WO2022056735A1

WO2022056735A1 - Cloud high-performance scientific calculation workflow design control system and graphical user interface

Info

Publication number: WO2022056735A1
Application number: PCT/CN2020/115613
Authority: WO
Inventors: 谈樑; 刘阳; 鄂同富; 姜子麒; 马健; 温书豪; 赖力鹏
Original assignee: 深圳晶泰科技有限公司
Priority date: 2020-09-16
Filing date: 2020-09-16
Publication date: 2022-03-24

Abstract

The present invention provides a cloud high-performance scientific calculation workflow design control system and a graphical user interface. The control system comprises three layers, wherein an underlying service layer provides, in the form of an SDK, a basic service for an upper-layer application; a parsing layer is an actual architectural support, and provides machine language conversion for an actual operation performed by a user on an interface, so that the system and an underlying service can operate according to an instruction of the user, and at the same time, data returned by the underlying service is converted according to a form designed by the system, such that the data conforms to an interface operation habit and display logic of the user; and a human-computer interaction layer is responsible for realizing operation behavior and parameter configuration functions for the user. The graphical user interface comprises a workflow management module, a workflow orchestration module, a workflow node control module and a task management module. According to the present invention, by means of cloud native software, human-computer interaction functions, such as dragging and configuration, are performed at a user terminal, so as to complete control over the design, deployment and execution of a cloud high-performance scientific calculation workflow.

Description

Cloud high-performance scientific computing workflow design control system and user graphical interface

technical field

The invention belongs to the technical field of high-performance computing and visualization, and particularly relates to a cloud-based high-performance scientific computing workflow design control system and a user graphical interface.

Background technique

Although the standardized workflow language provides scientific computing process developers with a standardized method of invoking cloud resources to perform tasks, it satisfies the needs of scientific computing process definition and design to complete high-throughput computing. However, in the process of designing the scientific computing process, the R&D personnel still need to master the writing of the language, understand the working mechanism of the cloud middleware and the abstract model of the complete process, especially in the scientific computing workflow R&D stage, the R&D personnel often constantly improve the computing nodes Algorithmic functions, adjustment and replacement of nodes in the workflow, if there is no interactive graphical interface, the research and development efficiency will be greatly reduced. At present, the main way to realize this process is mainly to write description files by engineers who have the foundation of workflow language writing and then transfer them to the workflow engine. The interactive graphical interface can effectively speed up this process and help R&D personnel to master the workflow. Panorama.

Mature scientific computing workflow software is deployed in the cloud. Because it is integrated by the software developer and the cloud manufacturer, the whole set of usage process is consistent with the functions of the local software. Only the process of deployment and installation in the cloud is free of installation and use in the browser. The usage time is charged. This method does not fully utilize the core advantages of scheduling cloud computing services, such as elastic scaling of computing resources, which can not only improve computing efficiency but also improve cost performance. The reason is that each algorithm module or computing analysis node is not a cloud-native architecture, but an architecture integrated with the cloud software. The computing resources use the software framework to schedule the computer where it is located. Although service function computing has many data statistics and analysis tools, it is at a disadvantage in large-scale parallel computing under high-throughput requirements.

In the actual scientific computing process design and development scenarios, considering the convenience and feasibility of cloud-native workflow construction, the applicant has adopted containerized packaging of mature software, and has made it consistent with the algorithms developed by researchers. The computer representation of the actual scientific computing workflow is thus actually available in the cloud. However, in the actual process of workflow execution, a workflow corresponds to a task instance, and it is difficult for users to judge the operation of each node through the status data returned by the task, and the total amount of data returned by the node is large, which makes the command It is difficult to find and identify in the running environment. In addition, users often need to dynamically control the task, or perform conditional manual intervention for a certain node, so as to achieve controllable task process, save computing resources, and visualize the dynamics of interactive operations. balance.

SUMMARY OF THE INVENTION

In view of the above technical problems, the present invention provides a cloud high-performance scientific computing workflow design control system and a user graphical interface, which can be used for cloud-based high-performance scientific computing by dragging and configuring human-computer interaction functions on a user terminal through cloud native software. Workflow design, deployment and execution control.

The present invention allows users to deploy the algorithm written in a containerized manner to the cloud through the interactive mode of real-time visual operation, design, arrange and combine different algorithms, and create and publish a set of workflows that can call mainstream public cloud computing resources. Collaborate and reuse with authorized users, conveniently test and execute computing tasks, and enable instant dynamic control through the interface during computing. .

The specific technical solutions are:

The cloud high-performance scientific computing workflow design control system includes three layers: the underlying service layer, the analysis layer and the human-computer interaction layer;

The underlying service layer: the underlying service provides basic services for upper-layer applications in the form of SDK, including: task scheduling and execution middleware, computing and storage resource management services, task monitoring and log services;

The main function of the task scheduling and execution middleware is to enable the cloud computing cluster to package and run tasks through packaged instructions;

The computing and storage resource service, the main function of which provides resource uploading, downloading, storage and distribution for the workflow and actual running tasks of the client;

The main function of the task monitoring and log service is to collect and store data such as status and error throws returned during the task running process, and to query through commands.

The parsing layer: the parsing layer is the actual support of the architecture, providing machine language conversion for the actual operation of the user on the interface, so that the system and the underlying service can operate according to the user's instructions, and at the same time, the data returned by the underlying service is in accordance with the The designed form of the system is converted to make the data conform to the user's interface operation habits and display logic; the parsing layer includes three modules: workflow description language interpretation parser, workflow generator, and task dispatcher;

The main function of the workflow description language interpretation parser is to apply a standardized workflow language to describe and convert cloud-native containerized software or algorithms. The node that can be dragged and controlled on the graphical interface, as well as the properties that the node needs to display, etc. On the underlying server, the user's operation is converted into the data structure required by the underlying SDK;

The main function of the workflow generator is to process and convert the workflow nodes and related configurations that users drag and drop to combine into workflow files, so that the task distributor can identify the workflow sequence relationship, distribution mode and parameter configuration. ;

The main function of the task dispatcher is to call the underlying service according to the configuration file generated by the workflow generator, and finally realize the analysis of the workflow, the execution and distribution of the task.

The human-computer interaction layer: the human-computer interaction layer is responsible for realizing the functions of operation behavior and parameter configuration for the user, including four modules: a workflow management module, a work node control panel, a workflow arrangement module, and a task manager;

The main function of the workflow management module is to download and load containers and workflow description files in the cloud, and manage, modify, or introduce workflows in the form of files on the client side;

The main function of the work node control panel is to allow the user to configure the work node on the client side, and to configure the execution properties and operation parameters of the node; the operation mode of the node is: single-point mode or parallel mode.

The main function of the workflow orchestration module is to allow users to introduce different workflow nodes and establish connection relationships between different workflow nodes;

The main function of the task manager is to be able to view the running status, input and output and task logs of each node's task or each sub-task in a concurrent task in real time.

The present invention also provides a user graphical interface for the above-mentioned cloud high-performance scientific computing workflow design control system, including a workflow management module, a workflow arrangement module, a workflow node control module, and a task management module;

The workflow management module: the workflow management module regards the description file of a workflow node as an independent file, and also regards the description file of a workflow as an independent file; The relationship between them is realized through the analysis of the workflow orchestration module;

The workflow of the workflow management module is:

(1.1) Import the local file; import the user's algorithm file through the system's file management system, and the cloud native description file of the algorithm will be automatically generated according to the algorithm name;

(1.2) Publish containerized software; by completing the cloud native description file, the user's local algorithm file or designed workflow is converted into an expression file that can be recognized by the task dispatcher through the workflow generator, and then the calculation in the underlying service is called. and middleware for storage resource services, containerize and package algorithms or designed workflows, and upload them to the cloud. Before containerization, users can configure the name of the container and the deployment environment;

(1.3) Import cloud files; by calling the computing and storage resource services in the underlying services, the description files of the containerized software deployed in the cloud are downloaded to the local client, allowing management and operation on the client.

(1.4) Export and save; save the user's file to the local disk environment through the system's file management system, so that the user can manage the file through the system resource manager.

The workflow orchestration module: the workflow orchestration module parses the opened workflow file by calling the workflow description language interpretation parser, or drags or imports the workflow node file, and correctly identifies it as each node graph, node graph The connection order and relationship of , are displayed in the orchestration panel; the workflow of the workflow orchestration module is as follows:

(2.1) Read the workflow file; by double-clicking the workflow suffix file in the workflow management module, load the workflow nodes and relationships between nodes contained in the file into the arrangement panel;

(2.2) Adding workflow nodes; new nodes can be added to the workflow by dragging and dropping node files from the workflow management module, or selecting Add Nodes in the toolbar;

(2.3) Connecting workflow nodes; by selecting the output end of a workflow node, dragging and pulling the arrow, and connecting to the input end of the target workflow node, the upper and lower steps of the scientific computing workflow can be determined;

(2.4) Configure the connection; by double-clicking the connection line, connect the output parameter of the previous step with the input parameter of the next step, the relationship is that one output parameter can be selected corresponding to multiple input parameters; if you want to configure multiple lower parameters for one output parameter One-step input parameters can be achieved by adding a link input parameter.

(2.5) Configure public parameters; public parameters are functions of fixed control of specific parameters in the workflow, which are connected to any step like a node, and carry out one-to-many assignment transfer with the input parameters of the node.

(2.6) Configure the workflow result; the workflow result is the final data or file that the designer hopes to obtain through the scientific computing workflow. After the configuration is completed, the result will be converted into a task management module or other high-level tasks through the workflow description language interpretation parser The interface description of the obtained result data required by the management system for users to easily obtain in each system;

(2.7) View the list and status of workflow nodes; by opening the workflow node panel, you can check whether each node file is correctly downloaded locally and added to the workflow to ensure that users can configure the nodes correctly;

(2.8) Debug and run the workflow; for the workflow that has been published, you can click the run function button in the orchestration module to call the task scheduling and execution middleware in the underlying service through the task dispatcher to allocate cloud computing cluster resources , to run the task.

The workflow node control module: the workflow node controller parses the node file introduced into the workflow by calling the workflow description language interpretation parser, and provides the user with the configuration function of the node characteristics; the process flow of the workflow node control module for:

(3.1) Read the workflow node configuration; double-click or right-click to open the selected node in the workflow orchestration module, and open the node control panel after parsing by the workflow description language interpretation parser, and you can see the scientific computing of the node. Description of capabilities, input and output parameters, and optional task operation mode;

(3.2) Task node attribute configuration At present, a node is divided into the following attributes, sequential operation, conditional branch and implementation intervention. The selection sequence operation is that if there is no manual intervention, the node will perform the operation of the next node after the operation is completed, so that the workflow can automatically run until an error is thrown or completed; the selection conditional branch is based on the output parameter results of a certain step. For different next-step path selections, usually the output of a conditional branch needs to be connected to at least two different nodes; when the scientific computing workflow reaches the current node, it will be automatically paused, and the parameters will be set separately after the user has reviewed the parameters. The input parameters of the next node can continue to execute;

(3.3) The task running mode is configured in the JobIn area. The user selects Scatter to perform concurrent computation on a specific input parameter. The scale of concurrency can be assigned by the specific value in Value or the function calculation relationship. In the JobOut area, the user selects Gather to converge multiple concurrently generated operation results to a certain output parameter to form data or files in dictionary format;

(3.4) Setting the output parameter as the result is similar to the configuration workflow result in the workflow orchestration module, and the process output result can be configured in the node controller.

The task management module: the task management module allows the user to learn the running status of a node's single-point task or concurrent task, input parameters, and output parameters and running logs through the operation of the orchestration panel. This module calls the interface of task monitoring and log service in the underlying service; after the task is started, the workflow orchestration module can visually see the workflow running stage and the running status of each node, and each node has five statuses : Standby (grey), Passed (green), Running (blue), Running but faulted (yellow) and Aborted (red). If all nodes turn green, the workflow has completed successfully. The process of the task management module is:

(4.1) View the running workflow node, select a workflow node in the orchestration module, and choose to view the node status details to open the task management module panel;

(4.2) Control the workflow node that is being executed. The control functions include the following three types: suspend the node task, modify the node parameters, and restart the node task; select a workflow node in the orchestration module, when it is found that the input and output parameters during the task running do not meet expectations , and check the log to find an exception but the task is still running, you can choose to pause to stop the running node task to avoid unnecessary waste of resources; when the workflow includes the implementation of the intervention node, you can modify the workflow node to control module, modify the output parameters, and then continue to run the selected workflow node, you can continue to execute the workflow from the current node; when any workflow is manually stopped at a node, it can be stopped at the node. Select Restart Node to resume the workflow from the current node.

The cloud high-performance scientific computing workflow design control system and user graphical interface provided by the present invention have the following technical advantages:

1. The design of the parsing layer encapsulates the standardized workflow language and the standardized method for executing tasks on cloud resources, which satisfies the concise definition and convenient design of the scientific computing process, and fulfills the needs of high-throughput computing. The ability to penetrate to the underlying service.

2. The design of the human-computer interaction layer allows users to quickly grasp the panorama of the scientific computing workflow by simply dragging, clicking, configuring and other graphical interface interactions in the process of designing the scientific computing process, reducing the design workflow. engineering threshold.

3. The cloud-native software architecture design ensures that in the research and development stage of scientific computing workflow, developers can continuously improve the algorithm functions of computing nodes, easily release self-developed algorithm software, adjust and replace nodes in the workflow, and improve research and development efficiency.

4. Comprehensively utilizes the combined advantages of local workflow software and containerized computing software, that is, it can fully reuse self-developed, cooperative authorization, third-party public data statistics and subject area analysis tools, and avoid huge software integration costs. It also guarantees massively parallel computing resources under high-throughput requirements.

5. The control module of the node task realizes or performs conditional manual intervention for a node, and gradually achieves the controllability of the task process, the saving of computing resources, and the dynamic balance of visual interaction.

Description of drawings

1 is a schematic diagram of a system architecture of the present invention;

Fig. 2 is the user graphical interface schematic diagram of the present invention;

Fig. 3 is the user graphic abstract interface diagram of the present invention;

FIG. 4 is a schematic flow chart of publishing containerized software according to an embodiment;

5 is a schematic diagram of the design and release workflow of an embodiment;

FIG. 6 is a schematic diagram of a workflow monitoring and management process in operation according to an embodiment.

detailed description

The specific technical solutions of the present invention are described with reference to the embodiments.

As shown in Figure 1, the cloud high-performance scientific computing workflow design control system includes three layers: the underlying service layer, the analysis layer and the human-computer interaction layer;

The parsing layer: the parsing layer is the actual support of the architecture, providing machine language conversion for the actual operation of the user on the interface, so that the system and the underlying service can operate according to the user's instructions, and at the same time, the data returned by the underlying service is in accordance with the Convert the designed form of the system to make the data conform to the user's interface operation habits and display logic;

The parsing layer consists of three modules: Workflow Description Language Interpretation Parser, Workflow Generator, and Task Dispatcher;

The main function of the task dispatcher is to call the underlying service according to the configuration file generated by the workflow generator, and finally realize workflow analysis, task execution and distribution.

The human-computer interaction layer: the human-computer interaction layer is responsible for implementing the functions of operation behavior and parameter configuration for users, including four modules: a workflow management module, a work node control panel, a workflow arrangement module, and a task manager.

This embodiment also provides a user graphical interface for the above-mentioned control system, and an abstract interface diagram is shown in FIG. 3 . The specific structure is shown in Figure 2, including a workflow management module, a workflow arrangement module, a workflow node control module, and a task management module;

The workflow of the workflow management module is:

(2.8) Debug and run the workflow; for the workflow that has been published, you can click the run function button in the orchestration module to call the task scheduling and execution middleware in the underlying service through the task dispatcher to allocate cloud computing cluster resources , run the task;

(3.2) Task node attribute configuration At present, a node is divided into the following attributes, sequential operation, conditional branch and implementation intervention. The selection sequence operation is that if no manual intervention is performed, the node will perform the operation of the next node after the operation is completed, so that the workflow can automatically run until an error is thrown or completed; the selection conditional branch is based on the output parameter results of a certain step. For different next-step path selections, usually the output of a conditional branch needs to be connected to at least two different nodes; when the scientific computing workflow reaches the current node, it will be automatically paused, and the parameters will be set separately after the user has reviewed the parameters. The input parameters of the next node can continue to execute;

Among them, the containerized software is released, as shown in Figure 4. In this embodiment, the user can deploy the locally written algorithm to the cloud through the user interface, turning it into a reusable cloud-native application, which can be easily invoked by a series of scientific computing workflows. for design and operation. The specific implementation process is as follows:

Step 1: The user imports the local algorithm file through the file import function of the workflow management module, and the left menu will generate a corresponding description file for it;

Step 2: The user double-clicks the description file, and after translation by the workflow description language interpretation parser, the user interface pops up the node control module to display the content in the description file;

Step 3: The user edits the parameters through the form provided by the user interface, and configures the attributes, types and default values of the parameters;

Step 4: After the configuration is complete, select the containerized software function on the toolbar to verify the written content;

Step 5: If the verification is passed, the client calls the underlying computing and storage services for cloud deployment. If the verification fails, go back to the third step to configure the description file.

Among them, the design and release workflow is shown in Figure 5. This embodiment enables users to use the deployed cloud software or algorithms to design and arrange scientific computing processes through dragging and visual form configuration, and deploy them to the cloud. Become a reusable cloud-native application that can be easily applied by members of the collaborative team. The specific implementation process is as follows:

The first step is to introduce cloud native applications: the user opens the cloud image library through the workflow management module, and imports a batch of required cloud algorithm or software description files. After the introduction, all the corresponding description files appear in the menu;

The second step is to introduce the editor: the user drags and drops the description files in sequence or selects a batch into the task orchestration module. After translation by the workflow description language interpretation parser, the orchestration module correctly displays the name, graph, status, etc. of the node. information;

The third step of connection and arrangement: The user can drag and drop the nodes in the canvas of the task arrangement module, or automatically layout, to arrange the arrangement positions of multiple nodes, and then the user can focus on the data and calculation sequence of the workflow. , connect different nodes in turn;

The fourth step is to configure the node: After the connection is completed, the user configures the node. In the node configuration, the concurrent and recycling configuration can be performed for one or more parameters of the nodes with data inflow and data outflow;

The fifth step is to save and publish: After the configuration is completed, save it. During the save process, the software calls the parsing layer to verify. If the connection and configuration are correct, it can be successfully saved. If there is an error, it will prompt the user to continue the configuration;

Step 6 Cloud deployment and construction: The saved files can be deployed in the cloud. After the user selects the cloud deployment, the basic cloud information is configured, the analysis layer calls the underlying computing and storage services, and the cloud service performs the software construction of the workflow. , which prompts the user on the user interface.

Among them, the running workflow monitoring and management, as shown in Figure 6, the running workflow monitoring and management module can be integrated by various types of user task management systems to realize the multiplexing support for each collaboration group or business department. In this embodiment, the user can open and view the running status of the workflow through the task management module in various systems. The specific implementation process is as follows:

The first step is to select the task of the user terminal task system: the user selects the task through the task management module integrated with other systems and invokes the interface for viewing details. The task management module invokes the underlying computing and services through the parsing layer, and requests the workflow file of the task carrier. After the download is successful, it is parsed by the workflow description language interpretation parser and displayed in the orchestration module.

The second step is to select a node in the task arrangement panel and view the detailed information: the workflow after parsing is completed, the overall workflow information and the connection and status of each node will be displayed. The user selects a node, right-clicks and selects to view the node information to open the task management module.

The third step is to select a task log in the concurrent tasks: after opening the management module of a node, if the node is in parallel mode, you can view the status and log of each concurrent subtask, right-click the selected subtask, And click View Log to open the subtask log for task debugging.

Claims

The cloud high-performance scientific computing workflow design control system is characterized in that it includes three layers: the bottom service layer, the analysis layer and the human-computer interaction layer;

The underlying service layer: the underlying service provides basic services for upper-layer applications in the form of SDK;

The parsing layer: provides machine language conversion for the user's actual operation on the interface, so that the system and the underlying service can operate according to the user's instructions, and at the same time, the data returned by the underlying service is converted in the form designed by the system , so that the data conforms to the user's interface operation habits and display logic;

The human-computer interaction layer is responsible for realizing operation behavior and parameter configuration functions for users.
The cloud high-performance scientific computing workflow design control system according to claim 1, wherein the underlying service layer includes: task scheduling and execution middleware, computing and storage resource management services, task monitoring and log services ;

The main function of the task scheduling and execution middleware is to enable the cloud computing cluster to package and run tasks through packaged instructions;

The computing and storage resource service, the main function of which provides resource uploading, downloading, storage and distribution for the workflow and actual running tasks of the client;

The main function of the task monitoring and log service is to collect and store the status and error-throwing data returned during the task running process, and to query through commands.
The cloud high-performance scientific computing workflow design control system according to claim 1, wherein the parsing layer comprises three modules: a workflow description language interpretation parser, a workflow generator, and a task dispatcher;

The main function of the workflow description language interpretation parser is to apply a standardized workflow language to describe and convert cloud-native containerized software or algorithms; on the underlying server, it converts user operations into those required by the underlying SDK. data structure;

The main function of the workflow generator is to process and convert the workflow nodes and related configurations that users drag and drop to combine into workflow files, so that the task distributor can identify the workflow sequence relationship, distribution mode and parameter configuration. ;

The main function of the task dispatcher is to call the underlying service according to the configuration file generated by the workflow generator, and finally realize the analysis of the workflow and the execution and distribution of tasks.
The cloud high-performance scientific computing workflow design control system according to claim 1, wherein the human-computer interaction layer includes four modules: a workflow management module, a work node control panel, a workflow orchestration module, task manager;

The main function of the workflow management module is to download and load containers and workflow description files in the cloud, and manage, modify, or introduce workflows in the form of files on the client side;

The main function of the work node control panel is to allow the user to configure the work node on the client side, and to configure the execution properties and operation parameters of the node;

The main function of the workflow orchestration module is to allow users to introduce different workflow nodes and establish connection relationships between different workflow nodes;

The main function of the task manager is to be able to view the running status, input and output and task log of each node's task or each sub-task in a concurrent task in real time.
A graphical user interface, characterized in that it is used in the cloud high-performance scientific computing workflow design control system according to any one of claims 1 to 4, including a workflow management module, a workflow orchestration module, a workflow node control module, a task management module;

The workflow management module: the workflow management module regards the description file of a workflow node as an independent file, and also regards the description file of a workflow as an independent file; The relationship between them is realized through the analysis of the workflow orchestration module;

The workflow orchestration module: the workflow orchestration module parses the opened workflow file by calling the workflow description language interpretation parser, or drags or imports the workflow node file, and correctly identifies it as each node graph, node graph The connection order and relationship of , displayed in the layout panel;

The workflow node control module: the workflow node controller parses the node file introduced into the workflow by calling the workflow description language interpretation parser, and provides the user with the configuration function of the node characteristics;

Described task management module: the task management module allows the user to learn the running status of a single-point task or concurrent task of a node, input parameters, and output parameters and operation logs through the operation of the orchestration panel; the task management module calls the The interface of task monitoring and log service in the underlying service; after the task is started, the workflow running stage and the running status of each node can be visually seen through the workflow orchestration module. Each node has five statuses: standby, passed, running If all nodes are passed, it means that the workflow is successfully completed.
The graphical user interface according to claim 5, wherein the process of the workflow management module is:

(1.1) Import the local file; import the user-based algorithm file through the system's file management system, and the cloud native description file of the algorithm will be automatically generated according to the algorithm name;

(1.2) Publishing containerized software; by completing the cloud native description file, the user's local algorithm file or designed workflow is converted into an expression file that can be recognized by the task dispatcher through the workflow generator, and then the calculation in the underlying service is called. and middleware for storage resource services, containerize the algorithm or designed workflow, and upload it to the cloud. Before containerization, the user configures the name of the container and the deployment environment;

(1.3) Import cloud files; by invoking the computing and storage resource services in the underlying services, download the description files of the containerized software deployed in the cloud to the local client, allowing management and operation on the client;

(1.4) Export and save; save the user's file to the local disk environment through the system's file management system, so that the user can manage the file through the system resource manager.
The graphical user interface according to claim 6, wherein the process of the workflow orchestration module is:

(2.1) Read the workflow file; by double-clicking the workflow suffix file in the workflow management module, load the workflow node and the relationship between the nodes contained in the file into the arrangement panel;

(2.2) Add a workflow node; add a new node to the workflow by dragging and dropping the node file from the workflow management module, or selecting Add Node in the toolbar;

(2.3) Connecting workflow nodes; by selecting the output end of a workflow node, dragging and pulling out the arrow, and connecting to the input end of the target workflow node, the upper and lower steps of the scientific computing workflow can be determined;

(2.4) Configure the connection; by double-clicking the connection line, connect the output parameter of the previous step with the input parameter of the next step, the relationship is that one output parameter can be selected corresponding to multiple input parameters; if you want to configure multiple lower parameters for one output parameter One-step input parameters can be achieved by adding a connection line input parameter;

(2.5) Configure public parameters; public parameters are the function of fixed control of specific parameters in the workflow, which are connected to any step like a node, and carry out a one-to-many assignment transfer with the input parameters of the node;

(2.6) Configure the workflow result; the workflow result is the final data or file that the designer hopes to obtain through the scientific computing workflow. After the configuration is completed, the result will be converted into a task management module or other high-level tasks through the workflow description language interpretation parser The interface description of the obtained result data required by the management system for users to easily obtain in each system;

(2.7) Check the list and status of workflow nodes; by opening the workflow node panel, check whether each node file is correctly downloaded locally and added to the workflow, so as to ensure that the user can configure the nodes correctly;

(2.8) Debug and run the workflow; for the workflow that has been published, click the run function button in the orchestration module to call the task scheduling and execution middleware in the underlying service through the task dispatcher, and allocate cloud computing cluster resources. Run the task.
The graphical user interface according to claim 7, wherein the process of the workflow node control module is:

(3.1) Read the workflow node configuration; double-click or right-click to open the selected node in the workflow orchestration module, and open the node control panel after parsing by the workflow description language interpretation parser, and you can see the scientific computing of the node. Description of capabilities, input and output parameters, and optional task operation mode;

(3.2) Task node attribute configuration At present, a node is divided into the following attributes: sequential operation, conditional branching and implementation intervention; selecting a conditional branch is to select different next-step paths according to the output parameter results of a certain step, and the output of a conditional branch needs to be Connect at least two different nodes; choose to implement the intervention, when the scientific computing workflow progresses to the current node, it will automatically pause, and after the user has reviewed the parameters, set the input parameters of the next node before continuing to execute;

(3.3) The task running mode is configured in the JobIn area. The user selects Scatter to perform concurrent calculation on a specific input parameter. The scale of concurrency can be assigned by the specific value in Value or the function calculation relationship; in the JobOut area, the user selects Gather That is, to converge multiple concurrently generated operation results to a certain output parameter to form data or files in dictionary format;

(3.4) Setting the output parameter as the result is similar to the configuration workflow result in the workflow orchestration module, and the process output result can be configured in the node controller.
The graphical user interface according to claim 8, wherein the process of the task management module is:

(4.1) View the running workflow node, select a workflow node in the orchestration module, and choose to view the node status details to open the task management module panel;

(4.2) Control the workflow node that is being executed. The control functions include the following three types: suspend the node task, modify the node parameters, and restart the node task; select a workflow node in the orchestration module, when it is found that the input and output parameters during the task running do not meet expectations , and check the log to find an exception but the task is still running, you can choose to pause to stop the running node task to avoid unnecessary waste of resources; when the workflow includes the implementation of the intervention node, by modifying the workflow node control module , modify the output parameters, and then continue the operation of the selected workflow node, you can continue to execute the workflow from the current node; when any workflow is manually stopped at a node, you can select it at the stop node Restart the node to resume the workflow from the current node.