Hybrid cloud native high computing power and high concurrency solution and device based on message queue

Through a hybrid cloud-native method based on message queues, combined with public and private cloud resource scheduling, the problem of high concurrency and high computing resource requirements is solved, and effective response and processing to massive concurrent computing tasks is achieved, ensuring the stability of the system Sex and resource security.

CN114741174BInactive Publication Date: 2024-11-15WUHAN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210388943.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-04-13
Publication Date
2024-11-15
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

In artificial intelligence training tasks, the demand for high concurrency and high computing resources has increased sharply, and it is difficult for existing technologies to effectively respond to and handle massive concurrent computing tasks, resulting in unstable system operation.

Method used

Adopt a hybrid cloud-native solution based on message queues, and create message queues, task scheduling algorithms and consumer container interactions to realize the distribution, processing and result return of computing tasks. Using resource scheduling of public and private clouds, ensure that computing tasks are in Respond within a reasonable time and restart and try in an error situation.

Benefits of technology

It realizes effective response and processing of massive concurrent computing tasks with high computing power, reduces costs, ensures resource security, and ensures the reliability and responsiveness of computing tasks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114741174B_ABST
    Figure CN114741174B_ABST
Patent Text Reader

Abstract

The present invention discloses a hybrid cloud native high computing power and high concurrency solution method and device based on a message queue, which relates to the cloud native field. The method includes creating a message queue, and storing different types of computing tasks in a high concurrent task scenario based on the created message queue; creating a task scheduling algorithm, and distributing the computing tasks stored in the message queue to consumers who listen to the message queue based on the created task scheduling algorithm; consumers and computing containers in the same POD interact to send computing tasks to the computing containers for computing, and consumers return the computing results to the message queue; the listener corresponding to the computing task obtains the output result from the reply queue, and writes the obtained output result into a database. The present invention can achieve effective response and processing of high computing power and massive concurrent computing tasks.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of cloud native, and in particular to a hybrid cloud native high-computing power and high-concurrency solution and device based on a message queue. Background Art

[0002] The problem of high concurrency and high computing power resource requirements has always been a pressing issue to be solved today. According to relevant analysis, the computing power used in artificial intelligence training tasks is growing exponentially, and its current speed is doubling every 3.5 months (compared to Moore's Law, which is doubling every 18 months). People's demand for computing power has now increased by more than 300,000 times.

[0003] At a time when artificial intelligence algorithms have such stringent requirements on computing resources, if tens of thousands of concurrent requests are initiated at the same time, a stable and available architecture solution is necessary to ensure the normal operation of the system. Therefore, how to effectively respond to and process high-computing power and massive concurrent computing tasks has become an urgent problem to be solved. Summary of the invention

[0004] In view of the defects existing in the prior art, the purpose of the present invention is to provide a hybrid cloud native high-computing power and high-concurrency solution and device based on message queues, which can effectively respond to and process high-computing power and massive concurrent computing tasks.

[0005] To achieve the above objectives, the present invention provides a hybrid cloud native high computing power and high concurrency solution based on message queue, which specifically includes the following steps:

[0006] Create a message queue and store different types of computing tasks in high-concurrency task scenarios based on the created message queue;

[0007] Create a task scheduling algorithm, and distribute the computing tasks stored in the message queue to consumers who monitor the message queue based on the created task scheduling algorithm;

[0008] Consumers and computing containers in the same POD interact to send computing tasks to the computing container for computing, and consumers return the computing results to the message queue;

[0009] The listener corresponding to the computing task obtains the output result from the reply queue and writes the obtained output result to the database.

[0010] Based on the above technical solutions,

[0011] The message queue includes a consumption queue, a reply queue and a dead letter queue;

[0012] The consumption queue is used to store computing tasks to be calculated;

[0013] The reply queue is used to store the computing tasks that have been correctly responded to by the consumer, and to store the computing results obtained after the computing tasks are correctly calculated by the computing container;

[0014] The dead letter queue is used to store computing tasks when an exception or unknown error occurs during the computing process of the computing container.

[0015] On the basis of the above technical solution, the storage of computing tasks in the message queue specifically includes:

[0016] The producer sends the computation task to the default switch;

[0017] The default switch forwards the computing tasks to the corresponding message queue according to the route.

[0018] Based on the above technical solutions,

[0019] An environment variable identifier is set in the consumer container where the consumer is located;

[0020] The environment variable identifier is used to declare the message queue that the consumer monitors;

[0021] The consumer obtains the status change of the computing task from the monitored queue, and the consumer can only take out the computing task from the message queue monitored by the consumer.

[0022] Based on the above technical solutions,

[0023] After the computing tasks stored in the message queue are distributed to the consumers who monitor the message queue based on the created task scheduling algorithm, it also includes: the consumers implement resource scheduling based on the preset strategy according to their own types;

[0024] The types of consumers include public clouds and private clouds;

[0025] The preset policy is that consumers of the private cloud type have priority in receiving computing tasks. When the private cloud resources are limited and the computing tasks cannot be completed, consumers of the public cloud type receive the computing tasks.

[0026] Based on the above technical solution, the consumers and computing containers in the same POD interact with each other to send computing tasks to the computing container for computing. The specific steps include:

[0027] After receiving the computing task, the consumer obtains the input parameter file corresponding to the computing task from the object storage service, where both the public cloud and the private cloud are deployed with the object storage service;

[0028] Establishing a start marker in the shared storage volume of the compute container and the consumer container;

[0029] The computing container monitors the shared storage volume in real time, and obtains the input parameter file of the computing task from the shared storage volume after the computing container learns the start identifier;

[0030] The computing container calls the computing model inside the computing container and uses the machine computing resources to calculate the computing tasks.

[0031] Based on the above technical solution, after the consumers and computing containers in the same POD interact to send computing tasks to the computing container for computing, the following is also included:

[0032] The computing container continues to monitor the computing model until the computing model stops computing;

[0033] The computing container stops the computing process and creates an end file in the shared storage volume as an end marker;

[0034] After the consumer container learns the end mark in the shared storage volume, it puts the calculation result of the computing task into the Reply queue and stores the input result file corresponding to the calculation result in the object storage service.

[0035] Based on the above technical solution, the listener corresponding to the computing task obtains the output result from the reply queue, and the specific steps include:

[0036] The listener monitors the reply queue in the message queue in real time;

[0037] The listener obtains the message information of the completed computing tasks from the reply queue;

[0038] Based on the message information, obtain the corresponding input result file from the object storage service.

[0039] Based on the above technical solution, the steps of writing the obtained output results into a database include:

[0040] The listener downloads the obtained input result file to the local computer, reads the input result file, and stores the read result data in the database.

[0041] The present invention provides a hybrid cloud native high computing power and high concurrency solution system based on a message queue, including:

[0042] The first creation module is used to create a message queue and store different types of computing tasks in a high-concurrency task scenario based on the created message queue;

[0043] The second creation module is used to create a task scheduling algorithm, and distribute the computing tasks stored in the message queue to consumers who monitor the message queue based on the created task scheduling algorithm;

[0044] The computing module is used to drive the interaction between consumers and computing containers in the same POD to send computing tasks to the computing container for computing, and the consumers return the computing results to the message queue;

[0045] The writing module is used to drive the listener corresponding to the computing task to obtain the output result from the reply queue and write the obtained output result into the database.

[0046] Compared with the prior art, the advantages of the present invention are: reasonable scheduling of public clouds and private clouds in a hybrid cloud environment, resource allocation for computing models with high computing power requirements to reduce costs while ensuring safety, queue planning for computing task requests in high concurrency scenarios to ensure that all tasks are responded to within a reasonable time, and timely processing and re-queuing of erroneous requests in the event of incorrect computing resource allocation so that all tasks are correctly responded to, continuous scanning of computing containers and timely restart in the event of excessive computing time timeout, thereby achieving effective response and processing of high-computing power and massive concurrent computing tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The present invention provides a flowchart of a hybrid cloud native high computing power and high concurrency solution based on a message queue in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0049] See also Figure 1 As shown, an embodiment of the present invention provides a hybrid cloud native high computing power and high concurrency solution based on a message queue. A computing cluster is arranged in a public cloud environment and a private cloud environment to undertake high-concurrency computing tasks. The computing cluster is implemented in the form of kuberntes (an open source containerized application for managing multiple hosts in a cloud platform), which can provide CPU (Central Processing Unit) computing nodes and GPU (Graphics Processing Unit) computing nodes to undertake different types of tasks. The requirements for different computing resources. In terms of massive task scheduling, message queues can be used to uniformly manage tasks, and algorithms can be designed to reasonably schedule tasks so that they can all be effectively responded to. There is a set of object storage services in the public cloud and private cloud to provide temporary data access during task computing. A cache database is used to store the status information of task computing in order to achieve real-time monitoring of message status, ensuring that this method has certain availability and versatility.

[0050] The hybrid cloud native high computing power and high concurrency solution based on message queue in the embodiment of the present invention specifically includes the following steps:

[0051] S1: Create a message queue and store different types of computing tasks in high-concurrency task scenarios based on the created message queue;

[0052] In the embodiment of the present invention, the message queue includes a consumption queue, a reply queue and a dead letter queue. The consumption queue is used to store the computing tasks to be calculated; the reply queue is used to store the computing tasks that have been correctly responded to by the consumer, and the computing results obtained after the computing tasks are correctly calculated by the computing container; that is, after the computing tasks are received by the corresponding consumers and the correct results are obtained through the computing container, the computing task information is stored in the reply queue and waits to be written into the database.

[0053] The dead letter queue is used to store computing tasks when an exception or unknown error occurs during the computing process of the computing container. That is, once a computing task fails to complete smoothly, it will be placed in the dead letter queue.

[0054] In the embodiment of the present invention, the storage of computing tasks in the message queue specifically includes:

[0055] S101: The producer sends the computing task to the default switch;

[0056] S102: The default switch forwards the computing task to the corresponding message queue according to the route, that is, when the producer sends the task, the task is forwarded to different queues according to the identifiers of different tasks.

[0057] S2: Create a task scheduling algorithm, and distribute the computing tasks stored in the message queue to consumers who monitor the message queue based on the created task scheduling algorithm;

[0058] In an embodiment of the present invention, an environment variable identifier is set in the consumer container where the consumer is located; the environment variable identifier is used to declare the message queue monitored by the consumer; the consumer obtains the state change of the computing task from the monitored queue, and the consumer can only take out the computing task from the message queue monitored by the consumer.

[0059] In an embodiment of the present invention, after the computing tasks stored in the message queue are distributed to consumers who have listened to the message queue based on the created task scheduling algorithm, it also includes: consumers implement resource scheduling based on preset strategies according to their own types; the types of consumers include public clouds and private clouds; the preset strategy is that consumers of the private cloud type have priority in receiving computing tasks, and when the private cloud resources are limited and the computing tasks cannot be completed, consumers of the public cloud type receive the computing tasks.

[0060] S3: Consumers and computing containers in the same POD (container) interact to send computing tasks to the computing container for computing, and the consumer returns the computing results to the message queue;

[0061] In the embodiment of the present invention, consumers and computing containers in the same POD interact to send computing tasks to the computing container for computing. The specific steps include:

[0062] S301: After receiving the computing task, the consumer obtains the input parameter file corresponding to the computing task from the object storage service, where both the public cloud and the private cloud are deployed with the object storage service;

[0063] S302: Establish a start identifier in the shared storage volume of the computing container and the consumer container;

[0064] S303: The computing container monitors the shared storage volume in real time, and obtains the input parameter file of the computing task from the shared storage volume after the computing container learns the start identifier;

[0065] S304: The computing container calls the computing model inside the computing container and uses the machine computing resources to calculate the computing task.

[0066] In the embodiment of the present invention, after the consumer and the computing container in the same POD interact to send the computing task to the computing container for computing, the following further includes:

[0067] S31: The computing container continues to monitor the computing model until the computing model stops computing;

[0068] S32: The computing container stops the computing process and creates an end file in the shared storage volume as an end marker;

[0069] S33: After the consumer container learns the end identifier in the shared storage volume, it puts the calculation result of the calculation task into the Reply queue, and stores the input result file corresponding to the calculation result into the object storage service.

[0070] In an embodiment of the present invention, the message header and message body of each computing task use the same ID. The consumption mode configured on the consumer side is to not respond when entering the consumption method normally. The computing task will be responded to normally only after the successful processing. When the calculation is abnormal or an unknown error occurs, the computing task will not be responded to, so as to ensure that the number of each computing task is unique and the caller is unique.

[0071] At the same time, a retry message header is defined for all computing tasks. When a computing task fails to be processed for the first time, it automatically enters the first-level dead letter queue. When a computing task fails to be processed and enters, if the retry attribute is greater than 3, the computing task is thrown into the second-level dead letter queue, and then the retry is incremented and forwarded to the first-level dead letter queue, so that all computing tasks in the dead letter queue are manually processed or responded normally, ensuring the reliability of the message.

[0072] S4: The listener corresponding to the computing task obtains the output result from the reply queue and writes the obtained output result into the database.

[0073] In the embodiment of the present invention, the listener corresponding to the computing task obtains the output result from the reply queue, and the specific steps include:

[0074] S401: The listener monitors the reply queue in the message queue in real time;

[0075] S402: The listener obtains the message information of the completed computing task from the reply queue;

[0076] S403: Based on the message information, obtain the corresponding input result file from the object storage service. Specifically, obtain the task ID (Identity Document) from the message body of the message information, concatenate the unique identifier according to the task ID, and obtain the output result file from the object storage service according to the unique identifier.

[0077] In the embodiment of the present invention, the obtained output result is written into the database, and the specific steps include: the listener downloads the obtained input result file to the local, reads the input result file, and stores the read result data into the database.

[0078] The present invention combines the current trend of rapid development of cloud computing, adopts open source stack for containerization, improves flexibility and maintainability based on microservice architecture, and uses cloud platform facilities to achieve elastic scaling, dynamic scheduling, and optimize resource utilization to solve massive computing tasks under high computing power requirements, so as to reasonably arrange machine computing resources and efficiently complete computing tasks. Public clouds and private clouds are mixed and matched to obtain the best results. Such a solution can reduce the cost while ensuring resource security. Therefore, the present invention adopts a hybrid cloud model for cloud computing resource planning and arrangement. At the same time, when high-concurrency computing task requests are sent simultaneously, message queue middleware is used for unified management of massive messages to ensure that the messages are safe, reliable, and not lost. The present invention can also effectively monitor the execution status of tasks, and can also ensure that concurrent tasks are effectively processed on the premise of ensuring reasonable arrangement and planning of resources required for computing tasks.

[0079] The hybrid cloud native high computing power and high concurrency solution of the present invention mainly includes the following three tasks:

[0080] Task 1: Design appropriate message queues to store different states of computing tasks. Different message queue types are used to manage tasks in different states. Design task scheduling algorithms to uniformly manage message queues. At the same time, different computing tasks are sent to public cloud clusters or private cloud clusters according to certain strategies.

[0081] Task 2: Deploy computing containers and service containers in private cloud and public cloud environments, allocate reasonable computing resources to computing containers, and facilitate communication between computing containers and service containers, so that the service container can send computing parameters to the computing container to complete the calculation;

[0082] Task three: Use the cache storage service service to record statistical indicators such as the number of transactions per second at the current moment and the distribution of computing time of completed tasks per unit time, so as to facilitate timely acquisition of system operation status during task scheduling and performance monitoring, and avoid losses that may be caused by monitoring data lags.

[0083] To achieve the effect of completing work one, there are several steps: Step 1, design different types of message queues to manage message queue information; Step 2, realize broadcast, multicast, and unicast effects; Step 3, consumers dynamically monitor message queues; Step 4, realize message idempotence; Step 5, design message return queues.

[0084] In work 2, the present invention selects kubernetes as a tool for multi-container orchestration management. The goal of Kubernetes is to make the deployment of containerized applications simple and efficient. Kubernetes provides a mechanism for application deployment, planning, updating, and maintenance. The essence of kubernetes is a group of server clusters. It can run specific programs on each node of the cluster to manage the containers in the nodes and realize the automation of resource management.

[0085] To achieve the desired effect of Work 2, the following steps are required: Step 1, realize the interaction between the task consumer container and the computing container; Step 2, container management in a private cloud environment; Step 3, container management in a public cloud environment.

[0086] The present invention uses Redis (remote dictionary service) as the technical selection of intermediate cache and access business monitoring indicators, and utilizes the unique fast access characteristics of Redis to meet the needs of high concurrent business and timely monitor system performance and business scheduling. To achieve the desired effect of work three, the following steps are required: Step 1, design a redis table structure that meets business requirements; Step 2, improve the concurrent performance of redis operations; Step 3, simplify redistemplate resource operations; Step 4, realize real-time reporting and monitoring of business data.

[0087] The present invention manages massive task requests in high-concurrency scenarios through message queue middleware. During the task processing, the execution status of the task can be monitored without any processing of the container. In addition, the message queue design of the present invention can support tasks of multiple task types to work simultaneously, and perform corresponding task processing for different types of computer tasks.

[0088] The message retry mechanism in the present invention will continuously monitor whether any exceptions are captured during the task execution process. Once an exception occurs in the task processing, the task will be re-queued and re-executed. This mechanism can prevent the situation where the task cannot be completed normally and cannot respond due to temporary network fluctuations or other reasons, thereby ensuring the reliability of task processing.

[0089] The present invention performs reasonable task scheduling on the hybrid cloud, first assigning tasks to the private cloud for processing, and only assigning tasks to the public cloud's paid resources for processing when the private cloud's task resources are insufficient. At the same time, there is also an elastic scaling strategy in the public cloud environment, thereby achieving the goal of both improving efficiency and reducing costs.

[0090] The present invention uses the cache service access business monitoring indicators, which can systematically view the execution information of each task in the high-concurrency task request, such as the task's operating environment (public cloud or private cloud), task execution time, task waiting time, etc. The present invention puts a large number of tasks in the message queue, which will not put pressure on the machine at the peak of traffic, and supports the ability to send and receive tens of millions of messages.

[0091] An embodiment of the present invention provides a hybrid cloud native high computing power and high concurrency solution system based on a message queue, comprising a first creation module, a second creation module, a calculation module and a writing module.

[0092] The first creation module is used to create a message queue, and based on the created message queue, store different types of computing tasks in high-concurrency task scenarios; the second creation module is used to create a task scheduling algorithm, and based on the created task scheduling algorithm, distribute the computing tasks stored in the message queue to consumers who listen to the message queue; the computing module is used to drive the interaction between consumers and computing containers in the same POD to send computing tasks to the computing containers for calculation, and the consumers return the calculation results to the message queue; the writing module is used to drive the listener corresponding to the computing task to obtain the output result from the reply queue, and write the obtained output result into the database.

[0093] The present invention is not limited to the above-mentioned embodiments. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention. The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

Claims

1. A hybrid cloud native high computing power and high concurrency solution based on message queue, characterized in that: The specific steps include: Create a message queue and store different types of computing tasks in high-concurrency task scenarios based on the created message queue; Create a task scheduling algorithm, and distribute the computing tasks stored in the message queue to consumers who monitor the message queue based on the created task scheduling algorithm; Consumers and computing containers in the same POD interact to send computing tasks to the computing container for computing, and consumers return the computing results to the message queue; The listener corresponding to the computing task obtains the output result from the reply queue and writes the obtained output result to the database; Among them, after the computing tasks stored in the message queue are distributed to the consumers who monitor the message queue based on the created task scheduling algorithm, it also includes: the consumers implement resource scheduling based on the preset strategy according to their own types; The types of consumers include public clouds and private clouds; The preset policy is that consumers of the private cloud type are given priority to receive computing tasks. When the private cloud resources are limited and the computing tasks cannot be completed, consumers of the public cloud type receive computing tasks. The consumer and the computing container in the same POD interact with each other to send the computing task to the computing container for computing. The specific steps include: After receiving the computing task, the consumer obtains the input parameter file corresponding to the computing task from the object storage service, where both the public cloud and the private cloud are deployed with the object storage service; Establishing a start marker in the shared storage volume of the compute container and the consumer container; The computing container monitors the shared storage volume in real time, and obtains the input parameter file of the computing task from the shared storage volume after the computing container learns the start identifier; The computing container calls the computing model inside the computing container and uses the machine computing resources to calculate the computing task; When consumers and computing containers in the same POD interact to send computing tasks to the computing container for computing, the following also occurs: The computing container continues to monitor the computing model until the computing model stops computing; The computing container stops the computing process and creates an end file in the shared storage volume as an end marker; After the consumer container learns the end mark in the shared storage volume, it puts the calculation result of the computing task into the Reply queue and stores the input result file corresponding to the calculation result in the object storage service.

2. A hybrid cloud native high computing power and high concurrency solution based on a message queue as claimed in claim 1, characterized in that: The message queue includes a consumption queue, a reply queue and a dead letter queue; The consumption queue is used to store computing tasks to be calculated; The reply queue is used to store the computing tasks that have been correctly responded to by the consumer, and to store the computing results obtained after the computing tasks are correctly calculated by the computing container; The dead letter queue is used to store computing tasks when an exception or unknown error occurs during the computing process of the computing container.

3. A hybrid cloud native high computing power and high concurrency solution based on message queue as claimed in claim 2, characterized in that: The storage of computing tasks in the message queue specifically includes: The producer sends the computation task to the default switch; The default switch forwards the computing tasks to the corresponding message queue according to the route.

4. A hybrid cloud native high computing power and high concurrency solution based on a message queue as claimed in claim 1, characterized in that: An environment variable identifier is set in the consumer container where the consumer is located; The environment variable identifier is used to declare the message queue that the consumer monitors; The consumer obtains the status change of the computing task from the monitored queue, and the consumer can only take out the computing task from the message queue monitored by the consumer.

5. A hybrid cloud native high computing power and high concurrency solution based on message queue as claimed in claim 1, characterized in that: The listener corresponding to the computing task obtains the output result from the reply queue. The specific steps include: The listener monitors the reply queue in the message queue in real time; The listener obtains the message information of the completed computing tasks from the reply queue; Based on the message information, obtain the corresponding input result file from the object storage service.

6. A hybrid cloud native high computing power and high concurrency solution based on message queue as claimed in claim 5, characterized in that: The step of writing the obtained output result into the database comprises: The listener downloads the obtained input result file to the local computer, reads the input result file, and stores the read result data in the database.

7. A hybrid cloud native high computing power and high concurrency solution system based on message queue, characterized in that: include: The first creation module is used to create a message queue and store different types of computing tasks in a high-concurrency task scenario based on the created message queue; The second creation module is used to create a task scheduling algorithm, and distribute the computing tasks stored in the message queue to consumers who monitor the message queue based on the created task scheduling algorithm; The computing module is used to drive the interaction between consumers and computing containers in the same POD to send computing tasks to the computing container for computing, and the consumers return the computing results to the message queue; The writing module is used to drive the listener corresponding to the computing task to obtain the output result from the reply queue and write the obtained output result into the database; Among them, after the computing tasks stored in the message queue are distributed to the consumers who monitor the message queue based on the created task scheduling algorithm, it also includes: the consumers implement resource scheduling based on the preset strategy according to their own types; The types of consumers include public clouds and private clouds; The preset policy is that consumers of the private cloud type are given priority to receive computing tasks. When the private cloud resources are limited and the computing tasks cannot be completed, consumers of the public cloud type receive computing tasks. The consumer and the computing container in the same POD interact with each other to send the computing task to the computing container for computing. The specific steps include: After receiving the computing task, the consumer obtains the input parameter file corresponding to the computing task from the object storage service, where both the public cloud and the private cloud are deployed with the object storage service; Establishing a start marker in the shared storage volume of the compute container and the consumer container; The computing container monitors the shared storage volume in real time, and obtains the input parameter file of the computing task from the shared storage volume after the computing container learns the start identifier; The computing container calls the computing model inside the computing container and uses the machine computing resources to calculate the computing task; When consumers and computing containers in the same POD interact to send computing tasks to the computing container for computing, the following also occurs: The computing container continues to monitor the computing model until the computing model stops computing; The computing container stops the computing process and creates an end file in the shared storage volume as an end marker; After the consumer container learns the end mark in the shared storage volume, it puts the calculation result of the computing task into the Reply queue and stores the input result file corresponding to the calculation result in the object storage service.