Container-based algorithm operation cluster deployment control method and system

Abstract

The invention provides a container-based algorithm operation cluster deployment control method and system, and the method comprises the steps: analyzing an algorithm operation request from a user side, determining the attribute information of an algorithm operation task expected to be executed by the user side, constructing a corresponding algorithm operation environment in a Docker container, and carrying out the operation of the algorithm operation cluster. Therefore, a compatible and matched algorithm operation environment is provided for different algorithm operation demand occasions; on the basis of GPU drive of the algorithm platform, the Docker container is driven, and when the Docker container is in a normal working state, the algorithm code and the algorithm model are mounted on the Docker container, so that the Docker container can independently carry out algorithm operation, and the working stability in the container is ensured; according to the technical scheme, the corresponding middleware is constructed for the Docker container based on the interface of the Docker container, the algorithm operation result in the Docker container is stored in the middleware, the data storage pressure of the container is effectively reduced, then the corresponding algorithm operation result is sent to the user side, and the compatibility of an algorithm platform to different algorithm operation occasions and the algorithm operation efficiency and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the field of computers, and in particular to a container-based algorithm operation cluster deployment control method and system. Background Art

[0002] The algorithm platform is used to provide different types of algorithms and drive algorithm operations to complete the corresponding algorithm operation tasks. The algorithm platform is usually deployed on a single machine, and the algorithm operation can only be implemented in a containerless environment. This results in a single operating environment for the algorithm platform, a small number of compatible and supported algorithms, and a high difficulty in clustered deployment. It is difficult to configure the algorithm operation environment according to different computing requirements, which reduces the compatibility of the algorithm platform with different algorithm operation scenarios and the efficiency and reliability of the algorithm operation. Summary of the invention

[0003] The purpose of the present invention is to provide a container-based algorithm operation cluster deployment control method and system, which analyzes the algorithm operation request from the user end, determines the algorithm operation task attribute information that the user end expects to execute, and thereby builds a corresponding algorithm operation environment in the Docker container, thereby providing a compatible and matching algorithm operation environment for different algorithm operation demand occasions; it also drives the Docker container based on the GPU driver of the algorithm platform, and when the Docker container is in normal working state, mounts the algorithm code and algorithm model in the Docker container, so that the Docker container can independently perform the algorithm operation, ensuring the working stability inside the container; based on the interface of the Dcoker container, builds the corresponding middleware for the Docker container, and stores the algorithm operation results inside the Docker container in the middleware, effectively reducing the data storage pressure of the container, and then sends the corresponding algorithm operation results to the user end, which is conducive to realizing the cluster deployment of algorithm operation and improving the compatibility of the algorithm platform for different algorithm operation occasions and the algorithm operation efficiency and reliability.

[0004] The present invention is achieved through the following technical solutions: The container-based algorithm operation cluster deployment control method includes: Analyze the algorithm running request from the user end to determine the attribute information of the algorithm running task that the user end expects to execute; and build a corresponding algorithm running environment in the Docker container based on the attribute information of the algorithm running task; Based on the GPU driver of the algorithm platform, the Docker container is driven, and it is determined whether the Docker container is in a normal working state; when the Docker container is in a normal working state, the algorithm code and the algorithm model are mounted on the Docker container; An interface is set for the Docker container, and based on the interface of the Docker container, a corresponding middleware is built for the Docker container; the algorithm running results inside the Docker container are stored in the middleware, and based on a running result acquisition request from the user end, the corresponding algorithm running results are sent to the user end.

[0005] Optionally, analyzing the algorithm running request from the user end to determine the attribute information of the algorithm running task that the user end expects to execute; and building a corresponding algorithm running environment in the Docker container based on the attribute information of the algorithm running task, including: Analyze the algorithm running request from the user end, extract the algorithm type information involved in all the pending algorithm running tasks from the pending algorithm running task queue of the user end; divide all the pending algorithm running tasks of the user end into a plurality of pending algorithm running task clusters based on the algorithm type information; wherein all the pending algorithm running tasks under each pending algorithm running task cluster have the same algorithm type information; Based on the amount of algorithm data to be run of all algorithm running task clusters to be processed, the processing order of all algorithm running task clusters to be processed is determined; the algorithm environment parameter attribute information compatible with the algorithm running task clusters to be processed currently needs to be processed is obtained, the corresponding algorithm running environment is built in the Docker container, and the algorithm running environment is debugged to make the algorithm running environment in a stable state.

[0006] Optionally, based on the GPU driver of the algorithm platform, the Docker container is driven, and it is determined whether the Docker container is in a normal working state; when the Docker container is in a normal working state, the algorithm code and the algorithm model are mounted on the Docker container, including: Determine whether the GPU driver of the algorithm platform is in a fully loaded running state, if so, do not drive the Docker container; if not, use the GPU driver to drive the Docker container; Extracting an average computing speed of the Docker container from the work log of the Docker container, and if the average computing speed is greater than or equal to a preset computing speed threshold, determining that the Docker container is in a normal working state; otherwise, determining that the Docker container is not in a normal working state; When the Docker container is in normal working state, the corresponding algorithm code and algorithm model are retrieved from the user end and mounted in the Docker container, and the Docker container is closed.

[0007] Optionally, when the Docker container is in a normal working state, determining the computing speed quality of the Docker container includes: When the average computing speed of the Docker container is greater than or equal to a preset computing speed threshold, all computing speed data recorded in the work log corresponding to the Docker container are retrieved; Obtaining a computing speed evaluation parameter using all computing speed data recorded in the work log corresponding to the Docker container; The calculation speed evaluation parameter is obtained by the following formula: Where K represents the computing speed evaluation parameter; n represents the number of all computing speed data recorded in the work log; S i represents the computing speed value corresponding to the i-th computing speed data; S p Indicates the average operation speed; S y Indicates the preset operation speed threshold; S b represents the standard deviation of the computing speed value corresponding to n computing speed data; k represents the parameter adjustment coefficient, and the parameter adjustment coefficient is obtained by the following formula: Where k represents the parameter adjustment coefficient; n represents the number of all operation speed data recorded in the work log; S i represents the computing speed value corresponding to the i-th computing speed data; S p Indicates the average operation speed; S y Indicates the preset operation speed threshold; S b Indicates the standard deviation of the computing speed values ​​corresponding to n computing speed data; Comparing the computing speed evaluation parameter with a preset first parameter threshold and a second parameter threshold; When the computing speed evaluation parameter exceeds a preset first parameter threshold, the computing speed quality of the Docker container is determined to be high; When the computing speed evaluation parameter does not exceed the preset first parameter threshold, but exceeds the preset second parameter threshold, the computing speed quality of the Docker container is determined to be medium; When the computing speed evaluation parameter does not exceed the preset second parameter threshold, the computing speed quality of the Docker container is determined to be low, and a low-level warning is issued.

[0008] Optionally, an interface is set for the Docker container, and corresponding middleware is constructed for the Docker container based on the interface of the Docker container; the algorithm operation result inside the Docker container is stored in the middleware, and the corresponding algorithm operation result is sent to the user terminal based on the operation result acquisition request of the user terminal, including: When the Docker container completes the algorithm running task, the available interface of the Docker container is identified, and the available interface with the largest bandwidth in the Docker container is used as the interface for data transmission of the Docker container; Based on the available interface with the maximum bandwidth, a corresponding Minio message middleware is constructed for the Docker container, and the algorithm running result inside the Docker container is stored in the Minio message middleware; The operation result acquisition request from the user terminal is analyzed, the matching algorithm operation result is located from the Minio message middleware, and the located algorithm operation result is sent to the user terminal through the available interface with the maximum bandwidth.

[0009] The container-based algorithm operation cluster deployment control system includes: The algorithm execution task identification module is used to analyze the algorithm execution request from the user end and determine the attribute information of the algorithm execution task that the user end expects to execute; An algorithm operation environment construction module is used to construct a corresponding algorithm operation environment in a Docker container based on the algorithm operation task attribute information; A container driver module, used for driving the Docker container based on the GPU driver of the algorithm platform, and determining whether the Docker container is in a normal working state; A file mounting module, used for mounting the algorithm code and the algorithm model in the Docker container when the Docker container is in a normal working state; A middleware construction module is used to set the interface of the Docker container and build corresponding middleware for the Docker container based on the interface of the Docker container; The operation result transmission module is used to store the algorithm operation result inside the Docker container in the middleware, and based on the operation result acquisition request from the user end, send the corresponding algorithm operation result to the user end.

[0010] Optionally, the algorithm execution task identification module is used to analyze the algorithm execution request from the user terminal to determine the attribute information of the algorithm execution task that the user terminal expects to execute, including: Analyze the algorithm running request from the user end, extract the algorithm type information involved in all the pending algorithm running tasks from the pending algorithm running task queue of the user end; divide all the pending algorithm running tasks of the user end into a plurality of pending algorithm running task clusters based on the algorithm type information; wherein all the pending algorithm running tasks under each pending algorithm running task cluster have the same algorithm type information; The algorithm operation environment construction module is used to construct a corresponding algorithm operation environment in a Docker container based on the algorithm operation task attribute information, including: Based on the amount of algorithm data to be run of all algorithm running task clusters to be processed, the processing order of all algorithm running task clusters to be processed is determined; the algorithm environment parameter attribute information compatible with the algorithm running task clusters to be processed currently needs to be processed is obtained, the corresponding algorithm running environment is built in the Docker container, and the algorithm running environment is debugged to make the algorithm running environment in a stable state.

[0011] Optionally, the container driver module is used to drive the Docker container based on a GPU driver of an algorithm platform, and to determine whether the Docker container is in a normal working state, including: Determine whether the GPU driver of the algorithm platform is in a fully loaded running state, if so, do not drive the Docker container; if not, use the GPU driver to drive the Docker container; Extracting an average computing speed of the Docker container from the work log of the Docker container, and if the average computing speed is greater than or equal to a preset computing speed threshold, determining that the Docker container is in a normal working state; otherwise, determining that the Docker container is not in a normal working state; The file mounting module is used to mount the algorithm code and the algorithm model in the Docker container when the Docker container is in a normal working state, including: When the Docker container is in normal working state, the corresponding algorithm code and algorithm model are retrieved from the user end and mounted in the Docker container, and the Docker container is closed.

[0012] Optionally, when the Docker container is in a normal working state, determining the computing speed quality of the Docker container includes: When the average computing speed of the Docker container is greater than or equal to a preset computing speed threshold, all computing speed data recorded in the work log corresponding to the Docker container are retrieved; Obtaining a computing speed evaluation parameter using all computing speed data recorded in the work log corresponding to the Docker container; The calculation speed evaluation parameter is obtained by the following formula: Where K represents the computing speed evaluation parameter; n represents the number of all computing speed data recorded in the work log; S i represents the computing speed value corresponding to the i-th computing speed data; S p Indicates the average operation speed; S y Indicates the preset operation speed threshold; S b represents the standard deviation of the computing speed value corresponding to n computing speed data; k represents the parameter adjustment coefficient, and the parameter adjustment coefficient is obtained by the following formula: Where k represents the parameter adjustment coefficient; n represents the number of all operation speed data recorded in the work log; S i represents the computing speed value corresponding to the i-th computing speed data; S p Indicates the average operation speed; S y Indicates the preset operation speed threshold; S b Indicates the standard deviation of the computing speed values ​​corresponding to n computing speed data; Comparing the computing speed evaluation parameter with a preset first parameter threshold and a second parameter threshold; When the computing speed evaluation parameter exceeds a preset first parameter threshold, the computing speed quality of the Docker container is determined to be high; When the computing speed evaluation parameter does not exceed the preset first parameter threshold, but exceeds the preset second parameter threshold, the computing speed quality of the Docker container is determined to be medium; When the computing speed evaluation parameter does not exceed the preset second parameter threshold, the computing speed quality of the Docker container is determined to be low, and a low-level warning is issued.

[0013] Optionally, the middleware construction module is used to set an interface for the Docker container, and based on the interface of the Docker container, to construct corresponding middleware for the Docker container, including: When the Docker container completes the algorithm running task, the available interface of the Docker container is identified, and the available interface with the largest bandwidth in the Docker container is used as the interface for data transmission of the Docker container; Based on the available interface with the maximum bandwidth, construct a corresponding Minio message middleware for the Docker container; The operation result transmission module is used to store the algorithm operation result inside the Docker container in the middleware, and send the corresponding algorithm operation result to the user terminal based on the operation result acquisition request from the user terminal, including: The algorithm running results inside the Docker container are stored in the Minio message middleware; the running result acquisition request from the user end is analyzed, the matching algorithm running results are located from the Minio message middleware, and the located algorithm running results are sent to the user end through the available interface with the maximum bandwidth.

[0014] Compared with the prior art, the present invention has the following beneficial effects: The container-based algorithm operation cluster deployment control method and system provided in the present application analyze the algorithm operation request from the user side, determine the attribute information of the algorithm operation task that the user side expects to execute, and build a corresponding algorithm operation environment in the Docker container, so as to provide a compatible and matching algorithm operation environment for different algorithm operation demand occasions; it also drives the Docker container based on the GPU driver of the algorithm platform, and when the Docker container is in normal working state, mounts the algorithm code and algorithm model in the Docker container, so that the Docker container can independently perform the algorithm operation, ensuring the working stability inside the container; based on the interface of the Dcoker container, builds the corresponding middleware for the Docker container, and stores the algorithm operation results inside the Docker container in the middleware, effectively reducing the data storage pressure of the container, and then sends the corresponding algorithm operation results to the user side, which is conducive to realizing the cluster deployment of algorithm operation and improving the compatibility of the algorithm platform with different algorithm operation occasions and the algorithm operation efficiency and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them: Figure 1 A schematic diagram of a flow chart of a container-based algorithm operation cluster deployment control method provided by the present invention.

[0016] Figure 2 This is a schematic diagram of the structure of the container-based algorithm operation cluster deployment control system provided by the present invention. DETAILED DESCRIPTION

[0017] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are described in detail below in conjunction with the accompanying drawings. It is to be understood that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should also be noted that, for ease of description, only some structures related to the present application are shown in the accompanying drawings, rather than all structures. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0018] The terms "including" and "having" and any variations thereof in this application are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices.

[0019] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0020] See also Figure 1 As shown, a container-based algorithm operation cluster deployment control method provided in an embodiment of the present application includes: Analyze the algorithm running request from the user end to determine the attribute information of the algorithm running task that the user end expects to execute; based on the attribute information of the algorithm running task, build the corresponding algorithm running environment in the Docker container; Based on the GPU driver of the algorithm platform, the Docker container is driven and it is determined whether the Docker container is in a normal working state; when the Docker container is in a normal working state, the algorithm code and the algorithm model are mounted in the Docker container; An interface is set for the Docker container, and based on the interface of the Docker container, a corresponding middleware is built for the Docker container; the algorithm running results inside the Docker container are stored in the middleware, and based on a running result acquisition request from the user end, the corresponding algorithm running results are sent to the user end.

[0021] The beneficial effects of the above embodiments are as follows: the container-based algorithm operation cluster deployment control method analyzes the algorithm operation request from the user end, determines the attribute information of the algorithm operation task that the user end expects to execute, and thereby builds a corresponding algorithm operation environment in the Docker container, thereby providing a compatible and matching algorithm operation environment for different algorithm operation demand occasions; it also drives the Docker container based on the GPU driver of the algorithm platform, and when the Docker container is in normal working state, mounts the algorithm code and algorithm model in the Docker container, so that the Docker container can independently perform the algorithm operation, ensuring the working stability inside the container; based on the interface of the Dcoker container, builds the corresponding middleware for the Docker container, and stores the algorithm operation results inside the Docker container in the middleware, effectively reducing the data storage pressure of the container, and then sends the corresponding algorithm operation results to the user end, which is conducive to realizing the cluster deployment of algorithm operation and improving the compatibility of the algorithm platform with different algorithm operation occasions and the algorithm operation efficiency and reliability.

[0022] In another embodiment, an algorithm running request from a user terminal is analyzed to determine attribute information of an algorithm running task that the user terminal expects to execute; and based on the attribute information of the algorithm running task, a corresponding algorithm running environment is constructed in a Docker container, including: Analyze the algorithm running request from the user end, extract the algorithm type information involved in all the pending algorithm running tasks from the pending algorithm running task queue of the user end; divide all the pending algorithm running tasks of the user end into a number of pending algorithm running task clusters based on the algorithm type information; wherein all the pending algorithm running tasks under each pending algorithm running task cluster have the same algorithm type information; Based on the amount of algorithm data to be run in all pending algorithm running task clusters, determine the processing order of all pending algorithm running task clusters; obtain the algorithm environment parameter attribute information compatible with the pending algorithm running task cluster that currently needs to be processed, build the corresponding algorithm running environment in the Docker container, and debug the algorithm running environment to make the algorithm running environment in a stable state.

[0023] The beneficial effect of the above embodiment is that different user terminals have different algorithm operation requirements, and the algorithm operation requirements may include but are not limited to the type of algorithm that the user terminal needs to run. The algorithm operation request from the user terminal is analyzed, and the algorithm type information involved in all pending algorithm operation tasks is extracted from the pending algorithm operation task queue of the user terminal. Based on the algorithm type information, all pending algorithm operation tasks of the user terminal are divided into several pending algorithm operation task clusters, so that the pending algorithm operation tasks with the same type of algorithm operation requirements can be centrally managed, that is, after the corresponding algorithm operation environment is constructed in the container, all corresponding pending algorithm tasks can be completed directly inside the container, without the need to repeatedly switch the algorithm operation environment of the container, thereby improving the processing efficiency of the container for the algorithm operation tasks. In addition, based on the amount of pending algorithm data of all pending algorithm operation task clusters, the processing order of all pending algorithm operation task clusters is determined. When the amount of pending algorithm data of the pending algorithm operation task cluster is smaller, the corresponding processing order is more prioritized. Then obtain the algorithm environment parameter attribute information that is compatible with the algorithm running task cluster that needs to be processed, so as to build the corresponding algorithm running environment in the Docker container and debug the algorithm running environment to make the algorithm running environment stable, which can improve the working reliability of the Docker container.

[0024] In another embodiment, based on the GPU driver of the algorithm platform, the Docker container is driven, and it is determined whether the Docker container is in a normal working state; when the Docker container is in a normal working state, the algorithm code and the algorithm model are mounted on the Docker container, including: Determine whether the GPU driver of the algorithm platform is in a fully loaded running state. If so, do not drive the Docker container; if not, use the GPU driver to drive the Docker container; Extracting an average computing speed of the Docker container from the work log of the Docker container, and if the average computing speed is greater than or equal to a preset computing speed threshold, determining that the Docker container is in a normal working state; otherwise, determining that the Docker container is not in a normal working state; When the Docker container is in normal working state, the corresponding algorithm code and algorithm model are retrieved from the user end and mounted in the Docker container, and the Docker container is closed.

[0025] The beneficial effect of the above embodiment is to determine whether the GPU driver of the algorithm platform is in a fully loaded operation state, so as to determine whether the algorithm platform has sufficient GPU driver capability to drive the Docker container, thereby providing a reliable driver guarantee for the Docker container. In addition, the average operation speed of the Docker container is compared with the preset operation speed threshold to accurately determine whether the Docker container is in a normal working state, and when the Docker container is in a normal working state, the corresponding algorithm code and algorithm model are retrieved from the user end and mounted in the Docker container, and the Docker container is closed to ensure that the Docker container can use the algorithm model to independently process the algorithm code, thereby avoiding external interference to the Docker container during operation.

[0026] In another embodiment, when the Docker container is in a normal working state, determining the computing speed quality of the Docker container includes: When the average computing speed of the Docker container is greater than or equal to a preset computing speed threshold, all computing speed data recorded in the work log corresponding to the Docker container are retrieved; Obtaining a computing speed evaluation parameter using all computing speed data recorded in the work log corresponding to the Docker container; The calculation speed evaluation parameter is obtained by the following formula: Where K represents the computing speed evaluation parameter; n represents the number of all computing speed data recorded in the work log; S i represents the computing speed value corresponding to the i-th computing speed data; S p Indicates the average operation speed; S y Indicates the preset operation speed threshold; S b represents the standard deviation of the computing speed value corresponding to n computing speed data; k represents the parameter adjustment coefficient, and the parameter adjustment coefficient is obtained by the following formula: Where k represents the parameter adjustment coefficient; n represents the number of all operation speed data recorded in the work log; S i represents the computing speed value corresponding to the i-th computing speed data; S p Indicates the average operation speed; S y Indicates the preset operation speed threshold; S b Indicates the standard deviation of the computing speed values ​​corresponding to n computing speed data; Comparing the computing speed evaluation parameter with a preset first parameter threshold and a second parameter threshold; When the computing speed evaluation parameter exceeds a preset first parameter threshold, the computing speed quality of the Docker container is determined to be high; When the computing speed evaluation parameter does not exceed the preset first parameter threshold, but exceeds the preset second parameter threshold, the computing speed quality of the Docker container is determined to be medium; When the computing speed evaluation parameter does not exceed the preset second parameter threshold, the computing speed quality of the Docker container is determined to be low, and a low-level warning is issued.

[0027] The beneficial effects of the above embodiments are as follows: the scheme calculates the computing speed evaluation parameter K, comprehensively considers multiple factors such as the average computing speed (Sp) of the Docker container, the preset computing speed threshold (Sy), the standard deviation of the computing speed data (Sb), and each computing speed data (Si), and realizes a refined evaluation of the computing speed quality of the Docker container. By introducing the parameter adjustment coefficient k, the flexibility and accuracy of the evaluation are further enhanced, so that the evaluation parameters can better reflect the actual computing performance of the Docker container. According to the comparison results of the computing speed evaluation parameter K with the preset first parameter threshold and second parameter threshold, the computing speed quality of the Docker container is divided into three levels: high, medium and low. When the computing speed quality is judged to be low, a low-level warning is triggered, which helps to timely discover and deal with potential computing performance problems to avoid further deterioration of the problem. The scheme uses all computing speed data recorded in the work log corresponding to the Docker container for evaluation, ensuring the objectivity and accuracy of the evaluation. By analyzing the data in the work log, the historical trajectory of the computing speed change can be traced, providing data support for further optimizing the computing performance of the Docker container. Through the automated evaluation process, the subjectivity of manual intervention and judgment is reduced, and the operation and maintenance efficiency is improved. The hierarchical judgment and early warning mechanism enables operation and maintenance personnel to quickly locate and deal with computing performance issues, reducing the risk of business interruption caused by poor computing performance. Accurate evaluation of the computing speed quality of Docker containers can provide a decision basis for resource allocation. For Docker containers with low computing speed quality, you can consider increasing resource allocation or optimizing performance; for Docker containers with high computing speed quality, you can maintain or appropriately reduce resource allocation to achieve optimal resource configuration and efficient utilization.

[0028] In summary, this technical solution provides strong support for the performance management and optimization of Docker containers by finely evaluating the computing speed quality of Docker containers, establishing a graded judgment and early warning mechanism, performing data analysis based on work logs, and improving operation and maintenance efficiency.

[0029] In another embodiment, an interface is set for the Docker container, and corresponding middleware is constructed for the Docker container based on the interface of the Docker container; the algorithm operation result inside the Docker container is stored in the middleware, and the corresponding algorithm operation result is sent to the user terminal based on the operation result acquisition request of the user terminal, including: When the Docker container completes the algorithm running task, the available interface of the Docker container is identified, and the available interface with the largest bandwidth in the Docker container is used as the interface for data transmission of the Docker container; Based on the available interface with the largest bandwidth, a corresponding Minio message middleware is constructed for the Docker container, and the algorithm operation result inside the Docker container is stored in the Minio message middleware; The operation result acquisition request from the user terminal is analyzed, the matching algorithm operation result is located from the Minio message middleware, and the located algorithm operation result is sent to the user terminal through the available interface with the maximum bandwidth.

[0030] The beneficial effect of the above embodiment is that when the Docker container completes the algorithm running task, the available interface of the Docker container is identified, and the available interface with the largest bandwidth in the Docker container is used as the interface for data transmission of the Docker container, so as to ensure that the Docker container can quickly transmit the algorithm running results to the outside. Based on the available interface with the largest bandwidth, the corresponding Minio message middleware is constructed for the Docker container, and the algorithm running results inside the Docker container are stored in the Minio message middleware. The Minio message middleware is used as the cache space for the algorithm running results to avoid the Docker container from storing too many algorithm running results, thereby improving the running efficiency of the Docker container. Then, the running result acquisition request from the user end is analyzed, the matching algorithm running results are located from the Minio message middleware, and the located algorithm running results are sent to the user end through the available interface with the largest bandwidth, so as to improve the compatibility of the algorithm platform for different algorithm running occasions and the algorithm running efficiency and reliability.

[0031] See also Figure 2 As shown, the algorithm operation cluster deployment control system based on containers provided in one embodiment of the present application includes: The algorithm operation task identification module is used to analyze the algorithm operation request from the user end and determine the attribute information of the algorithm operation task that the user end expects to execute; The algorithm running environment construction module is used to build the corresponding algorithm running environment in the Docker container based on the algorithm running task attribute information; The container driver module is used to drive the Docker container based on the GPU driver of the algorithm platform and determine whether the Docker container is in a normal working state; A file mounting module, used to mount the algorithm code and algorithm model in the Docker container when the Docker container is in normal working state; A middleware construction module is used to set the interface of the Docker container and build the corresponding middleware for the Docker container based on the interface of the Docker container; The operation result transmission module is used to store the algorithm operation result inside the Docker container to the middleware, and based on the operation result acquisition request from the user end, send the corresponding algorithm operation result to the user end.

[0032] The beneficial effects of the above embodiments are as follows: the container-based algorithm operation cluster deployment control system analyzes the algorithm operation request from the user end, determines the attribute information of the algorithm operation task that the user end expects to execute, and thereby builds a corresponding algorithm operation environment in the Docker container, thereby providing a compatible and matching algorithm operation environment for different algorithm operation demand occasions; it also drives the Docker container based on the GPU driver of the algorithm platform, and when the Docker container is in normal working state, mounts the algorithm code and algorithm model in the Docker container, so that the Docker container can independently perform the algorithm operation, ensuring the working stability inside the container; based on the interface of the Dcoker container, builds the corresponding middleware for the Docker container, and stores the algorithm operation results inside the Docker container in the middleware, effectively reducing the data storage pressure of the container, and then sends the corresponding algorithm operation results to the user end, which is conducive to realizing the cluster deployment of algorithm operation and improving the compatibility of the algorithm platform with different algorithm operation occasions and the algorithm operation efficiency and reliability.

[0033] In another embodiment, the algorithm execution task identification module is used to analyze the algorithm execution request from the user terminal to determine the attribute information of the algorithm execution task that the user terminal expects to execute, including: Analyze the algorithm running request from the user end, extract the algorithm type information involved in all the pending algorithm running tasks from the pending algorithm running task queue of the user end; divide all the pending algorithm running tasks of the user end into a number of pending algorithm running task clusters based on the algorithm type information; wherein all the pending algorithm running tasks under each pending algorithm running task cluster have the same algorithm type information; The algorithm runtime environment construction module is used to build the corresponding algorithm runtime environment in the Docker container based on the algorithm runtime task attribute information, including: Based on the amount of algorithm data to be run in all pending algorithm running task clusters, determine the processing order of all pending algorithm running task clusters; obtain the algorithm environment parameter attribute information compatible with the pending algorithm running task cluster that currently needs to be processed, build the corresponding algorithm running environment in the Docker container, and debug the algorithm running environment to make the algorithm running environment in a stable state.

[0034] The beneficial effect of the above embodiment is that different user terminals have different algorithm operation requirements, and the algorithm operation requirements may include but are not limited to the type of algorithm that the user terminal needs to run. The algorithm operation request from the user terminal is analyzed, and the algorithm type information involved in all pending algorithm operation tasks is extracted from the pending algorithm operation task queue of the user terminal. Based on the algorithm type information, all pending algorithm operation tasks of the user terminal are divided into several pending algorithm operation task clusters, so that the pending algorithm operation tasks with the same type of algorithm operation requirements can be centrally managed, that is, after the corresponding algorithm operation environment is constructed in the container, all corresponding pending algorithm tasks can be completed directly inside the container, without the need to repeatedly switch the algorithm operation environment of the container, thereby improving the processing efficiency of the container for the algorithm operation tasks. In addition, based on the amount of pending algorithm data of all pending algorithm operation task clusters, the processing order of all pending algorithm operation task clusters is determined. When the amount of pending algorithm data of the pending algorithm operation task cluster is smaller, the corresponding processing order is more prioritized. Then obtain the algorithm environment parameter attribute information that is compatible with the algorithm running task cluster that needs to be processed, so as to build the corresponding algorithm running environment in the Docker container and debug the algorithm running environment to make the algorithm running environment stable, which can improve the working reliability of the Docker container.

[0035] In another embodiment, the container driver module is used to drive the Docker container based on the GPU driver of the algorithm platform, and determine whether the Docker container is in a normal working state, including: Determine whether the GPU driver of the algorithm platform is in a fully loaded running state. If so, do not drive the Docker container; if not, use the GPU driver to drive the Docker container; Extracting an average computing speed of the Docker container from the work log of the Docker container, and if the average computing speed is greater than or equal to a preset computing speed threshold, determining that the Docker container is in a normal working state; otherwise, determining that the Docker container is not in a normal working state; The file mounting module is used to mount the algorithm code and algorithm model in the Docker container when the Docker container is in normal working state, including: When the Docker container is in normal working state, the corresponding algorithm code and algorithm model are retrieved from the user end and mounted in the Docker container, and the Docker container is closed.

[0036] The beneficial effect of the above embodiment is to determine whether the GPU driver of the algorithm platform is in a fully loaded operation state, so as to determine whether the algorithm platform has sufficient GPU driver capability to drive the Docker container, thereby providing a reliable driver guarantee for the Docker container. In addition, the average operation speed of the Docker container is compared with the preset operation speed threshold to accurately determine whether the Docker container is in a normal working state, and when the Docker container is in a normal working state, the corresponding algorithm code and algorithm model are retrieved from the user end and mounted in the Docker container, and the Docker container is closed to ensure that the Docker container can use the algorithm model to independently process the algorithm code, thereby avoiding external interference to the Docker container during operation.

[0037] In another embodiment, when the Docker container is in a normal working state, determining the computing speed quality of the Docker container includes: When the average computing speed of the Docker container is greater than or equal to a preset computing speed threshold, all computing speed data recorded in the work log corresponding to the Docker container are retrieved; Obtaining a computing speed evaluation parameter using all computing speed data recorded in the work log corresponding to the Docker container; The calculation speed evaluation parameter is obtained by the following formula: Where K represents the computing speed evaluation parameter; n represents the number of all computing speed data recorded in the work log; S i represents the computing speed value corresponding to the i-th computing speed data; S p Indicates the average operation speed; S y Indicates the preset operation speed threshold; S b represents the standard deviation of the computing speed value corresponding to n computing speed data; k represents the parameter adjustment coefficient, and the parameter adjustment coefficient is obtained by the following formula: Where k represents the parameter adjustment coefficient; n represents the number of all operation speed data recorded in the work log; S irepresents the computing speed value corresponding to the i-th computing speed data; S p Indicates the average operation speed; S y Indicates the preset operation speed threshold; S b Indicates the standard deviation of the computing speed values ​​corresponding to n computing speed data; Comparing the computing speed evaluation parameter with a preset first parameter threshold and a second parameter threshold; When the computing speed evaluation parameter exceeds a preset first parameter threshold, the computing speed quality of the Docker container is determined to be high; When the computing speed evaluation parameter does not exceed the preset first parameter threshold, but exceeds the preset second parameter threshold, the computing speed quality of the Docker container is determined to be medium; When the computing speed evaluation parameter does not exceed the preset second parameter threshold, the computing speed quality of the Docker container is determined to be low, and a low-level warning is issued.

[0038] The beneficial effects of the above embodiments are as follows: the scheme calculates the computing speed evaluation parameter K, comprehensively considers multiple factors such as the average computing speed (Sp) of the Docker container, the preset computing speed threshold (Sy), the standard deviation of the computing speed data (Sb), and each computing speed data (Si), and realizes a refined evaluation of the computing speed quality of the Docker container. By introducing the parameter adjustment coefficient k, the flexibility and accuracy of the evaluation are further enhanced, so that the evaluation parameters can better reflect the actual computing performance of the Docker container. According to the comparison results of the computing speed evaluation parameter K with the preset first parameter threshold and second parameter threshold, the computing speed quality of the Docker container is divided into three levels: high, medium and low. When the computing speed quality is judged to be low, a low-level warning is triggered, which helps to timely discover and deal with potential computing performance problems to avoid further deterioration of the problem. The scheme uses all computing speed data recorded in the work log corresponding to the Docker container for evaluation, ensuring the objectivity and accuracy of the evaluation. By analyzing the data in the work log, the historical trajectory of the computing speed change can be traced, providing data support for further optimizing the computing performance of the Docker container. Through the automated evaluation process, the subjectivity of manual intervention and judgment is reduced, and the operation and maintenance efficiency is improved. The hierarchical judgment and early warning mechanism enables operation and maintenance personnel to quickly locate and deal with computing performance issues, reducing the risk of business interruption caused by poor computing performance. Accurate evaluation of the computing speed quality of Docker containers can provide a decision basis for resource allocation. For Docker containers with low computing speed quality, you can consider increasing resource allocation or optimizing performance; for Docker containers with high computing speed quality, you can maintain or appropriately reduce resource allocation to achieve optimal resource configuration and efficient utilization.

[0039] In summary, this technical solution provides strong support for the performance management and optimization of Docker containers by finely evaluating the computing speed quality of Docker containers, establishing a graded judgment and early warning mechanism, performing data analysis based on work logs, and improving operation and maintenance efficiency.

[0040] In another embodiment, the middleware construction module is used to set the interface of the Docker container, and build the corresponding middleware for the Docker container based on the interface of the Docker container, including: When the Docker container completes the algorithm running task, the available interface of the Docker container is identified, and the available interface with the largest bandwidth in the Docker container is used as the interface for data transmission of the Docker container; Based on the available interface with the largest bandwidth, build the corresponding Minio message middleware for the Docker container; The operation result transmission module is used to store the algorithm operation result inside the Docker container to the middleware, and based on the operation result acquisition request from the user end, send the corresponding algorithm operation result to the user end, including: The algorithm running results inside the Docker container are stored in the Minio message middleware; the running result acquisition request from the user end is analyzed, the matching algorithm running results are located from the Minio message middleware, and the located algorithm running results are sent to the user end through the available interface with the maximum bandwidth.

[0041] The beneficial effect of the above embodiment is that when the Docker container completes the algorithm running task, the available interface of the Docker container is identified, and the available interface with the largest bandwidth in the Docker container is used as the interface for data transmission of the Docker container, so as to ensure that the Docker container can quickly transmit the algorithm running results to the outside. Based on the available interface with the largest bandwidth, the corresponding Minio message middleware is constructed for the Docker container, and the algorithm running results inside the Docker container are stored in the Minio message middleware. The Minio message middleware is used as the cache space for the algorithm running results to avoid the Docker container from storing too many algorithm running results, thereby improving the running efficiency of the Docker container. Then, the running result acquisition request from the user end is analyzed, the matching algorithm running results are located from the Minio message middleware, and the located algorithm running results are sent to the user end through the available interface with the largest bandwidth, so as to improve the compatibility of the algorithm platform for different algorithm running occasions and the algorithm running efficiency and reliability.

[0042] In general, the container-based algorithm operation cluster deployment control method and system analyzes the algorithm operation request from the user side, determines the attribute information of the algorithm operation task that the user side expects to execute, and builds the corresponding algorithm operation environment in the Docker container, thereby providing a compatible and matching algorithm operation environment for different algorithm operation demand occasions; it also drives the Docker container based on the GPU driver of the algorithm platform, and when the Docker container is in normal working state, mounts the algorithm code and algorithm model in the Docker container, so that the Docker container can independently perform the algorithm operation, ensuring the working stability inside the container; based on the interface of the Dcoker container, builds the corresponding middleware for the Docker container, and stores the algorithm operation results inside the Docker container to the middleware, effectively reducing the data storage pressure of the container, and then sends the corresponding algorithm operation results to the user side, which is conducive to realizing the cluster deployment of algorithm operation and improving the compatibility of the algorithm platform for different algorithm operation occasions and the algorithm operation efficiency and reliability.

[0043] The above is only a specific implementation of the present invention, and any other improvements made based on the concept of the present invention are considered to be within the protection scope of the present invention.

Claims

1. A container-based algorithm operation cluster deployment control method, characterized in that: include: Analyze the algorithm execution request from the user end to determine the attribute information of the algorithm execution task that the user end expects to execute; Based on the algorithm running task attribute information, build a corresponding algorithm running environment in the Docker container; Based on the GPU driver of the algorithm platform, the Docker container is driven, and it is determined whether the Docker container is in a normal working state; when the Docker container is in a normal working state, the algorithm code and the algorithm model are mounted on the Docker container; An interface is set for the Docker container, and based on the interface of the Docker container, a corresponding middleware is built for the Docker container; the algorithm running results inside the Docker container are stored in the middleware, and based on a running result acquisition request from the user end, the corresponding algorithm running results are sent to the user end.

2. The container-based algorithm operation cluster deployment control method according to claim 1, characterized in that: Analyze the algorithm execution request from the user end to determine the attribute information of the algorithm execution task that the user end expects to execute; Based on the algorithm running task attribute information, a corresponding algorithm running environment is constructed in the Docker container, including: Analyze the algorithm running request from the user end, extract the algorithm type information involved in all the pending algorithm running tasks from the pending algorithm running task queue of the user end; divide all the pending algorithm running tasks of the user end into a plurality of pending algorithm running task clusters based on the algorithm type information; wherein all the pending algorithm running tasks under each pending algorithm running task cluster have the same algorithm type information; Based on the amount of algorithm data to be run of all algorithm running task clusters to be processed, the processing order of all algorithm running task clusters to be processed is determined; the algorithm environment parameter attribute information compatible with the algorithm running task clusters to be processed currently needs to be processed is obtained, the corresponding algorithm running environment is built in the Docker container, and the algorithm running environment is debugged to make the algorithm running environment in a stable state.

3. The container-based algorithm operation cluster deployment control method according to claim 1, characterized in that: Based on the GPU driver of the algorithm platform, the Docker container is driven, and it is determined whether the Docker container is in a normal working state; when the Docker container is in a normal working state, the algorithm code and the algorithm model are mounted on the Docker container, including: Determine whether the GPU driver of the algorithm platform is in a fully loaded running state, if so, do not drive the Docker container; if not, use the GPU driver to drive the Docker container; Extracting an average computing speed of the Docker container from the work log of the Docker container, and if the average computing speed is greater than or equal to a preset computing speed threshold, determining that the Docker container is in a normal working state; otherwise, determining that the Docker container is not in a normal working state; When the Docker container is in normal working state, the corresponding algorithm code and algorithm model are retrieved from the user end and mounted in the Docker container, and the Docker container is closed.

4. The container-based algorithm operation cluster deployment control method according to claim 3, characterized in that: When the Docker container is in a normal working state, determining the computing speed quality of the Docker container includes: When the average computing speed of the Docker container is greater than or equal to a preset computing speed threshold, all computing speed data recorded in the work log corresponding to the Docker container are retrieved; Obtaining a computing speed evaluation parameter using all computing speed data recorded in the work log corresponding to the Docker container; The calculation speed evaluation parameter is obtained by the following formula: Where K represents the computing speed evaluation parameter; n represents the number of all computing speed data recorded in the work log; S i represents the computing speed value corresponding to the i-th computing speed data; S p Indicates the average operation speed; S y Indicates the preset operation speed threshold; S b represents the standard deviation of the computing speed value corresponding to n computing speed data; k represents the parameter adjustment coefficient, and the parameter adjustment coefficient is obtained by the following formula: Where k represents the parameter adjustment coefficient; n represents the number of all operation speed data recorded in the work log; S i represents the computing speed value corresponding to the i-th computing speed data; S p Indicates the average operation speed; S y Indicates the preset operation speed threshold; S b Indicates the standard deviation of the computing speed values ​​corresponding to n computing speed data; Comparing the computing speed evaluation parameter with a preset first parameter threshold and a second parameter threshold; When the computing speed evaluation parameter exceeds a preset first parameter threshold, the computing speed quality of the Docker container is determined to be high; When the computing speed evaluation parameter does not exceed the preset first parameter threshold, but exceeds the preset second parameter threshold, the computing speed quality of the Docker container is determined to be medium; When the computing speed evaluation parameter does not exceed the preset second parameter threshold, the computing speed quality of the Docker container is determined to be low, and a low-level warning is issued.

5. The container-based algorithm operation cluster deployment control method according to claim 1, characterized in that: The Docker container is interfaced and corresponding middleware is constructed for the Docker container based on the interface of the Docker container; the algorithm operation result inside the Docker container is stored in the middleware, and the corresponding algorithm operation result is sent to the user terminal based on the operation result acquisition request of the user terminal, including: When the Docker container completes the algorithm running task, the available interface of the Docker container is identified, and the available interface with the largest bandwidth in the Docker container is used as the interface for data transmission of the Docker container; Based on the available interface with the maximum bandwidth, a corresponding Minio message middleware is constructed for the Docker container, and the algorithm running result inside the Docker container is stored in the Minio message middleware; The operation result acquisition request from the user terminal is analyzed, the matching algorithm operation result is located from the Minio message middleware, and the located algorithm operation result is sent to the user terminal through the available interface with the maximum bandwidth.

6. The algorithm operation cluster deployment control system based on the container is characterized by: include: The algorithm execution task identification module is used to analyze the algorithm execution request from the user end and determine the attribute information of the algorithm execution task that the user end expects to execute; An algorithm operation environment construction module is used to construct a corresponding algorithm operation environment in a Docker container based on the algorithm operation task attribute information; A container driver module, used for driving the Docker container based on the GPU driver of the algorithm platform, and determining whether the Docker container is in a normal working state; A file mounting module, used for mounting the algorithm code and the algorithm model in the Docker container when the Docker container is in a normal working state; A middleware construction module, used to set the interface of the Docker container and build corresponding middleware for the Docker container based on the interface of the Docker container; The operation result transmission module is used to store the algorithm operation result inside the Docker container in the middleware, and based on the operation result acquisition request from the user end, send the corresponding algorithm operation result to the user end.

7. The container-based algorithm operation cluster deployment control system according to claim 6, characterized in that: The algorithm execution task identification module is used to analyze the algorithm execution request from the user terminal and determine the attribute information of the algorithm execution task that the user terminal expects to execute, including: Analyze the algorithm running request from the user end, extract the algorithm type information involved in all the pending algorithm running tasks from the pending algorithm running task queue of the user end; divide all the pending algorithm running tasks of the user end into a plurality of pending algorithm running task clusters based on the algorithm type information; wherein all the pending algorithm running tasks under each pending algorithm running task cluster have the same algorithm type information; The algorithm operation environment construction module is used to construct a corresponding algorithm operation environment in a Docker container based on the algorithm operation task attribute information, including: Based on the amount of algorithm data to be run of all algorithm running task clusters to be processed, the processing order of all algorithm running task clusters to be processed is determined; the algorithm environment parameter attribute information compatible with the algorithm running task clusters to be processed currently needs to be processed is obtained, the corresponding algorithm running environment is built in the Docker container, and the algorithm running environment is debugged to make the algorithm running environment in a stable state.

8. The container-based algorithm operation cluster deployment control system according to claim 6, characterized in that: The container driver module is used to drive the Docker container based on the GPU driver of the algorithm platform, and determine whether the Docker container is in a normal working state, including: Determine whether the GPU driver of the algorithm platform is in a fully loaded running state, if so, do not drive the Docker container; if not, use the GPU driver to drive the Docker container; Extracting an average computing speed of the Docker container from the work log of the Docker container, and if the average computing speed is greater than or equal to a preset computing speed threshold, determining that the Docker container is in a normal working state; otherwise, determining that the Docker container is not in a normal working state; The file mounting module is used to mount the algorithm code and the algorithm model in the Docker container when the Docker container is in a normal working state, including: When the Docker container is in normal working state, the corresponding algorithm code and algorithm model are retrieved from the user end and mounted in the Docker container, and the Docker container is closed.

9. The container-based algorithm operation cluster deployment control system according to claim 8, characterized in that: When the Docker container is in a normal working state, determining the computing speed quality of the Docker container includes: When the average computing speed of the Docker container is greater than or equal to a preset computing speed threshold, all computing speed data recorded in the work log corresponding to the Docker container are retrieved; Obtaining a computing speed evaluation parameter using all computing speed data recorded in the work log corresponding to the Docker container; The calculation speed evaluation parameter is obtained by the following formula: Where K represents the computing speed evaluation parameter; n represents the number of all computing speed data recorded in the work log; S i represents the computing speed value corresponding to the i-th computing speed data; S p Indicates the average operation speed; S y Indicates the preset operation speed threshold; S b represents the standard deviation of the computing speed value corresponding to n computing speed data; k represents the parameter adjustment coefficient, and the parameter adjustment coefficient is obtained by the following formula: Where k represents the parameter adjustment coefficient; n represents the number of all operation speed data recorded in the work log; S i represents the computing speed value corresponding to the i-th computing speed data; S p Indicates the average operation speed; S y Indicates the preset operation speed threshold; S b Indicates the standard deviation of the computing speed values ​​corresponding to n computing speed data; Comparing the computing speed evaluation parameter with a preset first parameter threshold and a second parameter threshold; When the computing speed evaluation parameter exceeds a preset first parameter threshold, the computing speed quality of the Docker container is determined to be high; When the computing speed evaluation parameter does not exceed the preset first parameter threshold, but exceeds the preset second parameter threshold, the computing speed quality of the Docker container is determined to be medium; When the computing speed evaluation parameter does not exceed the preset second parameter threshold, the computing speed quality of the Docker container is determined to be low, and a low-level warning is issued.

10. The container-based algorithm operation cluster deployment control system according to claim 6, characterized in that: The middleware construction module is used to set the interface of the Docker container and build the corresponding middleware for the Docker container based on the interface of the Docker container, including: When the Docker container completes the algorithm running task, the available interface of the Docker container is identified, and the available interface with the largest bandwidth in the Docker container is used as the interface for data transmission of the Docker container; Based on the available interface with the maximum bandwidth, construct a corresponding Minio message middleware for the Docker container; The operation result transmission module is used to store the algorithm operation result inside the Docker container in the middleware, and send the corresponding algorithm operation result to the user terminal based on the operation result acquisition request from the user terminal, including: The algorithm running results inside the Docker container are stored in the Minio message middleware; the running result acquisition request from the user end is analyzed, the matching algorithm running results are located from the Minio message middleware, and the located algorithm running results are sent to the user end through the available interface with the maximum bandwidth.

Images