCCESS;
 }
 
-static int32_t tdRSmaExecCheck(SSma *pSma) {
-  SRSmaStat *pRSmaStat = SMA_RSMA_STAT(pSma);
-
-  if (atomic_load_8(&pRSmaStat->nExecutor) >= TMIN(RSMA_EXECUTOR_MAX, tsNumOfVnodeQueryThreads / 2)) {
-    return TSDB_CODE_SUCCESS;
-  }
-
-  SRSmaExecMsg fetchMsg;
-  int32_t      contLen = sizeof(SMsgHead);
-  void        *pBuf = rpcMallocCont(0 + contLen);
-
-  ((SMsgHead *)pBuf)->vgId = SMA_VID(pSma);
-  ((SMsgHead *)pBuf)->contLen = sizeof(SMsgHead);
-
-  SRpcMsg rpcMsg = {
-      .code = 0,
-      .msgType = TDMT_VND_EXEC_RSMA,
-      .pCont = pBuf,
-      .contLen = contLen,
-  };
-
-  if ((terrno = tmsgPutToQueue(&pSma->pVnode->msgCb, QUERY_QUEUE, &rpcMsg)) != 0) {
-    smaError("vgId:%d, failed to put rsma exec msg into query-queue since %s", SMA_VID(pSma), terrstr());
-    goto _err;
-  }
-
-  smaDebug("vgId:%d, success to put rsma fetch msg into query-queue", SMA_VID(pSma));
-
-  return TSDB_CODE_SUCCESS;
-_err:
-  return TSDB_CODE_FAILED;
-}
-
 int32_t tdProcessRSmaSubmit(SSma *pSma, void *pMsg, int32_t inputType) {
   SSmaEnv *pEnv = SMA_RSMA_ENV(pSma);
   if (!pEnv) {
@@ -974,10 +943,6 @@ int32_t tdProcessRSmaSubmit(SSma *pSma, void *pMsg, int32_t inputType) {
           goto _err;
         }
       }
-
-      if (tdRSmaExecCheck(pSma) < 0) {
-        goto _err;
-      }
     }
   }
   tdUidStoreDestory(&uidStore);
@@ -1563,7 +1528,9 @@ static void tdRSmaFetchTrigger(void *param, void *tmrId) {
       ASSERT(qItem->level == pItem->level);
       ASSERT(qItem->fetchLevel == pItem->fetchLevel);
 #endif
-      tsem_post(&(pStat->notEmpty));
+      if (atomic_load_8(&pRSmaInfo->assigned) == 0) {
+        tsem_post(&(pStat->notEmpty));
+      }
       smaInfo("vgId:%d, rsma fetch task planned for level:%" PRIi8 " suid:%" PRIi64, SMA_VID(pSma), pItem->level,
               pRSmaInfo->suid);
     } break;
@@ -1591,9 +1558,11 @@ _end:
 }
 
 static void tdFreeRSmaSubmitItems(SArray *pItems) {
+  ASSERT(taosArrayGetSize(pItems) > 0);
   for (int32_t i = 0; i < taosArrayGetSize(pItems); ++i) {
     taosFreeQitem(*(void **)taosArrayGet(pItems, i));
   }
+  taosArrayClear(pItems);
 }
 
 /**
@@ -1703,6 +1672,7 @@ _err:
  * @param type
  * @return int32_t
  */
+
 int32_t tdRSmaProcessExecImpl(SSma *pSma, ERsmaExecType type) {
   SVnode    *pVnode = pSma->pVnode;
   SSmaEnv   *pEnv = SMA_RSMA_ENV(pSma);
@@ -1722,41 +1692,53 @@ int32_t tdRSmaProcessExecImpl(SSma *pSma, ERsmaExecType type) {
     goto _err;
   }
 
-  bool isBusy = false;
   while (true) {
-    isBusy = false;
     // step 1: rsma exec - consume data in buffer queue for all suids
     if (type == RSMA_EXEC_OVERFLOW || type == RSMA_EXEC_COMMIT) {
-      void *pIter = taosHashIterate(infoHash, NULL);  // infoHash has r/w lock
-      while (pIter) {
+      void *pIter = NULL;
+      while ((pIter = taosHashIterate(infoHash, pIter))) {
         SRSmaInfo *pInfo = *(SRSmaInfo **)pIter;
-        int64_t    itemSize = 0;
-        if ((itemSize = taosQueueItemSize(pInfo->queue)) || RSMA_INFO_ITEM(pInfo, 0)->fetchLevel ||
-            RSMA_INFO_ITEM(pInfo, 1)->fetchLevel) {
-          smaDebug("vgId:%d, queueItemSize is %" PRIi64 " execType:%" PRIi8, SMA_VID(pSma), itemSize, type);
-          if (atomic_val_compare_exchange_8(&pInfo->assigned, 0, 1) == 0) {
-            taosReadAllQitems(pInfo->queue, pInfo->qall);  // queue has mutex lock
-            int32_t qallItemSize = taosQallItemSize(pInfo->qall);
-            if (qallItemSize > 0) {
-              tdRSmaBatchExec(pSma, pInfo, pInfo->qall, pSubmitArr, type);
-            }
-
-            if (type == RSMA_EXEC_OVERFLOW) {
-              tdRSmaFetchAllResult(pSma, pInfo, pSubmitArr);
+        if (atomic_val_compare_exchange_8(&pInfo->assigned, 0, 1) == 0) {
+          if ((taosQueueItemSize(pInfo->queue) > 0) || RSMA_INFO_ITEM(pInfo, 0)->fetchLevel ||
+              RSMA_INFO_ITEM(pInfo, 1)->fetchLevel) {
+            int32_t batchCnt = -1;
+            int32_t batchMax = taosHashGetSize(infoHash) / tsNumOfVnodeRsmaThreads;
+            bool    occupied = (batchMax <= 1);
+            if (batchMax > 1) {
+              batchMax = 100 / batchMax;
             }
-
-            if (qallItemSize > 0) {
-              // subtract the item size after the task finished, commit should wait for all items be consumed
-              atomic_fetch_sub_64(&pRSmaStat->nBufItems, qallItemSize);
-              isBusy = true;
+            while (occupied || (++batchCnt < batchMax)) {    // greedy mode
+              taosReadAllQitems(pInfo->queue, pInfo->qall);  // queue has mutex lock
+              int32_t qallItemSize = taosQallItemSize(pInfo->qall);
+              if (qallItemSize > 0) {
+                tdRSmaBatchExec(pSma, pInfo, pInfo->qall, pSubmitArr, type);
+                smaDebug("vgId:%d, batchSize:%d, execType:%" PRIi8, SMA_VID(pSma), qallItemSize, type);
+              }
+
+              if (type == RSMA_EXEC_OVERFLOW) {
+                tdRSmaFetchAllResult(pSma, pInfo, pSubmitArr);
+              }
+
+              if (qallItemSize > 0) {
+                atomic_fetch_sub_64(&pRSmaStat->nBufItems, qallItemSize);
+                continue;
+              } else if (RSMA_INFO_ITEM(pInfo, 0)->fetchLevel || RSMA_INFO_ITEM(pInfo, 1)->fetchLevel) {
+                continue;
+              }
+
+              break;
             }
-            ASSERT(1 == atomic_val_compare_exchange_8(&pInfo->assigned, 1, 0));
           }
+          ASSERT(1 == atomic_val_compare_exchange_8(&pInfo->assigned, 1, 0));
         }
-        pIter = taosHashIterate(infoHash, pIter);
       }
       if (type == RSMA_EXEC_COMMIT) {
-        break;
+        if (atomic_load_64(&pRSmaStat->nBufItems) <= 0) {
+          break;
+        } else {
+          // commit should wait for all items be consumed
+          continue;
+        }
       }
     }
 #if 0
@@ -1790,16 +1772,19 @@ int32_t tdRSmaProcessExecImpl(SSma *pSma, ERsmaExecType type) {
     }
 
     if (atomic_load_64(&pRSmaStat->nBufItems) <= 0) {
-      if (pVnode->inClose) {
+      if (pEnv->flag & SMA_ENV_FLG_CLOSE) {
         break;
       }
+      
       tsem_wait(&pRSmaStat->notEmpty);
-      if (pVnode->inClose && (atomic_load_64(&pRSmaStat->nBufItems) <= 0)) {
-        smaInfo("vgId:%d, exec task end, inClose:%d, nBufItems:%" PRIi64, SMA_VID(pSma), pVnode->inClose,
+
+      if ((pEnv->flag & SMA_ENV_FLG_CLOSE) && (atomic_load_64(&pRSmaStat->nBufItems) <= 0)) {
+        smaInfo("vgId:%d, exec task end, flag:%" PRIi8 ", nBufItems:%" PRIi64, SMA_VID(pSma), pEnv->flag,
                 atomic_load_64(&pRSmaStat->nBufItems));
         break;
       }
     }
+
   }  // end of while(true)
 
 _end:
@@ -1809,39 +1794,3 @@ _err:
   taosArrayDestroy(pSubmitArr);
   return TSDB_CODE_FAILED;
 }
-
-/**
- * @brief exec rsma level 1data, fetch result of level 2/3 and submit
- *
- * @param pSma
- * @param pMsg
- * @return int32_t
- */
-int32_t smaProcessExec(SSma *pSma, void *pMsg) {
-  SRpcMsg   *pRpcMsg = (SRpcMsg *)pMsg;
-  SRSmaStat *pRSmaStat = SMA_RSMA_STAT(pSma);
-
-  if (!pRpcMsg || pRpcMsg->contLen < sizeof(SMsgHead)) {
-    terrno = TSDB_CODE_RSMA_FETCH_MSG_MSSED_UP;
-    goto _err;
-  }
-  smaDebug("vgId:%d, begin to process rsma exec msg by TID:%p", SMA_VID(pSma), (void *)taosGetSelfPthreadId());
-
-  int8_t nOld = atomic_fetch_add_8(&pRSmaStat->nExecutor, 1);
-
-  if (nOld < TMIN(RSMA_EXECUTOR_MAX, tsNumOfVnodeQueryThreads / 2)) {
-    if (tdRSmaProcessExecImpl(pSma, RSMA_EXEC_OVERFLOW) < 0) {
-      goto _err;
-    }
-  } else {
-    atomic_fetch_sub_8(&pRSmaStat->nExecutor, 1);
-  }
-
-  smaDebug("vgId:%d, success to process rsma exec msg by TID:%p", SMA_VID(pSma), (void *)taosGetSelfPthreadId());
-  return TSDB_CODE_SUCCESS;
-_err:
-  atomic_fetch_sub_8(&pRSmaStat->nExecutor, 1);
-  smaError("vgId:%d, failed to process rsma exec msg by TID:%p since %s", SMA_VID(pSma), (void *)taosGetSelfPthreadId(),
-           terrstr());
-  return TSDB_CODE_FAILED;
-}

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

@@ -79,6 +79,10 @@ STQ* tqOpen(const char* path, SVnode* pVnode) {
     ASSERT(0);
   }
 
+  if (streamLoadTasks(pTq->pStreamMeta) < 0) {
+    ASSERT(0);
+  }
+
   return pTq;
 }
 
@@ -648,17 +652,28 @@ int32_t tqExpandTask(STQ* pTq, SStreamTask* pTask) {
 
   // expand executor
   if (pTask->taskLevel == TASK_LEVEL__SOURCE) {
+    pTask->pState = streamStateOpen(pTq->pStreamMeta->path, pTask);
+    if (pTask->pState == NULL) {
+      return -1;
+    }
+
     SReadHandle handle = {
         .meta = pTq->pVnode->pMeta,
         .vnode = pTq->pVnode,
         .initTqReader = 1,
+        .pStateBackend = pTask->pState,
     };
     pTask->exec.executor = qCreateStreamExecTaskInfo(pTask->exec.qmsg, &handle);
     ASSERT(pTask->exec.executor);
   } else if (pTask->taskLevel == TASK_LEVEL__AGG) {
+    pTask->pState = streamStateOpen(pTq->pStreamMeta->path, pTask);
+    if (pTask->pState == NULL) {
+      return -1;
+    }
     SReadHandle mgHandle = {
         .vnode = NULL,
         .numOfVgroups = (int32_t)taosArrayGetSize(pTask->childEpInfo),
+        .pStateBackend = pTask->pState,
     };
     pTask->exec.executor = qCreateStreamExecTaskInfo(pTask->exec.qmsg, &mgHandle);
     ASSERT(pTask->exec.executor);

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

@@ -1416,7 +1416,7 @@ static int32_t doMergeBufAndFileRows_Rv(STsdbReader* pReader, STableBlockScanInf
   int64_t minKey = 0;
   if (pReader->order == TSDB_ORDER_ASC) {
     minKey = INT64_MAX;   // chosen the minimum value
-    if (minKey > tsLast && pLastBlockReader->lastBlockData.nRow > 0) {
+    if (minKey > tsLast && hasDataInLastBlock(pLastBlockReader)) {
       minKey = tsLast;
     }
 
@@ -1429,7 +1429,7 @@ static int32_t doMergeBufAndFileRows_Rv(STsdbReader* pReader, STableBlockScanInf
     }
   } else {
     minKey = INT64_MIN;
-    if (minKey < tsLast && pLastBlockReader->lastBlockData.nRow > 0) {
+    if (minKey < tsLast && hasDataInLastBlock(pLastBlockReader)) {
       minKey = tsLast;
     }
 

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

@@ -87,7 +87,6 @@ SVnode *vnodeOpen(const char *path, STfs *pTfs, SMsgCb msgCb) {
   pVnode->msgCb = msgCb;
   taosThreadMutexInit(&pVnode->lock, NULL);
   pVnode->blocked = false;
-  pVnode->inClose = false;
 
   tsem_init(&pVnode->syncSem, 0, 0);
   tsem_init(&(pVnode->canCommit), 0, 1);
@@ -182,8 +181,6 @@ _err:
 void vnodePreClose(SVnode *pVnode) {
   if (pVnode) {
     syncLeaderTransfer(pVnode->sync);
-    pVnode->inClose = true;
-    smaPreClose(pVnode->pSma);
   }
 }
 

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

@@ -301,8 +301,6 @@ int32_t vnodeProcessQueryMsg(SVnode *pVnode, SRpcMsg *pMsg) {
       return qWorkerProcessQueryMsg(&handle, pVnode->pQuery, pMsg, 0);
     case TDMT_SCH_QUERY_CONTINUE:
       return qWorkerProcessCQueryMsg(&handle, pVnode->pQuery, pMsg, 0);
-    case TDMT_VND_EXEC_RSMA:
-      return smaProcessExec(pVnode->pSma, pMsg);
     default:
       vError("unknown msg type:%d in query queue", pMsg->msgType);
       return TSDB_CODE_VND_APP_ERROR;
@@ -380,14 +378,14 @@ static int32_t vnodeProcessTrimReq(SVnode *pVnode, int64_t version, void *pReq,
   int32_t     code = 0;
   SVTrimDbReq trimReq = {0};
 
-  vInfo("vgId:%d, trim vnode request will be processed, time:%d", pVnode->config.vgId, trimReq.timestamp);
-
   // decode
   if (tDeserializeSVTrimDbReq(pReq, len, &trimReq) != 0) {
     code = TSDB_CODE_INVALID_MSG;
     goto _exit;
   }
 
+  vInfo("vgId:%d, trim vnode request will be processed, time:%d", pVnode->config.vgId, trimReq.timestamp);
+
   // process
   code = tsdbDoRetention(pVnode->pTsdb, trimReq.timestamp);
   if (code) goto _exit;

source/libs/executor/inc/executorimpl.h

@@ -150,6 +150,7 @@ typedef struct {
   SQueryTableDataCond tableCond;
   int64_t recoverStartVer;
   int64_t recoverEndVer;
+  SStreamState* pState;
 } SStreamTaskInfo;
 
 typedef struct {

source/libs/executor/src/executorimpl.c

@@ -148,20 +148,6 @@ static int32_t doCopyToSDataBlock(SExecTaskInfo* pTaskInfo, SSDataBlock* pBlock,
 static void initCtxOutputBuffer(SqlFunctionCtx* pCtx, int32_t size);
 static void doSetTableGroupOutputBuf(SOperatorInfo* pOperator, int32_t numOfOutput, uint64_t groupId);
 
-// setup the output buffer for each operator
-static bool hasNull(SColumn* pColumn, SColumnDataAgg* pStatis) {
-  if (TSDB_COL_IS_TAG(pColumn->flag) || TSDB_COL_IS_UD_COL(pColumn->flag) ||
-      pColumn->colId == PRIMARYKEY_TIMESTAMP_COL_ID) {
-    return false;
-  }
-
-  if (pStatis != NULL && pStatis->numOfNull == 0) {
-    return false;
-  }
-
-  return true;
-}
-
 #if 0
 static bool chkResultRowFromKey(STaskRuntimeEnv* pRuntimeEnv, SResultRowInfo* pResultRowInfo, char* pData,
                                 int16_t bytes, bool masterscan, uint64_t uid) {
@@ -392,7 +378,7 @@ static void functionCtxSave(SqlFunctionCtx* pCtx, SFunctionCtxStatus* pStatus) {
 
 static void functionCtxRestore(SqlFunctionCtx* pCtx, SFunctionCtxStatus* pStatus) {
   pCtx->input.colDataAggIsSet = pStatus->hasAgg;
-  pCtx->input.numOfRows  = pStatus->numOfRows;
+  pCtx->input.numOfRows = pStatus->numOfRows;
   pCtx->input.startRowIndex = pStatus->startOffset;
 }
 
@@ -3715,7 +3701,7 @@ static int32_t initFillInfo(SFillOperatorInfo* pInfo, SExprInfo* pExpr, int32_t
                             const char* id, SInterval* pInterval, int32_t fillType, int32_t order) {
   SFillColInfo* pColInfo = createFillColInfo(pExpr, numOfCols, pNotFillExpr, numOfNotFillCols, pValNode);
 
-  int64_t startKey = (order == TSDB_ORDER_ASC) ? win.skey : win.ekey;
+  int64_t     startKey = (order == TSDB_ORDER_ASC) ? win.skey : win.ekey;
   STimeWindow w = getAlignQueryTimeWindow(pInterval, pInterval->precision, startKey);
   w = getFirstQualifiedTimeWindow(startKey, &w, pInterval, order);
 
@@ -3988,15 +3974,15 @@ int32_t generateGroupIdMap(STableListInfo* pTableListInfo, SReadHandle* pHandle,
 
   bool assignUid = groupbyTbname(group);
 
-  size_t  numOfTables = taosArrayGetSize(pTableListInfo->pTableList);
+  size_t numOfTables = taosArrayGetSize(pTableListInfo->pTableList);
 
-  if(assignUid){
+  if (assignUid) {
     for (int32_t i = 0; i < numOfTables; i++) {
       STableKeyInfo* info = taosArrayGet(pTableListInfo->pTableList, i);
       info->groupId = info->uid;
       taosHashPut(pTableListInfo->map, &(info->uid), sizeof(uint64_t), &info->groupId, sizeof(uint64_t));
     }
-  }else{
+  } else {
     int32_t code = getColInfoResultForGroupby(pHandle->meta, group, pTableListInfo);
     if (code != TSDB_CODE_SUCCESS) {
       return code;
@@ -4615,6 +4601,10 @@ int32_t createExecTaskInfoImpl(SSubplan* pPlan, SExecTaskInfo** pTaskInfo, SRead
     goto _complete;
   }
 
+  if (pHandle && pHandle->pStateBackend) {
+    (*pTaskInfo)->streamInfo.pState = pHandle->pStateBackend;
+  }
+
   (*pTaskInfo)->sql = sql;
   sql = NULL;
   (*pTaskInfo)->pSubplan = pPlan;

source/libs/executor/src/scanoperator.c

@@ -1281,6 +1281,42 @@ static SSDataBlock* doStreamScan(SOperatorInfo* pOperator) {
   SExecTaskInfo*   pTaskInfo = pOperator->pTaskInfo;
   SStreamScanInfo* pInfo = pOperator->info;
 
+#if 0
+  SStreamState* pState = pTaskInfo->streamInfo.pState;
+  if (pState) {
+    printf(">>>>>>>> stream write backend\n");
+    SWinKey key = {
+        .ts = 1,
+        .groupId = 2,
+    };
+    char tmp[100] = "abcdefg1";
+    if (streamStatePut(pState, &key, &tmp, strlen(tmp) + 1) < 0) {
+      ASSERT(0);
+    }
+
+    key.ts = 2;
+    char tmp2[100] = "abcdefg2";
+    if (streamStatePut(pState, &key, &tmp2, strlen(tmp2) + 1) < 0) {
+      ASSERT(0);
+    }
+
+    key.groupId = 5;
+    key.ts = 1;
+    char tmp3[100] = "abcdefg3";
+    if (streamStatePut(pState, &key, &tmp3, strlen(tmp3) + 1) < 0) {
+      ASSERT(0);
+    }
+
+    char*   val2 = NULL;
+    int32_t sz;
+    if (streamStateGet(pState, &key, (void**)&val2, &sz) < 0) {
+      ASSERT(0);
+    }
+    printf("stream read %s %d\n", val2, sz);
+    streamFreeVal(val2);
+  }
+#endif
+
   qDebug("stream scan called");
   if (pTaskInfo->streamInfo.prepareStatus.type == TMQ_OFFSET__LOG) {
     while (1) {

source/libs/executor/src/tfill.c

@@ -36,6 +36,7 @@
 #define GET_DEST_SLOT_ID(_p) ((_p)->pExpr->base.resSchema.slotId)
 
 static void doSetVal(SColumnInfoData* pDstColInfoData, int32_t rowIndex, const SGroupKeys* pKey);
+static bool fillIfWindowPseudoColumn(SFillInfo* pFillInfo, SFillColInfo* pCol, SColumnInfoData* pDstColInfoData, int32_t rowIndex);
 
 static void setNullRow(SSDataBlock* pBlock, SFillInfo* pFillInfo, int32_t rowIndex) {
   for(int32_t i = 0; i < pFillInfo->numOfCols; ++i) {
@@ -43,9 +44,8 @@ static void setNullRow(SSDataBlock* pBlock, SFillInfo* pFillInfo, int32_t rowInd
     int32_t dstSlotId = GET_DEST_SLOT_ID(pCol);
     SColumnInfoData* pDstColInfo = taosArrayGet(pBlock->pDataBlock, dstSlotId);
     if (pCol->notFillCol) {
-      if (pDstColInfo->info.type == TSDB_DATA_TYPE_TIMESTAMP) {
-        colDataAppend(pDstColInfo, rowIndex, (const char*)&pFillInfo->currentKey, false);
-      } else {
+      bool filled = fillIfWindowPseudoColumn(pFillInfo, pCol, pDstColInfo, rowIndex);
+      if (!filled) {
         SArray* p = FILL_IS_ASC_FILL(pFillInfo) ? pFillInfo->prev.pRowVal : pFillInfo->next.pRowVal;
         SGroupKeys* pKey = taosArrayGet(p, i);
         doSetVal(pDstColInfo, rowIndex, pKey);
@@ -76,6 +76,35 @@ static void doSetUserSpecifiedValue(SColumnInfoData* pDst, SVariant* pVar, int32
   }
 }
 
+//fill windows pseudo column, _wstart, _wend, _wduration and return true, otherwise return false
+static bool fillIfWindowPseudoColumn(SFillInfo* pFillInfo, SFillColInfo* pCol, SColumnInfoData* pDstColInfoData, int32_t rowIndex) {
+  if (!pCol->notFillCol) {
+    return false;
+  }
+  if (pCol->pExpr->pExpr->nodeType == QUERY_NODE_COLUMN) {
+    if (pCol->pExpr->base.numOfParams != 1) {
+      return false;
+    }
+    if (pCol->pExpr->base.pParam[0].pCol->colType == COLUMN_TYPE_WINDOW_START) {
+      colDataAppend(pDstColInfoData, rowIndex, (const char*)&pFillInfo->currentKey, false);
+      return true;
+    } else if (pCol->pExpr->base.pParam[0].pCol->colType == COLUMN_TYPE_WINDOW_END) {
+      //TODO: include endpoint
+      SInterval* pInterval = &pFillInfo->interval;
+      int32_t step = (pFillInfo->order == TSDB_ORDER_ASC) ? 1 : -1;
+      int64_t windowEnd =
+          taosTimeAdd(pFillInfo->currentKey, pInterval->sliding * step, pInterval->slidingUnit, pInterval->precision);
+      colDataAppend(pDstColInfoData, rowIndex, (const char*)&windowEnd, false);
+      return true;
+    } else if (pCol->pExpr->base.pParam[0].pCol->colType == COLUMN_TYPE_WINDOW_DURATION) {
+      //TODO: include endpoint
+      colDataAppend(pDstColInfoData, rowIndex, (const char*)&pFillInfo->interval.sliding, false);
+      return true;
+    }
+  }
+  return false;
+}
+
 static void doFillOneRow(SFillInfo* pFillInfo, SSDataBlock* pBlock, SSDataBlock* pSrcBlock, int64_t ts,
                          bool outOfBound) {
   SPoint  point1, point2, point;
@@ -92,10 +121,8 @@ static void doFillOneRow(SFillInfo* pFillInfo, SSDataBlock* pBlock, SSDataBlock*
       SFillColInfo* pCol = &pFillInfo->pFillCol[i];
 
       SColumnInfoData* pDstColInfoData = taosArrayGet(pBlock->pDataBlock, GET_DEST_SLOT_ID(pCol));
-
-      if (pDstColInfoData->info.type == TSDB_DATA_TYPE_TIMESTAMP) {
-        colDataAppend(pDstColInfoData, index, (const char*)&pFillInfo->currentKey, false);
-      } else {
+      bool filled = fillIfWindowPseudoColumn(pFillInfo, pCol, pDstColInfoData, index);
+      if (!filled) {
         SGroupKeys* pKey = taosArrayGet(p, i);
         doSetVal(pDstColInfoData, index, pKey);
       }
@@ -106,10 +133,8 @@ static void doFillOneRow(SFillInfo* pFillInfo, SSDataBlock* pBlock, SSDataBlock*
     for (int32_t i = 0; i < pFillInfo->numOfCols; ++i) {
       SFillColInfo* pCol = &pFillInfo->pFillCol[i];
       SColumnInfoData* pDstColInfoData = taosArrayGet(pBlock->pDataBlock, GET_DEST_SLOT_ID(pCol));
-
-      if (pDstColInfoData->info.type == TSDB_DATA_TYPE_TIMESTAMP) {
-        colDataAppend(pDstColInfoData, index, (const char*)&pFillInfo->currentKey, false);
-      } else {
+      bool filled = fillIfWindowPseudoColumn(pFillInfo, pCol, pDstColInfoData, index);
+      if (!filled) {
         SGroupKeys* pKey = taosArrayGet(p, i);
         doSetVal(pDstColInfoData, index, pKey);
       }
@@ -127,9 +152,8 @@ static void doFillOneRow(SFillInfo* pFillInfo, SSDataBlock* pBlock, SSDataBlock*
         int16_t          type = pDstCol->info.type;
 
         if (pCol->notFillCol) {
-          if (type == TSDB_DATA_TYPE_TIMESTAMP) {
-            colDataAppend(pDstCol, index, (const char*)&pFillInfo->currentKey, false);
-          } else {
+          bool filled = fillIfWindowPseudoColumn(pFillInfo, pCol, pDstCol, index);
+          if (!filled) {
             SArray* p = FILL_IS_ASC_FILL(pFillInfo) ? pFillInfo->prev.pRowVal : pFillInfo->next.pRowVal;
             SGroupKeys* pKey = taosArrayGet(p, i);
             doSetVal(pDstCol, index, pKey);
@@ -170,9 +194,8 @@ static void doFillOneRow(SFillInfo* pFillInfo, SSDataBlock* pBlock, SSDataBlock*
       SColumnInfoData* pDst = taosArrayGet(pBlock->pDataBlock, slotId);
 
       if (pCol->notFillCol) {
-        if (pDst->info.type == TSDB_DATA_TYPE_TIMESTAMP) {
-          colDataAppend(pDst, index, (const char*)&pFillInfo->currentKey, false);
-        } else {
+        bool filled = fillIfWindowPseudoColumn(pFillInfo, pCol, pDst, index);
+        if (!filled) {
           SArray* p = FILL_IS_ASC_FILL(pFillInfo) ? pFillInfo->prev.pRowVal : pFillInfo->next.pRowVal;
           SGroupKeys* pKey = taosArrayGet(p, i);
           doSetVal(pDst, index, pKey);

source/libs/index/src/indexFilter.c

@@ -255,6 +255,13 @@ static int32_t sifInitOperParams(SIFParam **params, SOperatorNode *node, SIFCtx
   if (node->opType == OP_TYPE_JSON_GET_VALUE) {
     return code;
   }
+  if ((node->pLeft != NULL && nodeType(node->pLeft) == QUERY_NODE_COLUMN) &&
+      (node->pRight != NULL && nodeType(node->pRight) == QUERY_NODE_VALUE)) {
+    SColumnNode *cn = (SColumnNode *)(node->pLeft);
+    if (cn->node.resType.type == TSDB_DATA_TYPE_JSON) {
+      SIF_ERR_RET(TSDB_CODE_QRY_INVALID_INPUT);
+    }
+  }
   SIFParam *paramList = taosMemoryCalloc(nParam, sizeof(SIFParam));
 
   if (NULL == paramList) {

source/libs/planner/src/planLogicCreater.c

@@ -44,12 +44,15 @@ static void setColumnInfo(SFunctionNode* pFunc, SColumnNode* pCol) {
       pCol->colType = COLUMN_TYPE_TBNAME;
       break;
     case FUNCTION_TYPE_WSTART:
+      pCol->colId = PRIMARYKEY_TIMESTAMP_COL_ID;
+      pCol->colType = COLUMN_TYPE_WINDOW_START;
+      break;
     case FUNCTION_TYPE_WEND:
       pCol->colId = PRIMARYKEY_TIMESTAMP_COL_ID;
-      pCol->colType = COLUMN_TYPE_WINDOW_PC;
+      pCol->colType = COLUMN_TYPE_WINDOW_END;
       break;
     case FUNCTION_TYPE_WDURATION:
-      pCol->colType = COLUMN_TYPE_WINDOW_PC;
+      pCol->colType = COLUMN_TYPE_WINDOW_DURATION;
       break;
     case FUNCTION_TYPE_GROUP_KEY:
       pCol->colType = COLUMN_TYPE_GROUP_KEY;
@@ -784,7 +787,10 @@ static int32_t createWindowLogicNode(SLogicPlanContext* pCxt, SSelectStmt* pSele
 static EDealRes needFillValueImpl(SNode* pNode, void* pContext) {
   if (QUERY_NODE_COLUMN == nodeType(pNode)) {
     SColumnNode* pCol = (SColumnNode*)pNode;
-    if (COLUMN_TYPE_WINDOW_PC != pCol->colType && COLUMN_TYPE_GROUP_KEY != pCol->colType) {
+    if (COLUMN_TYPE_WINDOW_START != pCol->colType &&
+        COLUMN_TYPE_WINDOW_END != pCol->colType &&
+        COLUMN_TYPE_WINDOW_DURATION != pCol->colType &&
+        COLUMN_TYPE_GROUP_KEY != pCol->colType) {
       *(bool*)pContext = true;
       return DEAL_RES_END;
     }

source/libs/scheduler/inc/schInt.h

@@ -283,7 +283,7 @@ typedef struct SSchJob {
 } SSchJob;
 
 typedef struct SSchTaskCtx {
-  SSchJob  *pJob;
+  int64_t  jobRid;
   SSchTask *pTask;
 } SSchTaskCtx;
 

source/libs/scheduler/src/schTask.c

@@ -821,7 +821,13 @@ int32_t schProcessOnTaskStatusRsp(SQueryNodeEpId *pEpId, SArray *pStatusList) {
 
 int32_t schLaunchTaskImpl(void *param) {
   SSchTaskCtx *pCtx = (SSchTaskCtx *)param;
-  SSchJob *pJob = pCtx->pJob;
+  SSchJob *pJob = schAcquireJob(pCtx->jobRid);
+  if (NULL == pJob) {
+    taosMemoryFree(param);
+    qDebug("job refId 0x%" PRIx64 " already not exist", pCtx->jobRid);
+    SCH_RET(TSDB_CODE_SCH_JOB_IS_DROPPING);
+  }
+  
   SSchTask *pTask = pCtx->pTask;
   int8_t  status = 0;
   int32_t code = 0;
@@ -880,6 +886,8 @@ _return:
     }
   }
 
+  schReleaseJob(pJob->refId);
+
   SCH_RET(code);
 }
 
@@ -890,7 +898,7 @@ int32_t schAsyncLaunchTaskImpl(SSchJob *pJob, SSchTask *pTask) {
     SCH_ERR_RET(TSDB_CODE_OUT_OF_MEMORY);
   }
   
-  param->pJob = pJob;
+  param->jobRid = pJob->refId;
   param->pTask = pTask;
 
   if (pJob->taskNum >= SCH_MIN_AYSNC_EXEC_NUM) {

source/libs/stream/src/streamDispatch.c

@@ -358,7 +358,7 @@ int32_t streamDispatchAllBlocks(SStreamTask* pTask, const SStreamDataBlock* pDat
   FAIL_SHUFFLE_DISPATCH:
     if (pReqs) {
       for (int32_t i = 0; i < vgSz; i++) {
-        taosArrayDestroy(pReqs[i].data);
+        taosArrayDestroyP(pReqs[i].data, taosMemoryFree);
         taosArrayDestroy(pReqs[i].dataLen);
       }
       taosMemoryFree(pReqs);

source/libs/stream/src/streamExec.c

@@ -140,7 +140,6 @@ int32_t streamPipelineExec(SStreamTask* pTask, int32_t batchNum, bool dispatch)
   return 0;
 }
 
-// TODO: handle version
 int32_t streamExecForAll(SStreamTask* pTask) {
   while (1) {
     int32_t batchCnt = 1;

source/libs/stream/src/streamMeta.c

@@ -14,7 +14,7 @@
  */
 
 #include "executor.h"
-#include "tstream.h"
+#include "streamInc.h"
 #include "ttimer.h"
 
 SStreamMeta* streamMetaOpen(const char* path, void* ahandle, FTaskExpand expandFunc) {
@@ -23,17 +23,23 @@ SStreamMeta* streamMetaOpen(const char* path, void* ahandle, FTaskExpand expandF
     terrno = TSDB_CODE_OUT_OF_MEMORY;
     return NULL;
   }
-  pMeta->path = strdup(path);
+  int32_t len = strlen(path) + 20;
+  char*   streamPath = taosMemoryCalloc(1, len);
+  sprintf(streamPath, "%s/%s", path, "stream");
+  pMeta->path = strdup(streamPath);
   if (tdbOpen(pMeta->path, 16 * 1024, 1, &pMeta->db) < 0) {
     goto _err;
   }
 
+  sprintf(streamPath, "%s/%s", pMeta->path, "checkpoints");
+  mkdir(streamPath, 0755);
+  taosMemoryFree(streamPath);
+
   if (tdbTbOpen("task.db", sizeof(int32_t), -1, NULL, pMeta->db, &pMeta->pTaskDb) < 0) {
     goto _err;
   }
 
-  // open state storage backend
-  if (tdbTbOpen("state.db", sizeof(int32_t), -1, NULL, pMeta->db, &pMeta->pStateDb) < 0) {
+  if (tdbTbOpen("checkpoint.db", sizeof(int32_t), -1, NULL, pMeta->db, &pMeta->pCheckpointDb) < 0) {
     goto _err;
   }
 
@@ -49,16 +55,13 @@ SStreamMeta* streamMetaOpen(const char* path, void* ahandle, FTaskExpand expandF
   pMeta->ahandle = ahandle;
   pMeta->expandFunc = expandFunc;
 
-  if (streamLoadTasks(pMeta) < 0) {
-    goto _err;
-  }
   return pMeta;
 
 _err:
   if (pMeta->path) taosMemoryFree(pMeta->path);
   if (pMeta->pTasks) taosHashCleanup(pMeta->pTasks);
-  if (pMeta->pStateDb) tdbTbClose(pMeta->pStateDb);
   if (pMeta->pTaskDb) tdbTbClose(pMeta->pTaskDb);
+  if (pMeta->pCheckpointDb) tdbTbClose(pMeta->pCheckpointDb);
   if (pMeta->db) tdbClose(pMeta->db);
   taosMemoryFree(pMeta);
   return NULL;
@@ -67,7 +70,7 @@ _err:
 void streamMetaClose(SStreamMeta* pMeta) {
   tdbCommit(pMeta->db, &pMeta->txn);
   tdbTbClose(pMeta->pTaskDb);
-  tdbTbClose(pMeta->pStateDb);
+  tdbTbClose(pMeta->pCheckpointDb);
   tdbClose(pMeta->db);
 
   void* pIter = NULL;

source/libs/stream/src/streamRecover.c

@@ -176,6 +176,7 @@ int32_t tDecodeSStreamTaskRecoverRsp(SDecoder* pDecoder, SStreamRecoverDownstrea
 }
 
 int32_t streamSaveStateInfo(SStreamMeta* pMeta, SStreamTask* pTask) {
+#if 0
   void* buf = NULL;
 
   ASSERT(pTask->taskLevel == TASK_LEVEL__SINK);
@@ -224,10 +225,12 @@ int32_t streamSaveStateInfo(SStreamMeta* pMeta, SStreamTask* pTask) {
 FAIL:
   if (buf) taosMemoryFree(buf);
   return -1;
+#endif
   return 0;
 }
 
 int32_t streamLoadStateInfo(SStreamMeta* pMeta, SStreamTask* pTask) {
+#if 0
   void*   pVal = NULL;
   int32_t vLen = 0;
   if (tdbTbGet(pMeta->pStateDb, &pTask->taskId, sizeof(void*), &pVal, &vLen) < 0) {
@@ -241,7 +244,7 @@ int32_t streamLoadStateInfo(SStreamMeta* pMeta, SStreamTask* pTask) {
 
   pTask->nextCheckId = aggCheckpoint.checkpointId + 1;
   pTask->checkpointInfo = aggCheckpoint.checkpointVer;
-
+#endif
   return 0;
 }
 

source/libs/stream/src/streamState.c

+/*
+ * Copyright (c) 2019 TAOS Data, Inc. <jhtao@taosdata.com>
+ *
+ * This program is free software: you can use, redistribute, and/or modify
+ * it under the terms of the GNU Affero General Public License, version 3
+ * or later ("AGPL"), as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * You should have received a copy of the GNU Affero General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "executor.h"
+#include "streamInc.h"
+#include "tcommon.h"
+#include "ttimer.h"
+
+SStreamState* streamStateOpen(char* path, SStreamTask* pTask) {
+  SStreamState* pState = taosMemoryCalloc(1, sizeof(SStreamState));
+  if (pState == NULL) {
+    terrno = TSDB_CODE_OUT_OF_MEMORY;
+    return NULL;
+  }
+  char statePath[300];
+  sprintf(statePath, "%s/%d", path, pTask->taskId);
+  if (tdbOpen(statePath, 4096, 256, &pState->db) < 0) {
+    goto _err;
+  }
+
+  // open state storage backend
+  if (tdbTbOpen("state.db", sizeof(SWinKey), -1, SWinKeyCmpr, pState->db, &pState->pStateDb) < 0) {
+    goto _err;
+  }
+
+  if (streamStateBegin(pState) < 0) {
+    goto _err;
+  }
+
+  pState->pOwner = pTask;
+
+  return pState;
+
+_err:
+  if (pState->pStateDb) tdbTbClose(pState->pStateDb);
+  if (pState->db) tdbClose(pState->db);
+  taosMemoryFree(pState);
+  return NULL;
+}
+
+void streamStateClose(SStreamState* pState) {
+  tdbCommit(pState->db, &pState->txn);
+  tdbTbClose(pState->pStateDb);
+  tdbClose(pState->db);
+
+  taosMemoryFree(pState);
+}
+
+int32_t streamStateBegin(SStreamState* pState) {
+  if (tdbTxnOpen(&pState->txn, 0, tdbDefaultMalloc, tdbDefaultFree, NULL, TDB_TXN_WRITE | TDB_TXN_READ_UNCOMMITTED) <
+      0) {
+    return -1;
+  }
+
+  if (tdbBegin(pState->db, &pState->txn) < 0) {
+    tdbTxnClose(&pState->txn);
+    return -1;
+  }
+  return 0;
+}
+
+int32_t streamStateCommit(SStreamState* pState) {
+  if (tdbCommit(pState->db, &pState->txn) < 0) {
+    return -1;
+  }
+  memset(&pState->txn, 0, sizeof(TXN));
+  if (tdbTxnOpen(&pState->txn, 0, tdbDefaultMalloc, tdbDefaultFree, NULL, TDB_TXN_WRITE | TDB_TXN_READ_UNCOMMITTED) <
+      0) {
+    return -1;
+  }
+  if (tdbBegin(pState->db, &pState->txn) < 0) {
+    return -1;
+  }
+  return 0;
+}
+
+int32_t streamStateAbort(SStreamState* pState) {
+  if (tdbAbort(pState->db, &pState->txn) < 0) {
+    return -1;
+  }
+  memset(&pState->txn, 0, sizeof(TXN));
+  if (tdbTxnOpen(&pState->txn, 0, tdbDefaultMalloc, tdbDefaultFree, NULL, TDB_TXN_WRITE | TDB_TXN_READ_UNCOMMITTED) <
+      0) {
+    return -1;
+  }
+  if (tdbBegin(pState->db, &pState->txn) < 0) {
+    return -1;
+  }
+  return 0;
+}
+
+int32_t streamStatePut(SStreamState* pState, const SWinKey* key, const void* value, int32_t vLen) {
+  return tdbTbUpsert(pState->pStateDb, key, sizeof(SWinKey), value, vLen, &pState->txn);
+}
+int32_t streamStateGet(SStreamState* pState, const SWinKey* key, void** pVal, int32_t* pVLen) {
+  return tdbTbGet(pState->pStateDb, key, sizeof(SWinKey), pVal, pVLen);
+}
+
+int32_t streamStateDel(SStreamState* pState, const SWinKey* key) {
+  return tdbTbDelete(pState->pStateDb, key, sizeof(SWinKey), &pState->txn);
+}
+
+SStreamStateCur* streamStateGetCur(SStreamState* pState, const SWinKey* key) {
+  SStreamStateCur* pCur = taosMemoryCalloc(1, sizeof(SStreamStateCur));
+  if (pCur == NULL) return NULL;
+  tdbTbcOpen(pState->pStateDb, &pCur->pCur, NULL);
+
+  int32_t c;
+  tdbTbcMoveTo(pCur->pCur, key, sizeof(SWinKey), &c);
+  if (c != 0) {
+    taosMemoryFree(pCur);
+    return NULL;
+  }
+  return pCur;
+}
+
+int32_t streamStateGetKVByCur(SStreamStateCur* pCur, SWinKey* pKey, const void** pVal, int32_t* pVLen) {
+  const SWinKey* pKTmp = NULL;
+  int32_t        kLen;
+  if (tdbTbcGet(pCur->pCur, (const void**)&pKTmp, &kLen, pVal, pVLen) < 0) {
+    return -1;
+  }
+  *pKey = *pKTmp;
+  return 0;
+}
+
+int32_t streamStateSeekFirst(SStreamState* pState, SStreamStateCur* pCur) {
+  //
+  return tdbTbcMoveToFirst(pCur->pCur);
+}
+
+int32_t streamStateSeekLast(SStreamState* pState, SStreamStateCur* pCur) {
+  //
+  return tdbTbcMoveToLast(pCur->pCur);
+}
+
+SStreamStateCur* streamStateSeekKeyNext(SStreamState* pState, const SWinKey* key) {
+  SStreamStateCur* pCur = taosMemoryCalloc(1, sizeof(SStreamStateCur));
+  if (pCur == NULL) {
+    return NULL;
+  }
+
+  int32_t c;
+  if (tdbTbcMoveTo(pCur->pCur, key, sizeof(SWinKey), &c) < 0) {
+    taosMemoryFree(pCur);
+    return NULL;
+  }
+  if (c > 0) return pCur;
+
+  if (tdbTbcMoveToNext(pCur->pCur) < 0) {
+    taosMemoryFree(pCur);
+    return NULL;
+  }
+
+  return pCur;
+}
+
+SStreamStateCur* streamStateSeekKeyPrev(SStreamState* pState, const SWinKey* key) {
+  SStreamStateCur* pCur = taosMemoryCalloc(1, sizeof(SStreamStateCur));
+  if (pCur == NULL) {
+    return NULL;
+  }
+
+  int32_t c;
+  if (tdbTbcMoveTo(pCur->pCur, key, sizeof(SWinKey), &c) < 0) {
+    taosMemoryFree(pCur);
+    return NULL;
+  }
+  if (c < 0) return pCur;
+
+  if (tdbTbcMoveToPrev(pCur->pCur) < 0) {
+    taosMemoryFree(pCur);
+    return NULL;
+  }
+
+  return pCur;
+}
+
+int32_t streamStateCurNext(SStreamState* pState, SStreamStateCur* pCur) {
+  //
+  return tdbTbcMoveToNext(pCur->pCur);
+}
+
+int32_t streamStateCurPrev(SStreamState* pState, SStreamStateCur* pCur) {
+  //
+  return tdbTbcMoveToPrev(pCur->pCur);
+}
+void streamStateFreeCur(SStreamStateCur* pCur) {
+  tdbTbcClose(pCur->pCur);
+  taosMemoryFree(pCur);
+}
+
+void streamFreeVal(void* val) { tdbFree(val); }

source/libs/stream/src/streamTask.c

@@ -165,5 +165,8 @@ void tFreeSStreamTask(SStreamTask* pTask) {
   if (pTask->outputType == TASK_OUTPUT__SHUFFLE_DISPATCH) {
     taosArrayDestroy(pTask->shuffleDispatcher.dbInfo.pVgroupInfos);
   }
+
+  if (pTask->pState) streamStateClose(pTask->pState);
+
   taosMemoryFree(pTask);
 }

source/libs/tdb/src/db/tdbBtree.c

@@ -1401,7 +1401,7 @@ static int tdbBtreeDecodeCell(SPage *pPage, const SCell *pCell, SCellDecoder *pD
   pDecoder->pgno = 0;
   TDB_CELLDECODER_SET_FREE_NIL(pDecoder);
 
-  tdbDebug("tdb btc decoder set nil: %p/0x%x ", pDecoder, pDecoder->freeKV);
+  // tdbTrace("tdb btc decoder set nil: %p/0x%x ", pDecoder, pDecoder->freeKV);
 
   // 1. Decode header part
   if (!leaf) {

source/libs/tdb/src/db/tdbPCache.c

@@ -145,7 +145,7 @@ static SPage *tdbPCacheFetchImpl(SPCache *pCache, const SPgid *pPgid, TXN *pTxn)
   // 1. Search the hash table
   pPage = pCache->pgHash[tdbPCachePageHash(pPgid) % pCache->nHash];
   while (pPage) {
-    if (memcmp(pPage->pgid.fileid, pPgid->fileid, TDB_FILE_ID_LEN) == 0 && pPage->pgid.pgno == pPgid->pgno) break;
+    if (pPage->pgid.pgno == pPgid->pgno && memcmp(pPage->pgid.fileid, pPgid->fileid, TDB_FILE_ID_LEN) == 0) break;
     pPage = pPage->pHashNext;
   }
 
@@ -243,7 +243,7 @@ static void tdbPCachePinPage(SPCache *pCache, SPage *pPage) {
     pCache->nRecyclable--;
 
     // printf("pin page %d pgno %d pPage %p\n", pPage->id, TDB_PAGE_PGNO(pPage), pPage);
-    tdbTrace("pin page %d", pPage->id);
+    tdbDebug("pin page %d", pPage->id);
   }
 }
 
@@ -264,30 +264,33 @@ static void tdbPCacheUnpinPage(SPCache *pCache, SPage *pPage) {
   pCache->nRecyclable++;
 
   // printf("unpin page %d pgno %d pPage %p\n", pPage->id, TDB_PAGE_PGNO(pPage), pPage);
-  tdbTrace("unpin page %d", pPage->id);
+  tdbDebug("unpin page %d", pPage->id);
 }
 
 static void tdbPCacheRemovePageFromHash(SPCache *pCache, SPage *pPage) {
-  SPage  **ppPage;
-  uint32_t h;
-
-  h = tdbPCachePageHash(&(pPage->pgid));
-  for (ppPage = &(pCache->pgHash[h % pCache->nHash]); (*ppPage) && *ppPage != pPage; ppPage = &((*ppPage)->pHashNext))
-    ;
-
+  uint32_t h = tdbPCachePageHash(&(pPage->pgid)) % pCache->nHash;
+
+  SPage **ppPage = &(pCache->pgHash[h]);
+  if (*ppPage == pPage) {
+    pCache->pgHash[h] = pPage->pHashNext;
+  } else {
+    for (; (*ppPage) && (*ppPage)->pHashNext != pPage; ppPage = &((*ppPage)->pHashNext))
+      ;
+    if (*ppPage) {
+      (*ppPage)->pHashNext = pPage->pHashNext;
+    }
+  }
   if (*ppPage) {
-    *ppPage = pPage->pHashNext;
-    pCache->nPage--;
+    pPage->pHashNext = NULL;
+    --pCache->nPage;
     // printf("rmv page %d to hash, pgno %d, pPage %p\n", pPage->id, TDB_PAGE_PGNO(pPage), pPage);
   }
 
-  tdbTrace("remove page %d to hash", pPage->id);
+  tdbDebug("remove page %p/%d from hash", pPage, pPage->id);
 }
 
 static void tdbPCacheAddPageToHash(SPCache *pCache, SPage *pPage) {
-  int h;
-
-  h = tdbPCachePageHash(&(pPage->pgid)) % pCache->nHash;
+  uint32_t h = tdbPCachePageHash(&(pPage->pgid)) % pCache->nHash;
 
   pPage->pHashNext = pCache->pgHash[h];
   pCache->pgHash[h] = pPage;
@@ -295,7 +298,7 @@ static void tdbPCacheAddPageToHash(SPCache *pCache, SPage *pPage) {
   pCache->nPage++;
 
   // printf("add page %d to hash, pgno %d, pPage %p\n", pPage->id, TDB_PAGE_PGNO(pPage), pPage);
-  tdbTrace("add page %d to hash", pPage->id);
+  tdbDebug("add page %p/%d to hash", pPage, pPage->id);
 }
 
 static int tdbPCacheOpenImpl(SPCache *pCache) {

source/libs/tdb/src/db/tdbPage.c

@@ -74,6 +74,7 @@ int tdbPageCreate(int pageSize, SPage **ppPage, void *(*xMalloc)(void *, size_t)
 int tdbPageDestroy(SPage *pPage, void (*xFree)(void *arg, void *ptr), void *arg) {
   u8 *ptr;
 
+  tdbDebug("page/destroy: %p %p", pPage, xFree);
   ASSERT(xFree);
 
   for (int iOvfl = 0; iOvfl < pPage->nOverflow; iOvfl++) {
@@ -87,6 +88,7 @@ int tdbPageDestroy(SPage *pPage, void (*xFree)(void *arg, void *ptr), void *arg)
 }
 
 void tdbPageZero(SPage *pPage, u8 szAmHdr, int (*xCellSize)(const SPage *, SCell *, int, TXN *, SBTree *pBt)) {
+  tdbDebug("page/zero: %p %" PRIu8 " %p", pPage, szAmHdr, xCellSize);
   pPage->pPageHdr = pPage->pData + szAmHdr;
   TDB_PAGE_NCELLS_SET(pPage, 0);
   TDB_PAGE_CCELLS_SET(pPage, pPage->pageSize - sizeof(SPageFtr));
@@ -103,6 +105,7 @@ void tdbPageZero(SPage *pPage, u8 szAmHdr, int (*xCellSize)(const SPage *, SCell
 }
 
 void tdbPageInit(SPage *pPage, u8 szAmHdr, int (*xCellSize)(const SPage *, SCell *, int, TXN *, SBTree *pBt)) {
+  tdbDebug("page/init: %p %" PRIu8 " %p", pPage, szAmHdr, xCellSize);
   pPage->pPageHdr = pPage->pData + szAmHdr;
   pPage->pCellIdx = pPage->pPageHdr + TDB_PAGE_HDR_SIZE(pPage);
   pPage->pFreeStart = pPage->pCellIdx + TDB_PAGE_OFFSET_SIZE(pPage) * TDB_PAGE_NCELLS(pPage);

source/libs/transport/src/thttp.c

@@ -23,6 +23,7 @@
 
 
 #define HTTP_RECV_BUF_SIZE 1024
+
 typedef struct SHttpClient {
   uv_connect_t conn;
   uv_tcp_t     tcp;
@@ -143,9 +144,9 @@ static void clientAllocBuffCb(uv_handle_t *handle, size_t suggested_size, uv_buf
 static void clientRecvCb(uv_stream_t* handle, ssize_t nread, const uv_buf_t *buf) {
   SHttpClient* cli = handle->data; 
   if (nread < 0) {
-    uError("http-report read error:%s", uv_err_name(nread));
+    uError("http-report recv error:%s", uv_err_name(nread));
   } else {
-    uTrace("http-report succ to read %d bytes, just ignore it", nread);
+    uTrace("http-report succ to recv %d bytes, just ignore it", nread);
   }
   uv_close((uv_handle_t*)&cli->tcp, clientCloseCb);
 } 

source/libs/transport/src/transSvr.c

@@ -276,14 +276,16 @@ void uvOnRecvCb(uv_stream_t* cli, ssize_t nread, const uv_buf_t* buf) {
       while (transReadComplete(pBuf)) {
         tTrace("%s conn %p alread read complete packet", transLabel(pTransInst), conn);
         if (true == pBuf->invalid || false == uvHandleReq(conn)) {
-          tError("%s conn %p read invalid packet", transLabel(pTransInst), conn);
+          tError("%s conn %p read invalid packet, received from %s, local info:%s", transLabel(pTransInst), conn,
+                 conn->dst, conn->src);
           destroyConn(conn, true);
           return;
         }
       }
       return;
     } else {
-      tError("%s conn %p read invalid packet, exceed limit", transLabel(pTransInst), conn);
+      tError("%s conn %p read invalid packet, exceed limit, received from %s, local info:", transLabel(pTransInst),
+             conn, conn->dst, conn->src);
       destroyConn(conn, true);
       return;
     }
@@ -649,7 +651,7 @@ void uvOnAcceptCb(uv_stream_t* stream, int status) {
 
     pObj->workerIdx = (pObj->workerIdx + 1) % pObj->numOfThreads;
 
-    tTrace("new conntion accepted by main server, dispatch to %dth worker-thread", pObj->workerIdx);
+    tTrace("new connection accepted by main server, dispatch to %dth worker-thread", pObj->workerIdx);
 
     uv_write2(wr, (uv_stream_t*)&(pObj->pipe[pObj->workerIdx][0]), &buf, 1, (uv_stream_t*)cli, uvOnPipeWriteCb);
   } else {

source/libs/wal/src/walMeta.c

@@ -121,7 +121,7 @@ static FORCE_INLINE int64_t walScanLogGetLastVer(SWal* pWal) {
   if (found == NULL) {
     // file corrupted, no complete log
     // TODO delete and search in previous files
-    ASSERT(0);
+    /*ASSERT(0);*/
     terrno = TSDB_CODE_WAL_FILE_CORRUPTED;
     return -1;
   }
@@ -221,7 +221,6 @@ int walCheckAndRepairMeta(SWal* pWal) {
 
       int code = walSaveMeta(pWal);
       if (code < 0) {
-        taosArrayDestroy(actualLog);
         return -1;
       }
     }

source/os/src/osFile.c

@@ -203,10 +203,11 @@ int32_t taosRenameFile(const char *oldName, const char *newName) {
 }
 
 int32_t taosStatFile(const char *path, int64_t *size, int32_t *mtime) {
-  struct stat fileStat;
 #ifdef WINDOWS
-  int32_t code = _stat(path, &fileStat);
+  struct _stati64 fileStat;
+  int32_t code = _stati64(path, &fileStat);
 #else
+  struct stat fileStat;
   int32_t code = stat(path, &fileStat);
 #endif
   if (code < 0) {

source/os/src/osSysinfo.c

@@ -851,13 +851,12 @@ char *taosGetCmdlineByPID(int pid) {
 }
 
 void taosSetCoreDump(bool enable) {
+  if (!enable) return;
 #ifdef WINDOWS
-  // SetUnhandledExceptionFilter(exceptionHandler);
-  // SetUnhandledExceptionFilter(&FlCrashDump);
+  SetUnhandledExceptionFilter(exceptionHandler);
+  SetUnhandledExceptionFilter(&FlCrashDump);
 #elif defined(_TD_DARWIN_64)
 #else
-  if (!enable) return;
-
   // 1. set ulimit -c unlimited
   struct rlimit rlim;
   struct rlimit rlim_new;

tests/script/tsim/db/basic2.sim

@@ -4,7 +4,7 @@ system sh/exec.sh -n dnode1 -s start
 sql connect
 
 print =============== conflict stb
-sql create database db vgroups 1;
+sql create database db vgroups 4;
 sql use db;
 sql create table stb (ts timestamp, i int) tags (j int);
 sql_error create table stb using stb tags (1);
@@ -16,6 +16,9 @@ sql_error create table ctb (ts timestamp, i int) tags (j int);
 sql create table ntb (ts timestamp, i int);
 sql_error create table ntb (ts timestamp, i int) tags (j int);
 
+sql drop table ntb
+sql create table ntb (ts timestamp, i int) tags (j int);
+
 sql drop database db
 
 print =============== create database d1